diff --git "a/multilang/ggml-small-encoder-openvino-int4/ggml-small-encoder-openvino.xml" "b/multilang/ggml-small-encoder-openvino-int4/ggml-small-encoder-openvino.xml"
--- "a/multilang/ggml-small-encoder-openvino-int4/ggml-small-encoder-openvino.xml"
+++ "b/multilang/ggml-small-encoder-openvino-int4/ggml-small-encoder-openvino.xml"
@@ -2,12 +2,12 @@
 <net name="main_graph" version="11">
 	<layers>
 		<layer id="0" name="mel" type="Parameter" version="opset1">
-			<data shape="1,80,2000" element_type="f32" />
+			<data shape="1,80,3000" element_type="f32" />
 			<output>
 				<port id="0" precision="FP32" names="mel">
 					<dim>1</dim>
 					<dim>80</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -22,7 +22,7 @@
 			</output>
 		</layer>
 		<layer id="2" name="Convert_1225" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>768</dim>
@@ -31,7 +31,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>80</dim>
 					<dim>3</dim>
@@ -49,7 +49,7 @@
 			</output>
 		</layer>
 		<layer id="4" name="Convert_1228" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>768</dim>
@@ -58,29 +58,29 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="5" name="Subtract_1229" type="Subtract" version="opset1">
+		<layer id="5" name="onnx::Conv_1454/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>80</dim>
 					<dim>3</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>80</dim>
 					<dim>3</dim>
@@ -97,39 +97,39 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="7" name="onnx::Conv_1454/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="7" name="onnx::Conv_1454/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>80</dim>
+					<dim>3</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>80</dim>
+					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="8" name="onnx::Conv_1454/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="8" name="onnx::Conv_1454/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>80</dim>
 					<dim>3</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>768</dim>
-					<dim>1</dim>
-					<dim>1</dim>
-				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>80</dim>
 					<dim>3</dim>
@@ -142,7 +142,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>80</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>768</dim>
@@ -154,7 +154,7 @@
 				<port id="2" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -193,7 +193,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</input>
 			<output>
@@ -298,7 +298,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
@@ -310,7 +310,7 @@
 				<port id="2" precision="FP32" names="/conv1/Conv_output_0">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -326,7 +326,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -334,7 +334,7 @@
 				<port id="2" precision="FP32" names="/Div_output_0">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -343,14 +343,14 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/Erf_output_0">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -366,7 +366,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -374,7 +374,7 @@
 				<port id="2" precision="FP32" names="/Add_output_0">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -384,19 +384,19 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/Mul_output_0">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -412,7 +412,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -420,7 +420,7 @@
 				<port id="2" precision="FP32" names="/Mul_1_output_0">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -435,7 +435,7 @@
 			</output>
 		</layer>
 		<layer id="30" name="Convert_1236" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>768</dim>
@@ -444,7 +444,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 					<dim>3</dim>
@@ -462,7 +462,7 @@
 			</output>
 		</layer>
 		<layer id="32" name="Convert_1239" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>768</dim>
@@ -471,29 +471,29 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="33" name="Subtract_1240" type="Subtract" version="opset1">
+		<layer id="33" name="onnx::Conv_1456/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 					<dim>3</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 					<dim>3</dim>
@@ -510,39 +510,39 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="35" name="onnx::Conv_1456/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="35" name="onnx::Conv_1456/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+					<dim>3</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>768</dim>
+					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="36" name="onnx::Conv_1456/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="36" name="onnx::Conv_1456/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 					<dim>3</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>768</dim>
-					<dim>1</dim>
-					<dim>1</dim>
-				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 					<dim>3</dim>
@@ -555,7 +555,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>768</dim>
@@ -567,7 +567,7 @@
 				<port id="2" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -606,7 +606,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
@@ -711,7 +711,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
@@ -723,7 +723,7 @@
 				<port id="2" precision="FP32" names="/conv2/Conv_output_0">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -739,7 +739,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -747,7 +747,7 @@
 				<port id="2" precision="FP32" names="/Div_1_output_0">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -756,14 +756,14 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/Erf_1_output_0">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -779,7 +779,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -787,7 +787,7 @@
 				<port id="2" precision="FP32" names="/Add_1_output_0">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -797,19 +797,19 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/Mul_2_output_0">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -825,7 +825,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -833,7 +833,7 @@
 				<port id="2" precision="FP32" names="/Mul_3_output_0">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -850,7 +850,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="I64">
 					<dim>3</dim>
@@ -859,16 +859,16 @@
 			<output>
 				<port id="2" precision="FP32" names="/Transpose_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="59" name="onnx::Add_1462" type="Const" version="opset1">
-			<data element_type="f32" shape="1000, 768" offset="1964596" size="3072000" />
+			<data element_type="f32" shape="1500, 768" offset="1964596" size="4608000" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1462">
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -878,18 +878,18 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/Add_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -899,14 +899,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -916,20 +916,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.0/attn_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="63" name="onnx::Add_1487" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="5036596" size="3072" />
+			<data element_type="f32" shape="768" offset="6572596" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1487">
 					<dim>768</dim>
@@ -937,7 +937,7 @@
 			</output>
 		</layer>
 		<layer id="64" name="onnx::Add_1464" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="5039668" size="3072" />
+			<data element_type="f32" shape="768" offset="6575668" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1464">
 					<dim>768</dim>
@@ -945,13 +945,13 @@
 			</output>
 		</layer>
 		<layer id="65" name="Constant_244" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="66" name="Constant_243" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -980,7 +980,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -990,13 +990,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="70" name="blocks.0.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="5042756" size="3072" />
+			<data element_type="f32" shape="768" offset="6578756" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.0.attn_ln.weight">
 					<dim>768</dim>
@@ -1008,7 +1008,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -1018,13 +1018,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="72" name="blocks.0.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="5045828" size="3072" />
+			<data element_type="f32" shape="768" offset="6581828" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.0.attn_ln.bias">
 					<dim>768</dim>
@@ -1036,7 +1036,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -1046,7 +1046,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -1056,20 +1056,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.0/attn_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="75" name="onnx::MatMul_1465" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="5048900" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="6584900" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -1079,7 +1079,7 @@
 			</output>
 		</layer>
 		<layer id="76" name="Convert_1268" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -1088,7 +1088,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -1096,7 +1096,7 @@
 			</output>
 		</layer>
 		<layer id="77" name="onnx::MatMul_1465/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -1104,32 +1104,32 @@
 			</output>
 		</layer>
 		<layer id="78" name="Convert_1270" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="79" name="Subtract_1271" type="Subtract" version="opset1">
+		<layer id="79" name="onnx::MatMul_1465/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -1137,7 +1137,7 @@
 			</output>
 		</layer>
 		<layer id="80" name="onnx::MatMul_1465/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="5343813" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="6879813" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -1146,57 +1146,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="81" name="onnx::MatMul_1465/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="82" name="onnx::MatMul_1465/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="81" name="onnx::MatMul_1465/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="83" name="Constant_1276" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="82" name="Constant_1275" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="84" name="Reshape_1277" type="Reshape" version="opset1">
+		<layer id="83" name="Reshape_1276" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -1206,7 +1189,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="84" name="onnx::MatMul_1465/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -1217,7 +1215,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -1228,7 +1226,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/query/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -1241,20 +1239,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/query/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="87" name="/blocks.0/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.0/attn/Constant_2_output_0">
 					<dim>4</dim>
@@ -1266,7 +1264,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -1276,14 +1274,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/Reshape_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="89" name="Constant_260" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -1294,7 +1292,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -1306,19 +1304,19 @@
 				<port id="2" precision="FP32" names="/blocks.0/attn/Transpose_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="91" name="/blocks.0/attn/Constant" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353109" size="4" />
+			<data element_type="f32" shape="" offset="6889109" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.0/attn/Constant_output_0" />
 			</output>
 		</layer>
 		<layer id="92" name="/blocks.0/attn/Constant_1" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353113" size="4" />
+			<data element_type="f32" shape="" offset="6889113" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.0/attn/Constant_1_output_0" />
 			</output>
@@ -1339,7 +1337,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -1348,13 +1346,13 @@
 				<port id="2" precision="FP32" names="/blocks.0/attn/Mul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="95" name="onnx::MatMul_1467" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="5353117" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="6889117" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -1364,7 +1362,7 @@
 			</output>
 		</layer>
 		<layer id="96" name="Convert_1257" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -1373,7 +1371,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -1381,7 +1379,7 @@
 			</output>
 		</layer>
 		<layer id="97" name="onnx::MatMul_1467/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -1389,32 +1387,32 @@
 			</output>
 		</layer>
 		<layer id="98" name="Convert_1259" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="99" name="Subtract_1260" type="Subtract" version="opset1">
+		<layer id="99" name="onnx::MatMul_1467/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -1422,7 +1420,7 @@
 			</output>
 		</layer>
 		<layer id="100" name="onnx::MatMul_1467/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="5648029" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="7184030" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -1431,57 +1429,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="101" name="onnx::MatMul_1467/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="102" name="onnx::MatMul_1467/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="101" name="onnx::MatMul_1467/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="103" name="Constant_1265" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="102" name="Constant_1264" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="104" name="Reshape_1266" type="Reshape" version="opset1">
+		<layer id="103" name="Reshape_1265" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -1491,7 +1472,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="104" name="onnx::MatMul_1467/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -1502,7 +1498,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -1513,13 +1509,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/key/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="106" name="/blocks.0/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.0/attn/Constant_3_output_0">
 					<dim>4</dim>
@@ -1531,7 +1527,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -1541,14 +1537,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/Reshape_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="108" name="Constant_265" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5657245" size="32" />
+			<data element_type="i64" shape="4" offset="7193246" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -1559,7 +1555,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -1572,7 +1568,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -1583,7 +1579,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -1592,7 +1588,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -1602,22 +1598,22 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/MatMul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -1627,16 +1623,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.0/attn/Cast_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -1646,16 +1642,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.0/attn/Softmax_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -1665,21 +1661,21 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.0/attn/Cast_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="115" name="onnx::Add_1469" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="5657277" size="3072" />
+			<data element_type="f32" shape="768" offset="7193278" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1469">
 					<dim>768</dim>
@@ -1687,7 +1683,7 @@
 			</output>
 		</layer>
 		<layer id="116" name="onnx::MatMul_1470" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="5660349" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="7196350" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -1697,7 +1693,7 @@
 			</output>
 		</layer>
 		<layer id="117" name="Convert_1246" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -1706,7 +1702,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -1714,7 +1710,7 @@
 			</output>
 		</layer>
 		<layer id="118" name="onnx::MatMul_1470/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7491262" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -1722,32 +1718,32 @@
 			</output>
 		</layer>
 		<layer id="119" name="Convert_1248" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="120" name="Subtract_1249" type="Subtract" version="opset1">
+		<layer id="120" name="onnx::MatMul_1470/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -1755,7 +1751,7 @@
 			</output>
 		</layer>
 		<layer id="121" name="onnx::MatMul_1470/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="5955261" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="7491263" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -1764,57 +1760,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="122" name="onnx::MatMul_1470/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="123" name="onnx::MatMul_1470/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="122" name="onnx::MatMul_1470/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="124" name="Constant_1254" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="123" name="Constant_1253" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="125" name="Reshape_1255" type="Reshape" version="opset1">
+		<layer id="124" name="Reshape_1254" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -1824,7 +1803,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="125" name="onnx::MatMul_1470/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -1835,7 +1829,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -1846,7 +1840,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/value/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -1859,20 +1853,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/value/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="128" name="/blocks.0/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.0/attn/Constant_4_output_0">
 					<dim>4</dim>
@@ -1884,7 +1878,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -1894,14 +1888,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/Reshape_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="130" name="Constant_270" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -1912,7 +1906,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -1924,7 +1918,7 @@
 				<port id="2" precision="FP32" names="/blocks.0/attn/Transpose_2_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
@@ -1935,13 +1929,13 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</input>
@@ -1949,13 +1943,13 @@
 				<port id="2" precision="FP32" names="/blocks.0/attn/MatMul_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="133" name="Constant_277" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -1967,7 +1961,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -1977,7 +1971,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/Transpose_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -1988,7 +1982,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -2000,7 +1994,7 @@
 			</output>
 		</layer>
 		<layer id="136" name="/blocks.0/attn/Constant_6" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.0/attn/Constant_6_output_0">
 					<dim>1</dim>
@@ -2053,7 +2047,7 @@
 			</output>
 		</layer>
 		<layer id="141" name="/blocks.0/attn/Constant_5" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.0/attn/Constant_5_output_0">
 					<dim>1</dim>
@@ -2085,7 +2079,7 @@
 			</output>
 		</layer>
 		<layer id="143" name="/blocks.0/attn/Constant_8" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042740" size="8" />
+			<data element_type="i64" shape="1" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.0/attn/Constant_8_output_0">
 					<dim>1</dim>
@@ -2113,7 +2107,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -2124,13 +2118,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/Reshape_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="146" name="onnx::MatMul_1488" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="5964477" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="7500479" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -2140,7 +2134,7 @@
 			</output>
 		</layer>
 		<layer id="147" name="Convert_1279" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -2149,7 +2143,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -2157,7 +2151,7 @@
 			</output>
 		</layer>
 		<layer id="148" name="onnx::MatMul_1488/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -2165,32 +2159,32 @@
 			</output>
 		</layer>
 		<layer id="149" name="Convert_1281" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="150" name="Subtract_1282" type="Subtract" version="opset1">
+		<layer id="150" name="onnx::MatMul_1488/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -2198,7 +2192,7 @@
 			</output>
 		</layer>
 		<layer id="151" name="onnx::MatMul_1488/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="6259389" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="7795391" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -2207,57 +2201,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="152" name="onnx::MatMul_1488/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="153" name="onnx::MatMul_1488/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="152" name="onnx::MatMul_1488/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="154" name="Constant_1287" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="153" name="Constant_1286" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="155" name="Reshape_1288" type="Reshape" version="opset1">
+		<layer id="154" name="Reshape_1287" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -2267,7 +2244,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="155" name="onnx::MatMul_1488/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -2278,7 +2270,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -2289,7 +2281,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/out/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -2302,14 +2294,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/attn/out/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -2319,19 +2311,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -2341,20 +2333,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.0/mlp_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="160" name="onnx::Add_1493" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="6268605" size="3072" />
+			<data element_type="f32" shape="768" offset="7804607" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1493">
 					<dim>768</dim>
@@ -2362,7 +2354,7 @@
 			</output>
 		</layer>
 		<layer id="161" name="onnx::Add_1490" type="Const" version="opset1">
-			<data element_type="f32" shape="3072" offset="6271677" size="12288" />
+			<data element_type="f32" shape="3072" offset="7807679" size="12288" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1490">
 					<dim>3072</dim>
@@ -2370,13 +2362,13 @@
 			</output>
 		</layer>
 		<layer id="162" name="Constant_303" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="163" name="Constant_302" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -2405,7 +2397,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -2415,13 +2407,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="167" name="blocks.0.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="6283965" size="3072" />
+			<data element_type="f32" shape="768" offset="7819967" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.0.mlp_ln.weight">
 					<dim>768</dim>
@@ -2433,7 +2425,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -2443,13 +2435,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="169" name="blocks.0.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="6287037" size="3072" />
+			<data element_type="f32" shape="768" offset="7823039" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.0.mlp_ln.bias">
 					<dim>768</dim>
@@ -2461,7 +2453,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -2471,7 +2463,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/mlp_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -2481,20 +2473,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.0/mlp_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="172" name="onnx::MatMul_1491" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 3072" offset="6290109" size="1179648" />
+			<data element_type="u4" shape="6, 128, 3072" offset="7826111" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -2504,7 +2496,7 @@
 			</output>
 		</layer>
 		<layer id="173" name="Convert_1290" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -2513,7 +2505,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -2521,7 +2513,7 @@
 			</output>
 		</layer>
 		<layer id="174" name="onnx::MatMul_1491/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -2529,32 +2521,32 @@
 			</output>
 		</layer>
 		<layer id="175" name="Convert_1292" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="176" name="Subtract_1293" type="Subtract" version="opset1">
+		<layer id="176" name="onnx::MatMul_1491/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -2562,7 +2554,7 @@
 			</output>
 		</layer>
 		<layer id="177" name="onnx::MatMul_1491/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 3072" offset="7469757" size="36864" />
+			<data element_type="f16" shape="6, 1, 3072" offset="9005759" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -2571,57 +2563,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="178" name="onnx::MatMul_1491/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="178" name="onnx::MatMul_1491/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
-					<dim>1</dim>
+					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="179" name="onnx::MatMul_1491/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>6</dim>
-					<dim>128</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="180" name="Constant_1298" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="7506621" size="16" />
+		<layer id="179" name="Constant_1297" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="9042623" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="181" name="Reshape_1299" type="Reshape" version="opset1">
+		<layer id="180" name="Reshape_1298" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -2631,7 +2606,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="181" name="onnx::MatMul_1491/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
@@ -2642,7 +2632,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -2653,7 +2643,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/mlp/mlp.0/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -2666,14 +2656,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/mlp/mlp.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -2689,7 +2679,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -2697,7 +2687,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/mlp/mlp.1/Div_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -2706,14 +2696,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.0/mlp/mlp.1/Erf_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -2729,7 +2719,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -2737,7 +2727,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/mlp/mlp.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -2747,19 +2737,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/mlp/mlp.1/Mul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -2775,7 +2765,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -2783,13 +2773,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/mlp/mlp.1/Mul_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="192" name="onnx::MatMul_1494" type="Const" version="opset1">
-			<data element_type="u4" shape="24, 128, 768" offset="7506637" size="1179648" />
+			<data element_type="u4" shape="24, 128, 768" offset="9042639" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -2799,7 +2789,7 @@
 			</output>
 		</layer>
 		<layer id="193" name="Convert_1301" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -2808,7 +2798,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -2816,7 +2806,7 @@
 			</output>
 		</layer>
 		<layer id="194" name="onnx::MatMul_1494/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -2824,32 +2814,32 @@
 			</output>
 		</layer>
 		<layer id="195" name="Convert_1303" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="196" name="Subtract_1304" type="Subtract" version="opset1">
+		<layer id="196" name="onnx::MatMul_1494/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -2857,7 +2847,7 @@
 			</output>
 		</layer>
 		<layer id="197" name="onnx::MatMul_1494/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="24, 1, 768" offset="8686285" size="36864" />
+			<data element_type="f16" shape="24, 1, 768" offset="10222287" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>24</dim>
@@ -2866,57 +2856,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="198" name="onnx::MatMul_1494/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="199" name="onnx::MatMul_1494/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="198" name="onnx::MatMul_1494/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="200" name="Constant_1309" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="8723149" size="16" />
+		<layer id="199" name="Constant_1308" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="10259151" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="201" name="Reshape_1310" type="Reshape" version="opset1">
+		<layer id="200" name="Reshape_1309" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -2926,7 +2899,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="201" name="onnx::MatMul_1494/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
@@ -2937,7 +2925,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -2948,7 +2936,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/mlp/mlp.2/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -2961,14 +2949,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/mlp/mlp.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -2978,19 +2966,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.0/Add_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -3000,20 +2988,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.1/attn_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="206" name="onnx::Add_1519" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="8723165" size="3072" />
+			<data element_type="f32" shape="768" offset="10259167" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1519">
 					<dim>768</dim>
@@ -3021,7 +3009,7 @@
 			</output>
 		</layer>
 		<layer id="207" name="onnx::Add_1496" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="8726237" size="3072" />
+			<data element_type="f32" shape="768" offset="10262239" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1496">
 					<dim>768</dim>
@@ -3029,13 +3017,13 @@
 			</output>
 		</layer>
 		<layer id="208" name="Constant_325" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="209" name="Constant_324" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -3064,7 +3052,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -3074,13 +3062,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="213" name="blocks.1.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="8729309" size="3072" />
+			<data element_type="f32" shape="768" offset="10265311" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.1.attn_ln.weight">
 					<dim>768</dim>
@@ -3092,7 +3080,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -3102,13 +3090,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="215" name="blocks.1.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="8732381" size="3072" />
+			<data element_type="f32" shape="768" offset="10268383" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.1.attn_ln.bias">
 					<dim>768</dim>
@@ -3120,7 +3108,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -3130,7 +3118,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -3140,20 +3128,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.1/attn_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="218" name="onnx::MatMul_1497" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="8735453" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="10271455" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -3163,7 +3151,7 @@
 			</output>
 		</layer>
 		<layer id="219" name="Convert_1334" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -3172,7 +3160,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -3180,7 +3168,7 @@
 			</output>
 		</layer>
 		<layer id="220" name="onnx::MatMul_1497/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="10566367" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -3188,32 +3176,32 @@
 			</output>
 		</layer>
 		<layer id="221" name="Convert_1336" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="222" name="Subtract_1337" type="Subtract" version="opset1">
+		<layer id="222" name="onnx::MatMul_1497/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -3221,7 +3209,7 @@
 			</output>
 		</layer>
 		<layer id="223" name="onnx::MatMul_1497/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="9030365" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="10566368" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -3230,57 +3218,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="224" name="onnx::MatMul_1497/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="225" name="onnx::MatMul_1497/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="224" name="onnx::MatMul_1497/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="226" name="Constant_1342" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="225" name="Constant_1341" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="227" name="Reshape_1343" type="Reshape" version="opset1">
+		<layer id="226" name="Reshape_1342" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -3290,7 +3261,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="227" name="onnx::MatMul_1497/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -3301,7 +3287,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -3312,7 +3298,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/query/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -3325,20 +3311,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/query/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="230" name="/blocks.1/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.1/attn/Constant_2_output_0">
 					<dim>4</dim>
@@ -3350,7 +3336,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -3360,14 +3346,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/Reshape_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="232" name="Constant_341" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -3378,7 +3364,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -3390,19 +3376,19 @@
 				<port id="2" precision="FP32" names="/blocks.1/attn/Transpose_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="234" name="/blocks.1/attn/Constant" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353109" size="4" />
+			<data element_type="f32" shape="" offset="6889109" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.1/attn/Constant_output_0" />
 			</output>
 		</layer>
 		<layer id="235" name="/blocks.1/attn/Constant_1" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353113" size="4" />
+			<data element_type="f32" shape="" offset="6889113" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.1/attn/Constant_1_output_0" />
 			</output>
@@ -3423,7 +3409,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -3432,13 +3418,13 @@
 				<port id="2" precision="FP32" names="/blocks.1/attn/Mul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="238" name="onnx::MatMul_1499" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="9039581" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="10575584" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -3448,7 +3434,7 @@
 			</output>
 		</layer>
 		<layer id="239" name="Convert_1323" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -3457,7 +3443,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -3465,7 +3451,7 @@
 			</output>
 		</layer>
 		<layer id="240" name="onnx::MatMul_1499/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -3473,32 +3459,32 @@
 			</output>
 		</layer>
 		<layer id="241" name="Convert_1325" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="242" name="Subtract_1326" type="Subtract" version="opset1">
+		<layer id="242" name="onnx::MatMul_1499/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -3506,7 +3492,7 @@
 			</output>
 		</layer>
 		<layer id="243" name="onnx::MatMul_1499/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="9334493" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="10870496" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -3515,57 +3501,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="244" name="onnx::MatMul_1499/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="245" name="onnx::MatMul_1499/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="244" name="onnx::MatMul_1499/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="246" name="Constant_1331" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="245" name="Constant_1330" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="247" name="Reshape_1332" type="Reshape" version="opset1">
+		<layer id="246" name="Reshape_1331" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -3575,7 +3544,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="247" name="onnx::MatMul_1499/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -3586,7 +3570,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -3597,13 +3581,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/key/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="249" name="/blocks.1/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.1/attn/Constant_3_output_0">
 					<dim>4</dim>
@@ -3615,7 +3599,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -3625,14 +3609,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/Reshape_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="251" name="Constant_346" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5657245" size="32" />
+			<data element_type="i64" shape="4" offset="7193246" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -3643,7 +3627,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -3656,7 +3640,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -3667,7 +3651,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -3676,7 +3660,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -3686,22 +3670,22 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/MatMul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -3711,16 +3695,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.1/attn/Cast_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -3730,16 +3714,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.1/attn/Softmax_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -3749,21 +3733,21 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.1/attn/Cast_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="258" name="onnx::Add_1501" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="9343709" size="3072" />
+			<data element_type="f32" shape="768" offset="10879712" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1501">
 					<dim>768</dim>
@@ -3771,7 +3755,7 @@
 			</output>
 		</layer>
 		<layer id="259" name="onnx::MatMul_1502" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="9346781" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="10882784" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -3781,7 +3765,7 @@
 			</output>
 		</layer>
 		<layer id="260" name="Convert_1312" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -3790,7 +3774,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -3798,7 +3782,7 @@
 			</output>
 		</layer>
 		<layer id="261" name="onnx::MatMul_1502/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -3806,32 +3790,32 @@
 			</output>
 		</layer>
 		<layer id="262" name="Convert_1314" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="263" name="Subtract_1315" type="Subtract" version="opset1">
+		<layer id="263" name="onnx::MatMul_1502/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -3839,7 +3823,7 @@
 			</output>
 		</layer>
 		<layer id="264" name="onnx::MatMul_1502/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="9641693" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="11177696" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -3848,57 +3832,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="265" name="onnx::MatMul_1502/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="266" name="onnx::MatMul_1502/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="265" name="onnx::MatMul_1502/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="267" name="Constant_1320" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="266" name="Constant_1319" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="268" name="Reshape_1321" type="Reshape" version="opset1">
+		<layer id="267" name="Reshape_1320" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -3908,7 +3875,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="268" name="onnx::MatMul_1502/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -3919,7 +3901,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -3930,7 +3912,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/value/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -3943,20 +3925,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/value/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="271" name="/blocks.1/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.1/attn/Constant_4_output_0">
 					<dim>4</dim>
@@ -3968,7 +3950,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -3978,14 +3960,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/Reshape_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="273" name="Constant_351" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -3996,7 +3978,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -4008,7 +3990,7 @@
 				<port id="2" precision="FP32" names="/blocks.1/attn/Transpose_2_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
@@ -4019,13 +4001,13 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</input>
@@ -4033,13 +4015,13 @@
 				<port id="2" precision="FP32" names="/blocks.1/attn/MatMul_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="276" name="Constant_358" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -4051,7 +4033,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -4061,7 +4043,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/Transpose_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -4072,7 +4054,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -4084,7 +4066,7 @@
 			</output>
 		</layer>
 		<layer id="279" name="/blocks.1/attn/Constant_6" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.1/attn/Constant_6_output_0">
 					<dim>1</dim>
@@ -4137,7 +4119,7 @@
 			</output>
 		</layer>
 		<layer id="284" name="/blocks.1/attn/Constant_5" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.1/attn/Constant_5_output_0">
 					<dim>1</dim>
@@ -4169,7 +4151,7 @@
 			</output>
 		</layer>
 		<layer id="286" name="/blocks.1/attn/Constant_8" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042740" size="8" />
+			<data element_type="i64" shape="1" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.1/attn/Constant_8_output_0">
 					<dim>1</dim>
@@ -4197,7 +4179,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -4208,13 +4190,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/Reshape_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="289" name="onnx::MatMul_1520" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="9650909" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="11186912" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -4224,7 +4206,7 @@
 			</output>
 		</layer>
 		<layer id="290" name="Convert_1345" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -4233,7 +4215,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -4241,7 +4223,7 @@
 			</output>
 		</layer>
 		<layer id="291" name="onnx::MatMul_1520/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="11481824" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -4249,32 +4231,32 @@
 			</output>
 		</layer>
 		<layer id="292" name="Convert_1347" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="293" name="Subtract_1348" type="Subtract" version="opset1">
+		<layer id="293" name="onnx::MatMul_1520/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -4282,7 +4264,7 @@
 			</output>
 		</layer>
 		<layer id="294" name="onnx::MatMul_1520/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="9945821" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="11481825" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -4291,57 +4273,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="295" name="onnx::MatMul_1520/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="296" name="onnx::MatMul_1520/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="295" name="onnx::MatMul_1520/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="297" name="Constant_1353" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="296" name="Constant_1352" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="298" name="Reshape_1354" type="Reshape" version="opset1">
+		<layer id="297" name="Reshape_1353" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -4351,7 +4316,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="298" name="onnx::MatMul_1520/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -4362,7 +4342,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -4373,7 +4353,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/out/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -4386,14 +4366,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/attn/out/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -4403,19 +4383,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -4425,20 +4405,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.1/mlp_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="303" name="onnx::Add_1525" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="9955037" size="3072" />
+			<data element_type="f32" shape="768" offset="11491041" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1525">
 					<dim>768</dim>
@@ -4446,7 +4426,7 @@
 			</output>
 		</layer>
 		<layer id="304" name="onnx::Add_1522" type="Const" version="opset1">
-			<data element_type="f32" shape="3072" offset="9958109" size="12288" />
+			<data element_type="f32" shape="3072" offset="11494113" size="12288" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1522">
 					<dim>3072</dim>
@@ -4454,13 +4434,13 @@
 			</output>
 		</layer>
 		<layer id="305" name="Constant_384" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="306" name="Constant_383" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -4489,7 +4469,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -4499,13 +4479,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="310" name="blocks.1.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="9970397" size="3072" />
+			<data element_type="f32" shape="768" offset="11506401" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.1.mlp_ln.weight">
 					<dim>768</dim>
@@ -4517,7 +4497,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -4527,13 +4507,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="312" name="blocks.1.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="9973469" size="3072" />
+			<data element_type="f32" shape="768" offset="11509473" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.1.mlp_ln.bias">
 					<dim>768</dim>
@@ -4545,7 +4525,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -4555,7 +4535,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/mlp_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -4565,20 +4545,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.1/mlp_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="315" name="onnx::MatMul_1523" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 3072" offset="9976541" size="1179648" />
+			<data element_type="u4" shape="6, 128, 3072" offset="11512545" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -4588,7 +4568,7 @@
 			</output>
 		</layer>
 		<layer id="316" name="Convert_1356" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -4597,7 +4577,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -4605,7 +4585,7 @@
 			</output>
 		</layer>
 		<layer id="317" name="onnx::MatMul_1523/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="12692193" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -4613,32 +4593,32 @@
 			</output>
 		</layer>
 		<layer id="318" name="Convert_1358" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="319" name="Subtract_1359" type="Subtract" version="opset1">
+		<layer id="319" name="onnx::MatMul_1523/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -4646,7 +4626,7 @@
 			</output>
 		</layer>
 		<layer id="320" name="onnx::MatMul_1523/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 3072" offset="11156189" size="36864" />
+			<data element_type="f16" shape="6, 1, 3072" offset="12692194" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -4655,57 +4635,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="321" name="onnx::MatMul_1523/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="322" name="onnx::MatMul_1523/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="321" name="onnx::MatMul_1523/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="323" name="Constant_1364" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="7506621" size="16" />
+		<layer id="322" name="Constant_1363" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="9042623" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="324" name="Reshape_1365" type="Reshape" version="opset1">
+		<layer id="323" name="Reshape_1364" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -4715,7 +4678,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="324" name="onnx::MatMul_1523/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
@@ -4726,7 +4704,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -4737,7 +4715,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/mlp/mlp.0/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -4750,14 +4728,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/mlp/mlp.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -4773,7 +4751,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -4781,7 +4759,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/mlp/mlp.1/Div_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -4790,14 +4768,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.1/mlp/mlp.1/Erf_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -4813,7 +4791,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -4821,7 +4799,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/mlp/mlp.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -4831,19 +4809,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/mlp/mlp.1/Mul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -4859,7 +4837,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -4867,13 +4845,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/mlp/mlp.1/Mul_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="335" name="onnx::MatMul_1526" type="Const" version="opset1">
-			<data element_type="u4" shape="24, 128, 768" offset="11193053" size="1179648" />
+			<data element_type="u4" shape="24, 128, 768" offset="12729058" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -4883,7 +4861,7 @@
 			</output>
 		</layer>
 		<layer id="336" name="Convert_1367" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -4892,7 +4870,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -4900,7 +4878,7 @@
 			</output>
 		</layer>
 		<layer id="337" name="onnx::MatMul_1526/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -4908,32 +4886,32 @@
 			</output>
 		</layer>
 		<layer id="338" name="Convert_1369" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="339" name="Subtract_1370" type="Subtract" version="opset1">
+		<layer id="339" name="onnx::MatMul_1526/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -4941,7 +4919,7 @@
 			</output>
 		</layer>
 		<layer id="340" name="onnx::MatMul_1526/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="24, 1, 768" offset="12372701" size="36864" />
+			<data element_type="f16" shape="24, 1, 768" offset="13908706" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>24</dim>
@@ -4950,57 +4928,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="341" name="onnx::MatMul_1526/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="342" name="onnx::MatMul_1526/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="341" name="onnx::MatMul_1526/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="343" name="Constant_1375" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="8723149" size="16" />
+		<layer id="342" name="Constant_1374" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="10259151" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="344" name="Reshape_1376" type="Reshape" version="opset1">
+		<layer id="343" name="Reshape_1375" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -5010,7 +4971,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="344" name="onnx::MatMul_1526/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
@@ -5021,7 +4997,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -5032,7 +5008,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/mlp/mlp.2/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -5045,14 +5021,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/mlp/mlp.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -5062,19 +5038,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.1/Add_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -5084,20 +5060,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.2/attn_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="349" name="onnx::Add_1551" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="12409565" size="3072" />
+			<data element_type="f32" shape="768" offset="13945570" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1551">
 					<dim>768</dim>
@@ -5105,7 +5081,7 @@
 			</output>
 		</layer>
 		<layer id="350" name="onnx::Add_1528" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="12412637" size="3072" />
+			<data element_type="f32" shape="768" offset="13948642" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1528">
 					<dim>768</dim>
@@ -5113,13 +5089,13 @@
 			</output>
 		</layer>
 		<layer id="351" name="Constant_406" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="352" name="Constant_405" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -5148,7 +5124,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -5158,13 +5134,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="356" name="blocks.2.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="12415709" size="3072" />
+			<data element_type="f32" shape="768" offset="13951714" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.2.attn_ln.weight">
 					<dim>768</dim>
@@ -5176,7 +5152,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -5186,13 +5162,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="358" name="blocks.2.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="12418781" size="3072" />
+			<data element_type="f32" shape="768" offset="13954786" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.2.attn_ln.bias">
 					<dim>768</dim>
@@ -5204,7 +5180,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -5214,7 +5190,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -5224,20 +5200,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.2/attn_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="361" name="onnx::MatMul_1529" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="12421853" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="13957858" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -5247,7 +5223,7 @@
 			</output>
 		</layer>
 		<layer id="362" name="Convert_1400" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -5256,7 +5232,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -5264,7 +5240,7 @@
 			</output>
 		</layer>
 		<layer id="363" name="onnx::MatMul_1529/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -5272,32 +5248,32 @@
 			</output>
 		</layer>
 		<layer id="364" name="Convert_1402" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="365" name="Subtract_1403" type="Subtract" version="opset1">
+		<layer id="365" name="onnx::MatMul_1529/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -5305,7 +5281,7 @@
 			</output>
 		</layer>
 		<layer id="366" name="onnx::MatMul_1529/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="12716765" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="14252770" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -5314,57 +5290,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="367" name="onnx::MatMul_1529/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="368" name="onnx::MatMul_1529/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="367" name="onnx::MatMul_1529/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="369" name="Constant_1408" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="368" name="Constant_1407" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="370" name="Reshape_1409" type="Reshape" version="opset1">
+		<layer id="369" name="Reshape_1408" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -5374,7 +5333,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="370" name="onnx::MatMul_1529/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -5385,7 +5359,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -5396,7 +5370,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/query/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -5409,20 +5383,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/query/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="373" name="/blocks.2/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.2/attn/Constant_2_output_0">
 					<dim>4</dim>
@@ -5434,7 +5408,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -5444,14 +5418,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/Reshape_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="375" name="Constant_422" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -5462,7 +5436,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -5474,19 +5448,19 @@
 				<port id="2" precision="FP32" names="/blocks.2/attn/Transpose_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="377" name="/blocks.2/attn/Constant" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353109" size="4" />
+			<data element_type="f32" shape="" offset="6889109" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.2/attn/Constant_output_0" />
 			</output>
 		</layer>
 		<layer id="378" name="/blocks.2/attn/Constant_1" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353113" size="4" />
+			<data element_type="f32" shape="" offset="6889113" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.2/attn/Constant_1_output_0" />
 			</output>
@@ -5507,7 +5481,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -5516,13 +5490,13 @@
 				<port id="2" precision="FP32" names="/blocks.2/attn/Mul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="381" name="onnx::MatMul_1531" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="12725981" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="14261986" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -5532,7 +5506,7 @@
 			</output>
 		</layer>
 		<layer id="382" name="Convert_1389" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -5541,7 +5515,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -5549,7 +5523,7 @@
 			</output>
 		</layer>
 		<layer id="383" name="onnx::MatMul_1531/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -5557,32 +5531,32 @@
 			</output>
 		</layer>
 		<layer id="384" name="Convert_1391" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="385" name="Subtract_1392" type="Subtract" version="opset1">
+		<layer id="385" name="onnx::MatMul_1531/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -5590,7 +5564,7 @@
 			</output>
 		</layer>
 		<layer id="386" name="onnx::MatMul_1531/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="13020893" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="14556898" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -5599,57 +5573,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="387" name="onnx::MatMul_1531/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="388" name="onnx::MatMul_1531/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="387" name="onnx::MatMul_1531/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="389" name="Constant_1397" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="388" name="Constant_1396" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="390" name="Reshape_1398" type="Reshape" version="opset1">
+		<layer id="389" name="Reshape_1397" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -5659,7 +5616,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="390" name="onnx::MatMul_1531/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -5670,7 +5642,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -5681,13 +5653,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/key/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="392" name="/blocks.2/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.2/attn/Constant_3_output_0">
 					<dim>4</dim>
@@ -5699,7 +5671,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -5709,14 +5681,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/Reshape_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="394" name="Constant_427" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5657245" size="32" />
+			<data element_type="i64" shape="4" offset="7193246" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -5727,7 +5699,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -5740,7 +5712,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -5751,7 +5723,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -5760,7 +5732,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -5770,22 +5742,22 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/MatMul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -5795,16 +5767,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.2/attn/Cast_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -5814,16 +5786,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.2/attn/Softmax_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -5833,21 +5805,21 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.2/attn/Cast_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="401" name="onnx::Add_1533" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="13030109" size="3072" />
+			<data element_type="f32" shape="768" offset="14566114" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1533">
 					<dim>768</dim>
@@ -5855,7 +5827,7 @@
 			</output>
 		</layer>
 		<layer id="402" name="onnx::MatMul_1534" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="13033181" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="14569186" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -5865,7 +5837,7 @@
 			</output>
 		</layer>
 		<layer id="403" name="Convert_1378" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -5874,7 +5846,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -5882,7 +5854,7 @@
 			</output>
 		</layer>
 		<layer id="404" name="onnx::MatMul_1534/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="14864098" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -5890,32 +5862,32 @@
 			</output>
 		</layer>
 		<layer id="405" name="Convert_1380" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="406" name="Subtract_1381" type="Subtract" version="opset1">
+		<layer id="406" name="onnx::MatMul_1534/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -5923,7 +5895,7 @@
 			</output>
 		</layer>
 		<layer id="407" name="onnx::MatMul_1534/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="13328093" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="14864099" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -5932,57 +5904,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="408" name="onnx::MatMul_1534/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="409" name="onnx::MatMul_1534/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="408" name="onnx::MatMul_1534/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="410" name="Constant_1386" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="409" name="Constant_1385" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="411" name="Reshape_1387" type="Reshape" version="opset1">
+		<layer id="410" name="Reshape_1386" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -5992,7 +5947,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="411" name="onnx::MatMul_1534/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -6003,7 +5973,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -6014,7 +5984,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/value/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -6027,20 +5997,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/value/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="414" name="/blocks.2/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.2/attn/Constant_4_output_0">
 					<dim>4</dim>
@@ -6052,7 +6022,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -6062,14 +6032,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/Reshape_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="416" name="Constant_432" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -6080,7 +6050,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -6092,7 +6062,7 @@
 				<port id="2" precision="FP32" names="/blocks.2/attn/Transpose_2_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
@@ -6103,13 +6073,13 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</input>
@@ -6117,13 +6087,13 @@
 				<port id="2" precision="FP32" names="/blocks.2/attn/MatMul_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="419" name="Constant_439" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -6135,7 +6105,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -6145,7 +6115,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/Transpose_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -6156,7 +6126,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -6168,7 +6138,7 @@
 			</output>
 		</layer>
 		<layer id="422" name="/blocks.2/attn/Constant_6" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.2/attn/Constant_6_output_0">
 					<dim>1</dim>
@@ -6221,7 +6191,7 @@
 			</output>
 		</layer>
 		<layer id="427" name="/blocks.2/attn/Constant_5" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.2/attn/Constant_5_output_0">
 					<dim>1</dim>
@@ -6253,7 +6223,7 @@
 			</output>
 		</layer>
 		<layer id="429" name="/blocks.2/attn/Constant_8" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042740" size="8" />
+			<data element_type="i64" shape="1" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.2/attn/Constant_8_output_0">
 					<dim>1</dim>
@@ -6281,7 +6251,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -6292,13 +6262,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/Reshape_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="432" name="onnx::MatMul_1552" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="13337309" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="14873315" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -6308,7 +6278,7 @@
 			</output>
 		</layer>
 		<layer id="433" name="Convert_1411" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -6317,7 +6287,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -6325,7 +6295,7 @@
 			</output>
 		</layer>
 		<layer id="434" name="onnx::MatMul_1552/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="12692193" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -6333,32 +6303,32 @@
 			</output>
 		</layer>
 		<layer id="435" name="Convert_1413" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="436" name="Subtract_1414" type="Subtract" version="opset1">
+		<layer id="436" name="onnx::MatMul_1552/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -6366,7 +6336,7 @@
 			</output>
 		</layer>
 		<layer id="437" name="onnx::MatMul_1552/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="13632221" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="15168227" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -6375,57 +6345,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="438" name="onnx::MatMul_1552/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="439" name="onnx::MatMul_1552/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="438" name="onnx::MatMul_1552/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="440" name="Constant_1419" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="439" name="Constant_1418" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="441" name="Reshape_1420" type="Reshape" version="opset1">
+		<layer id="440" name="Reshape_1419" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -6435,7 +6388,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="441" name="onnx::MatMul_1552/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -6446,7 +6414,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -6457,7 +6425,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/out/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -6470,14 +6438,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/attn/out/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -6487,19 +6455,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -6509,20 +6477,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.2/mlp_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="446" name="onnx::Add_1557" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="13641437" size="3072" />
+			<data element_type="f32" shape="768" offset="15177443" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1557">
 					<dim>768</dim>
@@ -6530,7 +6498,7 @@
 			</output>
 		</layer>
 		<layer id="447" name="onnx::Add_1554" type="Const" version="opset1">
-			<data element_type="f32" shape="3072" offset="13644509" size="12288" />
+			<data element_type="f32" shape="3072" offset="15180515" size="12288" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1554">
 					<dim>3072</dim>
@@ -6538,13 +6506,13 @@
 			</output>
 		</layer>
 		<layer id="448" name="Constant_465" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="449" name="Constant_464" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -6573,7 +6541,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -6583,13 +6551,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="453" name="blocks.2.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="13656797" size="3072" />
+			<data element_type="f32" shape="768" offset="15192803" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.2.mlp_ln.weight">
 					<dim>768</dim>
@@ -6601,7 +6569,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -6611,13 +6579,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="455" name="blocks.2.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="13659869" size="3072" />
+			<data element_type="f32" shape="768" offset="15195875" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.2.mlp_ln.bias">
 					<dim>768</dim>
@@ -6629,7 +6597,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -6639,7 +6607,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/mlp_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -6649,20 +6617,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.2/mlp_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="458" name="onnx::MatMul_1555" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 3072" offset="13662941" size="1179648" />
+			<data element_type="u4" shape="6, 128, 3072" offset="15198947" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -6672,7 +6640,7 @@
 			</output>
 		</layer>
 		<layer id="459" name="Convert_1422" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -6681,7 +6649,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -6689,7 +6657,7 @@
 			</output>
 		</layer>
 		<layer id="460" name="onnx::MatMul_1555/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="11481824" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -6697,32 +6665,32 @@
 			</output>
 		</layer>
 		<layer id="461" name="Convert_1424" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="462" name="Subtract_1425" type="Subtract" version="opset1">
+		<layer id="462" name="onnx::MatMul_1555/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -6730,7 +6698,7 @@
 			</output>
 		</layer>
 		<layer id="463" name="onnx::MatMul_1555/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 3072" offset="14842589" size="36864" />
+			<data element_type="f16" shape="6, 1, 3072" offset="16378595" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -6739,57 +6707,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="464" name="onnx::MatMul_1555/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="465" name="onnx::MatMul_1555/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="464" name="onnx::MatMul_1555/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="466" name="Constant_1430" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="7506621" size="16" />
+		<layer id="465" name="Constant_1429" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="9042623" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="467" name="Reshape_1431" type="Reshape" version="opset1">
+		<layer id="466" name="Reshape_1430" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -6799,7 +6750,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="467" name="onnx::MatMul_1555/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
@@ -6810,7 +6776,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -6821,7 +6787,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/mlp/mlp.0/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -6834,14 +6800,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/mlp/mlp.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -6857,7 +6823,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -6865,7 +6831,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/mlp/mlp.1/Div_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -6874,14 +6840,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.2/mlp/mlp.1/Erf_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -6897,7 +6863,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -6905,7 +6871,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/mlp/mlp.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -6915,19 +6881,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/mlp/mlp.1/Mul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -6943,7 +6909,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -6951,13 +6917,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/mlp/mlp.1/Mul_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="478" name="onnx::MatMul_1558" type="Const" version="opset1">
-			<data element_type="u4" shape="24, 128, 768" offset="14879453" size="1179648" />
+			<data element_type="u4" shape="24, 128, 768" offset="16415459" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -6967,7 +6933,7 @@
 			</output>
 		</layer>
 		<layer id="479" name="Convert_1433" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -6976,7 +6942,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -6984,7 +6950,7 @@
 			</output>
 		</layer>
 		<layer id="480" name="onnx::MatMul_1558/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -6992,32 +6958,32 @@
 			</output>
 		</layer>
 		<layer id="481" name="Convert_1435" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="482" name="Subtract_1436" type="Subtract" version="opset1">
+		<layer id="482" name="onnx::MatMul_1558/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -7025,7 +6991,7 @@
 			</output>
 		</layer>
 		<layer id="483" name="onnx::MatMul_1558/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="24, 1, 768" offset="16059101" size="36864" />
+			<data element_type="f16" shape="24, 1, 768" offset="17595107" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>24</dim>
@@ -7034,57 +7000,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="484" name="onnx::MatMul_1558/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="485" name="onnx::MatMul_1558/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="484" name="onnx::MatMul_1558/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="486" name="Constant_1441" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="8723149" size="16" />
+		<layer id="485" name="Constant_1440" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="10259151" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="487" name="Reshape_1442" type="Reshape" version="opset1">
+		<layer id="486" name="Reshape_1441" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -7094,7 +7043,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="487" name="onnx::MatMul_1558/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
@@ -7105,7 +7069,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -7116,7 +7080,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/mlp/mlp.2/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -7129,14 +7093,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/mlp/mlp.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -7146,19 +7110,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.2/Add_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -7168,20 +7132,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.3/attn_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="492" name="onnx::Add_1583" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="16095965" size="3072" />
+			<data element_type="f32" shape="768" offset="17631971" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1583">
 					<dim>768</dim>
@@ -7189,7 +7153,7 @@
 			</output>
 		</layer>
 		<layer id="493" name="onnx::Add_1560" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="16099037" size="3072" />
+			<data element_type="f32" shape="768" offset="17635043" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1560">
 					<dim>768</dim>
@@ -7197,13 +7161,13 @@
 			</output>
 		</layer>
 		<layer id="494" name="Constant_487" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="495" name="Constant_486" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -7232,7 +7196,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -7242,13 +7206,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="499" name="blocks.3.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="16102109" size="3072" />
+			<data element_type="f32" shape="768" offset="17638115" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.3.attn_ln.weight">
 					<dim>768</dim>
@@ -7260,7 +7224,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -7270,13 +7234,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="501" name="blocks.3.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="16105181" size="3072" />
+			<data element_type="f32" shape="768" offset="17641187" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.3.attn_ln.bias">
 					<dim>768</dim>
@@ -7288,7 +7252,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -7298,7 +7262,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -7308,20 +7272,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.3/attn_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="504" name="onnx::MatMul_1561" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="16108253" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="17644259" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -7331,7 +7295,7 @@
 			</output>
 		</layer>
 		<layer id="505" name="Convert_1466" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -7340,7 +7304,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -7348,7 +7312,7 @@
 			</output>
 		</layer>
 		<layer id="506" name="onnx::MatMul_1561/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="10566367" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -7356,32 +7320,32 @@
 			</output>
 		</layer>
 		<layer id="507" name="Convert_1468" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="508" name="Subtract_1469" type="Subtract" version="opset1">
+		<layer id="508" name="onnx::MatMul_1561/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -7389,7 +7353,7 @@
 			</output>
 		</layer>
 		<layer id="509" name="onnx::MatMul_1561/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="16403165" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="17939171" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -7398,57 +7362,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="510" name="onnx::MatMul_1561/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="511" name="onnx::MatMul_1561/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="510" name="onnx::MatMul_1561/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="512" name="Constant_1474" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="511" name="Constant_1473" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="513" name="Reshape_1475" type="Reshape" version="opset1">
+		<layer id="512" name="Reshape_1474" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -7458,7 +7405,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="513" name="onnx::MatMul_1561/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -7469,7 +7431,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -7480,7 +7442,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/query/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -7493,20 +7455,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/query/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="516" name="/blocks.3/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.3/attn/Constant_2_output_0">
 					<dim>4</dim>
@@ -7518,7 +7480,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -7528,14 +7490,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/Reshape_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="518" name="Constant_503" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -7546,7 +7508,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -7558,19 +7520,19 @@
 				<port id="2" precision="FP32" names="/blocks.3/attn/Transpose_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="520" name="/blocks.3/attn/Constant" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353109" size="4" />
+			<data element_type="f32" shape="" offset="6889109" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.3/attn/Constant_output_0" />
 			</output>
 		</layer>
 		<layer id="521" name="/blocks.3/attn/Constant_1" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353113" size="4" />
+			<data element_type="f32" shape="" offset="6889113" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.3/attn/Constant_1_output_0" />
 			</output>
@@ -7591,7 +7553,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -7600,13 +7562,13 @@
 				<port id="2" precision="FP32" names="/blocks.3/attn/Mul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="524" name="onnx::MatMul_1563" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="16412381" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="17948387" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -7616,7 +7578,7 @@
 			</output>
 		</layer>
 		<layer id="525" name="Convert_1455" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -7625,7 +7587,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -7633,7 +7595,7 @@
 			</output>
 		</layer>
 		<layer id="526" name="onnx::MatMul_1563/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -7641,32 +7603,32 @@
 			</output>
 		</layer>
 		<layer id="527" name="Convert_1457" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="528" name="Subtract_1458" type="Subtract" version="opset1">
+		<layer id="528" name="onnx::MatMul_1563/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -7674,7 +7636,7 @@
 			</output>
 		</layer>
 		<layer id="529" name="onnx::MatMul_1563/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="16707293" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="18243299" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -7683,57 +7645,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="530" name="onnx::MatMul_1563/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="531" name="onnx::MatMul_1563/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="530" name="onnx::MatMul_1563/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="532" name="Constant_1463" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="531" name="Constant_1462" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="533" name="Reshape_1464" type="Reshape" version="opset1">
+		<layer id="532" name="Reshape_1463" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -7743,7 +7688,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="533" name="onnx::MatMul_1563/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -7754,7 +7714,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -7765,13 +7725,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/key/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="535" name="/blocks.3/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.3/attn/Constant_3_output_0">
 					<dim>4</dim>
@@ -7783,7 +7743,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -7793,14 +7753,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/Reshape_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="537" name="Constant_508" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5657245" size="32" />
+			<data element_type="i64" shape="4" offset="7193246" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -7811,7 +7771,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -7824,7 +7784,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -7835,7 +7795,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -7844,7 +7804,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -7854,22 +7814,22 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/MatMul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -7879,16 +7839,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.3/attn/Cast_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -7898,16 +7858,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.3/attn/Softmax_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -7917,21 +7877,21 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.3/attn/Cast_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="544" name="onnx::Add_1565" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="16716509" size="3072" />
+			<data element_type="f32" shape="768" offset="18252515" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1565">
 					<dim>768</dim>
@@ -7939,7 +7899,7 @@
 			</output>
 		</layer>
 		<layer id="545" name="onnx::MatMul_1566" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="16719581" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="18255587" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -7949,7 +7909,7 @@
 			</output>
 		</layer>
 		<layer id="546" name="Convert_1444" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -7958,7 +7918,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -7966,7 +7926,7 @@
 			</output>
 		</layer>
 		<layer id="547" name="onnx::MatMul_1566/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -7974,32 +7934,32 @@
 			</output>
 		</layer>
 		<layer id="548" name="Convert_1446" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="549" name="Subtract_1447" type="Subtract" version="opset1">
+		<layer id="549" name="onnx::MatMul_1566/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -8007,7 +7967,7 @@
 			</output>
 		</layer>
 		<layer id="550" name="onnx::MatMul_1566/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="17014493" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="18550499" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -8016,57 +7976,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="551" name="onnx::MatMul_1566/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="552" name="onnx::MatMul_1566/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="551" name="onnx::MatMul_1566/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="553" name="Constant_1452" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="552" name="Constant_1451" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="554" name="Reshape_1453" type="Reshape" version="opset1">
+		<layer id="553" name="Reshape_1452" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -8076,7 +8019,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="554" name="onnx::MatMul_1566/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -8087,7 +8045,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -8098,7 +8056,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/value/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -8111,20 +8069,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/value/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="557" name="/blocks.3/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.3/attn/Constant_4_output_0">
 					<dim>4</dim>
@@ -8136,7 +8094,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -8146,14 +8104,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/Reshape_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="559" name="Constant_513" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -8164,7 +8122,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -8176,7 +8134,7 @@
 				<port id="2" precision="FP32" names="/blocks.3/attn/Transpose_2_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
@@ -8187,13 +8145,13 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</input>
@@ -8201,13 +8159,13 @@
 				<port id="2" precision="FP32" names="/blocks.3/attn/MatMul_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="562" name="Constant_520" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -8219,7 +8177,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -8229,7 +8187,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/Transpose_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -8240,7 +8198,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -8252,7 +8210,7 @@
 			</output>
 		</layer>
 		<layer id="565" name="/blocks.3/attn/Constant_6" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.3/attn/Constant_6_output_0">
 					<dim>1</dim>
@@ -8305,7 +8263,7 @@
 			</output>
 		</layer>
 		<layer id="570" name="/blocks.3/attn/Constant_5" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.3/attn/Constant_5_output_0">
 					<dim>1</dim>
@@ -8337,7 +8295,7 @@
 			</output>
 		</layer>
 		<layer id="572" name="/blocks.3/attn/Constant_8" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042740" size="8" />
+			<data element_type="i64" shape="1" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.3/attn/Constant_8_output_0">
 					<dim>1</dim>
@@ -8365,7 +8323,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -8376,13 +8334,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/Reshape_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="575" name="onnx::MatMul_1584" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="17023709" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="18559715" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -8392,7 +8350,7 @@
 			</output>
 		</layer>
 		<layer id="576" name="Convert_1477" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -8401,7 +8359,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -8409,7 +8367,7 @@
 			</output>
 		</layer>
 		<layer id="577" name="onnx::MatMul_1584/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="10566367" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -8417,32 +8375,32 @@
 			</output>
 		</layer>
 		<layer id="578" name="Convert_1479" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="579" name="Subtract_1480" type="Subtract" version="opset1">
+		<layer id="579" name="onnx::MatMul_1584/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -8450,7 +8408,7 @@
 			</output>
 		</layer>
 		<layer id="580" name="onnx::MatMul_1584/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="17318621" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="18854627" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -8459,57 +8417,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="581" name="onnx::MatMul_1584/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="581" name="onnx::MatMul_1584/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
+					<dim>6</dim>
+					<dim>128</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
-					<dim>1</dim>
+					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="582" name="onnx::MatMul_1584/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>6</dim>
-					<dim>128</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="2" precision="FP32">
-					<dim>6</dim>
-					<dim>128</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="583" name="Constant_1485" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="582" name="Constant_1484" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="584" name="Reshape_1486" type="Reshape" version="opset1">
+		<layer id="583" name="Reshape_1485" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -8519,7 +8460,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="584" name="onnx::MatMul_1584/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -8530,7 +8486,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -8541,7 +8497,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/out/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -8554,14 +8510,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/attn/out/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -8571,19 +8527,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -8593,20 +8549,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.3/mlp_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="589" name="onnx::Add_1589" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="17327837" size="3072" />
+			<data element_type="f32" shape="768" offset="18863843" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1589">
 					<dim>768</dim>
@@ -8614,7 +8570,7 @@
 			</output>
 		</layer>
 		<layer id="590" name="onnx::Add_1586" type="Const" version="opset1">
-			<data element_type="f32" shape="3072" offset="17330909" size="12288" />
+			<data element_type="f32" shape="3072" offset="18866915" size="12288" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1586">
 					<dim>3072</dim>
@@ -8622,13 +8578,13 @@
 			</output>
 		</layer>
 		<layer id="591" name="Constant_546" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="592" name="Constant_545" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -8657,7 +8613,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -8667,13 +8623,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="596" name="blocks.3.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="17343197" size="3072" />
+			<data element_type="f32" shape="768" offset="18879203" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.3.mlp_ln.weight">
 					<dim>768</dim>
@@ -8685,7 +8641,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -8695,13 +8651,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="598" name="blocks.3.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="17346269" size="3072" />
+			<data element_type="f32" shape="768" offset="18882275" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.3.mlp_ln.bias">
 					<dim>768</dim>
@@ -8713,7 +8669,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -8723,7 +8679,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/mlp_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -8733,20 +8689,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.3/mlp_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="601" name="onnx::MatMul_1587" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 3072" offset="17349341" size="1179648" />
+			<data element_type="u4" shape="6, 128, 3072" offset="18885347" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -8756,7 +8712,7 @@
 			</output>
 		</layer>
 		<layer id="602" name="Convert_1488" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -8765,7 +8721,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -8773,7 +8729,7 @@
 			</output>
 		</layer>
 		<layer id="603" name="onnx::MatMul_1587/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="20064995" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -8781,32 +8737,32 @@
 			</output>
 		</layer>
 		<layer id="604" name="Convert_1490" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="605" name="Subtract_1491" type="Subtract" version="opset1">
+		<layer id="605" name="onnx::MatMul_1587/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -8814,7 +8770,7 @@
 			</output>
 		</layer>
 		<layer id="606" name="onnx::MatMul_1587/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 3072" offset="18528989" size="36864" />
+			<data element_type="f16" shape="6, 1, 3072" offset="20064996" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -8823,57 +8779,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="607" name="onnx::MatMul_1587/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="608" name="onnx::MatMul_1587/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="607" name="onnx::MatMul_1587/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="609" name="Constant_1496" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="7506621" size="16" />
+		<layer id="608" name="Constant_1495" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="9042623" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="610" name="Reshape_1497" type="Reshape" version="opset1">
+		<layer id="609" name="Reshape_1496" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -8883,7 +8822,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="610" name="onnx::MatMul_1587/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
@@ -8894,7 +8848,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -8905,7 +8859,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/mlp/mlp.0/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -8918,14 +8872,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/mlp/mlp.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -8941,7 +8895,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -8949,7 +8903,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/mlp/mlp.1/Div_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -8958,14 +8912,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.3/mlp/mlp.1/Erf_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -8981,7 +8935,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -8989,7 +8943,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/mlp/mlp.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -8999,19 +8953,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/mlp/mlp.1/Mul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -9027,7 +8981,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -9035,13 +8989,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/mlp/mlp.1/Mul_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="621" name="onnx::MatMul_1590" type="Const" version="opset1">
-			<data element_type="u4" shape="24, 128, 768" offset="18565853" size="1179648" />
+			<data element_type="u4" shape="24, 128, 768" offset="20101860" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -9051,7 +9005,7 @@
 			</output>
 		</layer>
 		<layer id="622" name="Convert_1499" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -9060,7 +9014,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -9068,7 +9022,7 @@
 			</output>
 		</layer>
 		<layer id="623" name="onnx::MatMul_1590/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -9076,32 +9030,32 @@
 			</output>
 		</layer>
 		<layer id="624" name="Convert_1501" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="625" name="Subtract_1502" type="Subtract" version="opset1">
+		<layer id="625" name="onnx::MatMul_1590/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -9109,7 +9063,7 @@
 			</output>
 		</layer>
 		<layer id="626" name="onnx::MatMul_1590/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="24, 1, 768" offset="19745501" size="36864" />
+			<data element_type="f16" shape="24, 1, 768" offset="21281508" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>24</dim>
@@ -9118,57 +9072,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="627" name="onnx::MatMul_1590/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="628" name="onnx::MatMul_1590/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="627" name="onnx::MatMul_1590/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="629" name="Constant_1507" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="8723149" size="16" />
+		<layer id="628" name="Constant_1506" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="10259151" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="630" name="Reshape_1508" type="Reshape" version="opset1">
+		<layer id="629" name="Reshape_1507" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -9178,7 +9115,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="630" name="onnx::MatMul_1590/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
@@ -9189,7 +9141,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -9200,7 +9152,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/mlp/mlp.2/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -9213,14 +9165,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/mlp/mlp.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -9230,19 +9182,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.3/Add_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -9252,20 +9204,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.4/attn_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="635" name="onnx::Add_1615" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="19782365" size="3072" />
+			<data element_type="f32" shape="768" offset="21318372" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1615">
 					<dim>768</dim>
@@ -9273,7 +9225,7 @@
 			</output>
 		</layer>
 		<layer id="636" name="onnx::Add_1592" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="19785437" size="3072" />
+			<data element_type="f32" shape="768" offset="21321444" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1592">
 					<dim>768</dim>
@@ -9281,13 +9233,13 @@
 			</output>
 		</layer>
 		<layer id="637" name="Constant_568" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="638" name="Constant_567" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -9316,7 +9268,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -9326,13 +9278,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="642" name="blocks.4.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="19788509" size="3072" />
+			<data element_type="f32" shape="768" offset="21324516" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.4.attn_ln.weight">
 					<dim>768</dim>
@@ -9344,7 +9296,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -9354,13 +9306,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="644" name="blocks.4.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="19791581" size="3072" />
+			<data element_type="f32" shape="768" offset="21327588" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.4.attn_ln.bias">
 					<dim>768</dim>
@@ -9372,7 +9324,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -9382,7 +9334,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -9392,20 +9344,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.4/attn_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="647" name="onnx::MatMul_1593" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="19794653" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="21330660" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -9415,7 +9367,7 @@
 			</output>
 		</layer>
 		<layer id="648" name="Convert_1532" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -9424,7 +9376,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -9432,7 +9384,7 @@
 			</output>
 		</layer>
 		<layer id="649" name="onnx::MatMul_1593/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -9440,32 +9392,32 @@
 			</output>
 		</layer>
 		<layer id="650" name="Convert_1534" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="651" name="Subtract_1535" type="Subtract" version="opset1">
+		<layer id="651" name="onnx::MatMul_1593/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -9473,7 +9425,7 @@
 			</output>
 		</layer>
 		<layer id="652" name="onnx::MatMul_1593/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="20089565" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="21625572" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -9482,57 +9434,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="653" name="onnx::MatMul_1593/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="654" name="onnx::MatMul_1593/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="653" name="onnx::MatMul_1593/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="655" name="Constant_1540" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="654" name="Constant_1539" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="656" name="Reshape_1541" type="Reshape" version="opset1">
+		<layer id="655" name="Reshape_1540" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -9542,7 +9477,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="656" name="onnx::MatMul_1593/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -9553,7 +9503,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -9564,7 +9514,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/query/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -9577,20 +9527,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/query/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="659" name="/blocks.4/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.4/attn/Constant_2_output_0">
 					<dim>4</dim>
@@ -9602,7 +9552,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -9612,14 +9562,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/Reshape_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="661" name="Constant_584" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -9630,7 +9580,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -9642,19 +9592,19 @@
 				<port id="2" precision="FP32" names="/blocks.4/attn/Transpose_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="663" name="/blocks.4/attn/Constant" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353109" size="4" />
+			<data element_type="f32" shape="" offset="6889109" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.4/attn/Constant_output_0" />
 			</output>
 		</layer>
 		<layer id="664" name="/blocks.4/attn/Constant_1" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353113" size="4" />
+			<data element_type="f32" shape="" offset="6889113" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.4/attn/Constant_1_output_0" />
 			</output>
@@ -9675,7 +9625,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -9684,13 +9634,13 @@
 				<port id="2" precision="FP32" names="/blocks.4/attn/Mul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="667" name="onnx::MatMul_1595" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="20098781" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="21634788" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -9700,7 +9650,7 @@
 			</output>
 		</layer>
 		<layer id="668" name="Convert_1521" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -9709,7 +9659,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -9717,7 +9667,7 @@
 			</output>
 		</layer>
 		<layer id="669" name="onnx::MatMul_1595/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -9725,32 +9675,32 @@
 			</output>
 		</layer>
 		<layer id="670" name="Convert_1523" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="671" name="Subtract_1524" type="Subtract" version="opset1">
+		<layer id="671" name="onnx::MatMul_1595/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -9758,7 +9708,7 @@
 			</output>
 		</layer>
 		<layer id="672" name="onnx::MatMul_1595/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="20393693" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="21929700" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -9767,57 +9717,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="673" name="onnx::MatMul_1595/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="674" name="onnx::MatMul_1595/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="673" name="onnx::MatMul_1595/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="675" name="Constant_1529" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="674" name="Constant_1528" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="676" name="Reshape_1530" type="Reshape" version="opset1">
+		<layer id="675" name="Reshape_1529" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -9827,7 +9760,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="676" name="onnx::MatMul_1595/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -9838,7 +9786,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -9849,13 +9797,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/key/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="678" name="/blocks.4/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.4/attn/Constant_3_output_0">
 					<dim>4</dim>
@@ -9867,7 +9815,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -9877,14 +9825,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/Reshape_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="680" name="Constant_589" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5657245" size="32" />
+			<data element_type="i64" shape="4" offset="7193246" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -9895,7 +9843,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -9908,7 +9856,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -9919,7 +9867,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -9928,7 +9876,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -9938,22 +9886,22 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/MatMul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -9963,16 +9911,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.4/attn/Cast_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -9982,16 +9930,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.4/attn/Softmax_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -10001,21 +9949,21 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.4/attn/Cast_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="687" name="onnx::Add_1597" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="20402909" size="3072" />
+			<data element_type="f32" shape="768" offset="21938916" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1597">
 					<dim>768</dim>
@@ -10023,7 +9971,7 @@
 			</output>
 		</layer>
 		<layer id="688" name="onnx::MatMul_1598" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="20405981" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="21941988" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -10033,7 +9981,7 @@
 			</output>
 		</layer>
 		<layer id="689" name="Convert_1510" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -10042,7 +9990,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -10050,7 +9998,7 @@
 			</output>
 		</layer>
 		<layer id="690" name="onnx::MatMul_1598/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -10058,32 +10006,32 @@
 			</output>
 		</layer>
 		<layer id="691" name="Convert_1512" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="692" name="Subtract_1513" type="Subtract" version="opset1">
+		<layer id="692" name="onnx::MatMul_1598/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -10091,7 +10039,7 @@
 			</output>
 		</layer>
 		<layer id="693" name="onnx::MatMul_1598/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="20700893" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="22236900" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -10100,57 +10048,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="694" name="onnx::MatMul_1598/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="695" name="onnx::MatMul_1598/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="694" name="onnx::MatMul_1598/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="696" name="Constant_1518" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="695" name="Constant_1517" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="697" name="Reshape_1519" type="Reshape" version="opset1">
+		<layer id="696" name="Reshape_1518" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -10160,7 +10091,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="697" name="onnx::MatMul_1598/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -10171,7 +10117,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -10182,7 +10128,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/value/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -10195,20 +10141,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/value/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="700" name="/blocks.4/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.4/attn/Constant_4_output_0">
 					<dim>4</dim>
@@ -10220,7 +10166,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -10230,14 +10176,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/Reshape_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="702" name="Constant_594" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -10248,7 +10194,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -10260,7 +10206,7 @@
 				<port id="2" precision="FP32" names="/blocks.4/attn/Transpose_2_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
@@ -10271,13 +10217,13 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</input>
@@ -10285,13 +10231,13 @@
 				<port id="2" precision="FP32" names="/blocks.4/attn/MatMul_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="705" name="Constant_601" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -10303,7 +10249,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -10313,7 +10259,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/Transpose_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -10324,7 +10270,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -10336,7 +10282,7 @@
 			</output>
 		</layer>
 		<layer id="708" name="/blocks.4/attn/Constant_6" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.4/attn/Constant_6_output_0">
 					<dim>1</dim>
@@ -10389,7 +10335,7 @@
 			</output>
 		</layer>
 		<layer id="713" name="/blocks.4/attn/Constant_5" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.4/attn/Constant_5_output_0">
 					<dim>1</dim>
@@ -10421,7 +10367,7 @@
 			</output>
 		</layer>
 		<layer id="715" name="/blocks.4/attn/Constant_8" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042740" size="8" />
+			<data element_type="i64" shape="1" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.4/attn/Constant_8_output_0">
 					<dim>1</dim>
@@ -10449,7 +10395,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -10460,13 +10406,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/Reshape_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="718" name="onnx::MatMul_1616" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="20710109" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="22246116" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -10476,7 +10422,7 @@
 			</output>
 		</layer>
 		<layer id="719" name="Convert_1543" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -10485,7 +10431,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -10493,7 +10439,7 @@
 			</output>
 		</layer>
 		<layer id="720" name="onnx::MatMul_1616/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -10501,32 +10447,32 @@
 			</output>
 		</layer>
 		<layer id="721" name="Convert_1545" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="722" name="Subtract_1546" type="Subtract" version="opset1">
+		<layer id="722" name="onnx::MatMul_1616/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -10534,7 +10480,7 @@
 			</output>
 		</layer>
 		<layer id="723" name="onnx::MatMul_1616/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="21005021" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="22541028" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -10543,57 +10489,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="724" name="onnx::MatMul_1616/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="725" name="onnx::MatMul_1616/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="724" name="onnx::MatMul_1616/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="726" name="Constant_1551" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="725" name="Constant_1550" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="727" name="Reshape_1552" type="Reshape" version="opset1">
+		<layer id="726" name="Reshape_1551" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -10603,7 +10532,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="727" name="onnx::MatMul_1616/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -10614,7 +10558,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -10625,7 +10569,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/out/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -10638,14 +10582,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/attn/out/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -10655,19 +10599,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -10677,20 +10621,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.4/mlp_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="732" name="onnx::Add_1621" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="21014237" size="3072" />
+			<data element_type="f32" shape="768" offset="22550244" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1621">
 					<dim>768</dim>
@@ -10698,7 +10642,7 @@
 			</output>
 		</layer>
 		<layer id="733" name="onnx::Add_1618" type="Const" version="opset1">
-			<data element_type="f32" shape="3072" offset="21017309" size="12288" />
+			<data element_type="f32" shape="3072" offset="22553316" size="12288" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1618">
 					<dim>3072</dim>
@@ -10706,13 +10650,13 @@
 			</output>
 		</layer>
 		<layer id="734" name="Constant_627" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="735" name="Constant_626" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -10741,7 +10685,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -10751,13 +10695,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="739" name="blocks.4.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="21029597" size="3072" />
+			<data element_type="f32" shape="768" offset="22565604" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.4.mlp_ln.weight">
 					<dim>768</dim>
@@ -10769,7 +10713,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -10779,13 +10723,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="741" name="blocks.4.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="21032669" size="3072" />
+			<data element_type="f32" shape="768" offset="22568676" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.4.mlp_ln.bias">
 					<dim>768</dim>
@@ -10797,7 +10741,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -10807,7 +10751,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/mlp_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -10817,20 +10761,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.4/mlp_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="744" name="onnx::MatMul_1619" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 3072" offset="21035741" size="1179648" />
+			<data element_type="u4" shape="6, 128, 3072" offset="22571748" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -10840,7 +10784,7 @@
 			</output>
 		</layer>
 		<layer id="745" name="Convert_1554" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -10849,7 +10793,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -10857,7 +10801,7 @@
 			</output>
 		</layer>
 		<layer id="746" name="onnx::MatMul_1619/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -10865,32 +10809,32 @@
 			</output>
 		</layer>
 		<layer id="747" name="Convert_1556" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="748" name="Subtract_1557" type="Subtract" version="opset1">
+		<layer id="748" name="onnx::MatMul_1619/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -10898,7 +10842,7 @@
 			</output>
 		</layer>
 		<layer id="749" name="onnx::MatMul_1619/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 3072" offset="22215389" size="36864" />
+			<data element_type="f16" shape="6, 1, 3072" offset="23751396" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -10907,57 +10851,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="750" name="onnx::MatMul_1619/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="751" name="onnx::MatMul_1619/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="750" name="onnx::MatMul_1619/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="752" name="Constant_1562" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="7506621" size="16" />
+		<layer id="751" name="Constant_1561" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="9042623" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="753" name="Reshape_1563" type="Reshape" version="opset1">
+		<layer id="752" name="Reshape_1562" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -10967,7 +10894,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="753" name="onnx::MatMul_1619/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
@@ -10978,7 +10920,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -10989,7 +10931,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/mlp/mlp.0/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -11002,14 +10944,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/mlp/mlp.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -11025,7 +10967,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -11033,7 +10975,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/mlp/mlp.1/Div_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -11042,14 +10984,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.4/mlp/mlp.1/Erf_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -11065,7 +11007,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -11073,7 +11015,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/mlp/mlp.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -11083,19 +11025,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/mlp/mlp.1/Mul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -11111,7 +11053,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -11119,13 +11061,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/mlp/mlp.1/Mul_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="764" name="onnx::MatMul_1622" type="Const" version="opset1">
-			<data element_type="u4" shape="24, 128, 768" offset="22252253" size="1179648" />
+			<data element_type="u4" shape="24, 128, 768" offset="23788260" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -11135,7 +11077,7 @@
 			</output>
 		</layer>
 		<layer id="765" name="Convert_1565" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -11144,7 +11086,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -11152,7 +11094,7 @@
 			</output>
 		</layer>
 		<layer id="766" name="onnx::MatMul_1622/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -11160,32 +11102,32 @@
 			</output>
 		</layer>
 		<layer id="767" name="Convert_1567" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="768" name="Subtract_1568" type="Subtract" version="opset1">
+		<layer id="768" name="onnx::MatMul_1622/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -11193,7 +11135,7 @@
 			</output>
 		</layer>
 		<layer id="769" name="onnx::MatMul_1622/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="24, 1, 768" offset="23431901" size="36864" />
+			<data element_type="f16" shape="24, 1, 768" offset="24967908" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>24</dim>
@@ -11202,57 +11144,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="770" name="onnx::MatMul_1622/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="771" name="onnx::MatMul_1622/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="770" name="onnx::MatMul_1622/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="772" name="Constant_1573" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="8723149" size="16" />
+		<layer id="771" name="Constant_1572" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="10259151" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="773" name="Reshape_1574" type="Reshape" version="opset1">
+		<layer id="772" name="Reshape_1573" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -11262,7 +11187,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="773" name="onnx::MatMul_1622/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
@@ -11273,7 +11213,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -11284,7 +11224,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/mlp/mlp.2/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -11297,14 +11237,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/mlp/mlp.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -11314,19 +11254,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.4/Add_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -11336,20 +11276,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.5/attn_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="778" name="onnx::Add_1647" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="23468765" size="3072" />
+			<data element_type="f32" shape="768" offset="25004772" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1647">
 					<dim>768</dim>
@@ -11357,7 +11297,7 @@
 			</output>
 		</layer>
 		<layer id="779" name="onnx::Add_1624" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="23471837" size="3072" />
+			<data element_type="f32" shape="768" offset="25007844" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1624">
 					<dim>768</dim>
@@ -11365,13 +11305,13 @@
 			</output>
 		</layer>
 		<layer id="780" name="Constant_649" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="781" name="Constant_648" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -11400,7 +11340,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -11410,13 +11350,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="785" name="blocks.5.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="23474909" size="3072" />
+			<data element_type="f32" shape="768" offset="25010916" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.5.attn_ln.weight">
 					<dim>768</dim>
@@ -11428,7 +11368,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -11438,13 +11378,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="787" name="blocks.5.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="23477981" size="3072" />
+			<data element_type="f32" shape="768" offset="25013988" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.5.attn_ln.bias">
 					<dim>768</dim>
@@ -11456,7 +11396,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -11466,7 +11406,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -11476,20 +11416,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.5/attn_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="790" name="onnx::MatMul_1625" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="23481053" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="25017060" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -11499,7 +11439,7 @@
 			</output>
 		</layer>
 		<layer id="791" name="Convert_1598" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -11508,7 +11448,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -11516,7 +11456,7 @@
 			</output>
 		</layer>
 		<layer id="792" name="onnx::MatMul_1625/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -11524,32 +11464,32 @@
 			</output>
 		</layer>
 		<layer id="793" name="Convert_1600" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="794" name="Subtract_1601" type="Subtract" version="opset1">
+		<layer id="794" name="onnx::MatMul_1625/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -11557,7 +11497,7 @@
 			</output>
 		</layer>
 		<layer id="795" name="onnx::MatMul_1625/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="23775965" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="25311972" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -11566,57 +11506,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="796" name="onnx::MatMul_1625/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="797" name="onnx::MatMul_1625/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="796" name="onnx::MatMul_1625/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="798" name="Constant_1606" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="797" name="Constant_1605" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="799" name="Reshape_1607" type="Reshape" version="opset1">
+		<layer id="798" name="Reshape_1606" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -11626,7 +11549,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="799" name="onnx::MatMul_1625/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -11637,7 +11575,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -11648,7 +11586,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/query/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -11661,20 +11599,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/query/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="802" name="/blocks.5/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.5/attn/Constant_2_output_0">
 					<dim>4</dim>
@@ -11686,7 +11624,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -11696,14 +11634,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/Reshape_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="804" name="Constant_665" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -11714,7 +11652,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -11726,19 +11664,19 @@
 				<port id="2" precision="FP32" names="/blocks.5/attn/Transpose_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="806" name="/blocks.5/attn/Constant" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353109" size="4" />
+			<data element_type="f32" shape="" offset="6889109" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.5/attn/Constant_output_0" />
 			</output>
 		</layer>
 		<layer id="807" name="/blocks.5/attn/Constant_1" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353113" size="4" />
+			<data element_type="f32" shape="" offset="6889113" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.5/attn/Constant_1_output_0" />
 			</output>
@@ -11759,7 +11697,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -11768,13 +11706,13 @@
 				<port id="2" precision="FP32" names="/blocks.5/attn/Mul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="810" name="onnx::MatMul_1627" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="23785181" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="25321188" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -11784,7 +11722,7 @@
 			</output>
 		</layer>
 		<layer id="811" name="Convert_1587" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -11793,7 +11731,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -11801,7 +11739,7 @@
 			</output>
 		</layer>
 		<layer id="812" name="onnx::MatMul_1627/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -11809,32 +11747,32 @@
 			</output>
 		</layer>
 		<layer id="813" name="Convert_1589" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="814" name="Subtract_1590" type="Subtract" version="opset1">
+		<layer id="814" name="onnx::MatMul_1627/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -11842,7 +11780,7 @@
 			</output>
 		</layer>
 		<layer id="815" name="onnx::MatMul_1627/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="24080093" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="25616100" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -11851,57 +11789,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="816" name="onnx::MatMul_1627/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="817" name="onnx::MatMul_1627/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="816" name="onnx::MatMul_1627/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="818" name="Constant_1595" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="817" name="Constant_1594" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="819" name="Reshape_1596" type="Reshape" version="opset1">
+		<layer id="818" name="Reshape_1595" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -11911,7 +11832,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="819" name="onnx::MatMul_1627/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -11922,7 +11858,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -11933,13 +11869,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/key/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="821" name="/blocks.5/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.5/attn/Constant_3_output_0">
 					<dim>4</dim>
@@ -11951,7 +11887,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -11961,14 +11897,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/Reshape_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="823" name="Constant_670" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5657245" size="32" />
+			<data element_type="i64" shape="4" offset="7193246" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -11979,7 +11915,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -11992,7 +11928,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -12003,7 +11939,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -12012,7 +11948,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -12022,22 +11958,22 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/MatMul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -12047,16 +11983,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.5/attn/Cast_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -12066,16 +12002,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.5/attn/Softmax_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -12085,21 +12021,21 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.5/attn/Cast_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="830" name="onnx::Add_1629" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="24089309" size="3072" />
+			<data element_type="f32" shape="768" offset="25625316" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1629">
 					<dim>768</dim>
@@ -12107,7 +12043,7 @@
 			</output>
 		</layer>
 		<layer id="831" name="onnx::MatMul_1630" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="24092381" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="25628388" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -12117,7 +12053,7 @@
 			</output>
 		</layer>
 		<layer id="832" name="Convert_1576" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -12126,7 +12062,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -12134,7 +12070,7 @@
 			</output>
 		</layer>
 		<layer id="833" name="onnx::MatMul_1630/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -12142,32 +12078,32 @@
 			</output>
 		</layer>
 		<layer id="834" name="Convert_1578" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="835" name="Subtract_1579" type="Subtract" version="opset1">
+		<layer id="835" name="onnx::MatMul_1630/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -12175,7 +12111,7 @@
 			</output>
 		</layer>
 		<layer id="836" name="onnx::MatMul_1630/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="24387293" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="25923300" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -12184,57 +12120,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="837" name="onnx::MatMul_1630/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="838" name="onnx::MatMul_1630/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="837" name="onnx::MatMul_1630/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="839" name="Constant_1584" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="838" name="Constant_1583" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="840" name="Reshape_1585" type="Reshape" version="opset1">
+		<layer id="839" name="Reshape_1584" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -12244,7 +12163,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="840" name="onnx::MatMul_1630/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -12255,7 +12189,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -12266,7 +12200,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/value/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -12279,20 +12213,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/value/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="843" name="/blocks.5/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.5/attn/Constant_4_output_0">
 					<dim>4</dim>
@@ -12304,7 +12238,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -12314,14 +12248,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/Reshape_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="845" name="Constant_675" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -12332,7 +12266,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -12344,7 +12278,7 @@
 				<port id="2" precision="FP32" names="/blocks.5/attn/Transpose_2_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
@@ -12355,13 +12289,13 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</input>
@@ -12369,13 +12303,13 @@
 				<port id="2" precision="FP32" names="/blocks.5/attn/MatMul_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="848" name="Constant_682" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -12387,7 +12321,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -12397,7 +12331,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/Transpose_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -12408,7 +12342,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -12420,7 +12354,7 @@
 			</output>
 		</layer>
 		<layer id="851" name="/blocks.5/attn/Constant_6" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.5/attn/Constant_6_output_0">
 					<dim>1</dim>
@@ -12473,7 +12407,7 @@
 			</output>
 		</layer>
 		<layer id="856" name="/blocks.5/attn/Constant_5" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.5/attn/Constant_5_output_0">
 					<dim>1</dim>
@@ -12505,7 +12439,7 @@
 			</output>
 		</layer>
 		<layer id="858" name="/blocks.5/attn/Constant_8" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042740" size="8" />
+			<data element_type="i64" shape="1" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.5/attn/Constant_8_output_0">
 					<dim>1</dim>
@@ -12533,7 +12467,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -12544,13 +12478,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/Reshape_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="861" name="onnx::MatMul_1648" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="24396509" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="25932516" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -12560,7 +12494,7 @@
 			</output>
 		</layer>
 		<layer id="862" name="Convert_1609" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -12569,7 +12503,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -12577,7 +12511,7 @@
 			</output>
 		</layer>
 		<layer id="863" name="onnx::MatMul_1648/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -12585,99 +12519,82 @@
 			</output>
 		</layer>
 		<layer id="864" name="Convert_1611" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="865" name="Subtract_1612" type="Subtract" version="opset1">
+		<layer id="865" name="onnx::MatMul_1648/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>6</dim>
-					<dim>128</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-				</port>
-			</input>
-			<output>
-				<port id="2" precision="FP32">
-					<dim>6</dim>
-					<dim>128</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="866" name="onnx::MatMul_1648/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="24691421" size="9216" />
-			<output>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="867" name="onnx::MatMul_1648/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
 			<input>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
-					<dim>1</dim>
+					<dim>128</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>6</dim>
+					<dim>128</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="866" name="onnx::MatMul_1648/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="6, 1, 768" offset="26227428" size="9216" />
+			<output>
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="868" name="onnx::MatMul_1648/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="867" name="onnx::MatMul_1648/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="869" name="Constant_1617" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="868" name="Constant_1616" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="870" name="Reshape_1618" type="Reshape" version="opset1">
+		<layer id="869" name="Reshape_1617" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -12687,7 +12604,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="870" name="onnx::MatMul_1648/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -12698,7 +12630,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -12709,7 +12641,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/out/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -12722,14 +12654,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/attn/out/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -12739,19 +12671,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -12761,20 +12693,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.5/mlp_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="875" name="onnx::Add_1653" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="24700637" size="3072" />
+			<data element_type="f32" shape="768" offset="26236644" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1653">
 					<dim>768</dim>
@@ -12782,7 +12714,7 @@
 			</output>
 		</layer>
 		<layer id="876" name="onnx::Add_1650" type="Const" version="opset1">
-			<data element_type="f32" shape="3072" offset="24703709" size="12288" />
+			<data element_type="f32" shape="3072" offset="26239716" size="12288" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1650">
 					<dim>3072</dim>
@@ -12790,13 +12722,13 @@
 			</output>
 		</layer>
 		<layer id="877" name="Constant_708" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="878" name="Constant_707" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -12825,7 +12757,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -12835,13 +12767,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="882" name="blocks.5.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="24715997" size="3072" />
+			<data element_type="f32" shape="768" offset="26252004" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.5.mlp_ln.weight">
 					<dim>768</dim>
@@ -12853,7 +12785,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -12863,13 +12795,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="884" name="blocks.5.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="24719069" size="3072" />
+			<data element_type="f32" shape="768" offset="26255076" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.5.mlp_ln.bias">
 					<dim>768</dim>
@@ -12881,7 +12813,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -12891,7 +12823,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/mlp_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -12901,20 +12833,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.5/mlp_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="887" name="onnx::MatMul_1651" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 3072" offset="24722141" size="1179648" />
+			<data element_type="u4" shape="6, 128, 3072" offset="26258148" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -12924,7 +12856,7 @@
 			</output>
 		</layer>
 		<layer id="888" name="Convert_1620" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -12933,7 +12865,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -12941,7 +12873,7 @@
 			</output>
 		</layer>
 		<layer id="889" name="onnx::MatMul_1651/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -12949,32 +12881,32 @@
 			</output>
 		</layer>
 		<layer id="890" name="Convert_1622" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="891" name="Subtract_1623" type="Subtract" version="opset1">
+		<layer id="891" name="onnx::MatMul_1651/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -12982,7 +12914,7 @@
 			</output>
 		</layer>
 		<layer id="892" name="onnx::MatMul_1651/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 3072" offset="25901789" size="36864" />
+			<data element_type="f16" shape="6, 1, 3072" offset="27437796" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -12991,57 +12923,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="893" name="onnx::MatMul_1651/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="894" name="onnx::MatMul_1651/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="893" name="onnx::MatMul_1651/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="895" name="Constant_1628" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="7506621" size="16" />
+		<layer id="894" name="Constant_1627" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="9042623" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="896" name="Reshape_1629" type="Reshape" version="opset1">
+		<layer id="895" name="Reshape_1628" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -13051,7 +12966,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="896" name="onnx::MatMul_1651/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
@@ -13062,7 +12992,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -13073,7 +13003,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/mlp/mlp.0/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -13086,14 +13016,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/mlp/mlp.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -13109,7 +13039,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -13117,7 +13047,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/mlp/mlp.1/Div_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -13126,14 +13056,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.5/mlp/mlp.1/Erf_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -13149,7 +13079,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -13157,7 +13087,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/mlp/mlp.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -13167,19 +13097,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/mlp/mlp.1/Mul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -13195,7 +13125,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -13203,13 +13133,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/mlp/mlp.1/Mul_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="907" name="onnx::MatMul_1654" type="Const" version="opset1">
-			<data element_type="u4" shape="24, 128, 768" offset="25938653" size="1179648" />
+			<data element_type="u4" shape="24, 128, 768" offset="27474660" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -13219,7 +13149,7 @@
 			</output>
 		</layer>
 		<layer id="908" name="Convert_1631" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -13228,7 +13158,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -13236,7 +13166,7 @@
 			</output>
 		</layer>
 		<layer id="909" name="onnx::MatMul_1654/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -13244,32 +13174,32 @@
 			</output>
 		</layer>
 		<layer id="910" name="Convert_1633" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="911" name="Subtract_1634" type="Subtract" version="opset1">
+		<layer id="911" name="onnx::MatMul_1654/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -13277,7 +13207,7 @@
 			</output>
 		</layer>
 		<layer id="912" name="onnx::MatMul_1654/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="24, 1, 768" offset="27118301" size="36864" />
+			<data element_type="f16" shape="24, 1, 768" offset="28654308" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>24</dim>
@@ -13286,57 +13216,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="913" name="onnx::MatMul_1654/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="914" name="onnx::MatMul_1654/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="913" name="onnx::MatMul_1654/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="915" name="Constant_1639" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="8723149" size="16" />
+		<layer id="914" name="Constant_1638" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="10259151" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="916" name="Reshape_1640" type="Reshape" version="opset1">
+		<layer id="915" name="Reshape_1639" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -13346,7 +13259,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="916" name="onnx::MatMul_1654/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
@@ -13357,7 +13285,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -13368,7 +13296,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/mlp/mlp.2/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -13381,14 +13309,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/mlp/mlp.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -13398,19 +13326,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.5/Add_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -13420,20 +13348,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.6/attn_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="921" name="onnx::Add_1679" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="27155165" size="3072" />
+			<data element_type="f32" shape="768" offset="28691172" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1679">
 					<dim>768</dim>
@@ -13441,7 +13369,7 @@
 			</output>
 		</layer>
 		<layer id="922" name="onnx::Add_1656" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="27158237" size="3072" />
+			<data element_type="f32" shape="768" offset="28694244" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1656">
 					<dim>768</dim>
@@ -13449,13 +13377,13 @@
 			</output>
 		</layer>
 		<layer id="923" name="Constant_730" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="924" name="Constant_729" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -13484,7 +13412,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -13494,13 +13422,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="928" name="blocks.6.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="27161309" size="3072" />
+			<data element_type="f32" shape="768" offset="28697316" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.6.attn_ln.weight">
 					<dim>768</dim>
@@ -13512,7 +13440,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -13522,13 +13450,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="930" name="blocks.6.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="27164381" size="3072" />
+			<data element_type="f32" shape="768" offset="28700388" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.6.attn_ln.bias">
 					<dim>768</dim>
@@ -13540,7 +13468,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -13550,7 +13478,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -13560,20 +13488,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.6/attn_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="933" name="onnx::MatMul_1657" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="27167453" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="28703460" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -13583,7 +13511,7 @@
 			</output>
 		</layer>
 		<layer id="934" name="Convert_1664" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -13592,7 +13520,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -13600,7 +13528,7 @@
 			</output>
 		</layer>
 		<layer id="935" name="onnx::MatMul_1657/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -13608,32 +13536,32 @@
 			</output>
 		</layer>
 		<layer id="936" name="Convert_1666" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="937" name="Subtract_1667" type="Subtract" version="opset1">
+		<layer id="937" name="onnx::MatMul_1657/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -13641,7 +13569,7 @@
 			</output>
 		</layer>
 		<layer id="938" name="onnx::MatMul_1657/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="27462365" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="28998372" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -13650,57 +13578,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="939" name="onnx::MatMul_1657/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="940" name="onnx::MatMul_1657/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="939" name="onnx::MatMul_1657/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="941" name="Constant_1672" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="940" name="Constant_1671" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="942" name="Reshape_1673" type="Reshape" version="opset1">
+		<layer id="941" name="Reshape_1672" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -13710,7 +13621,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="942" name="onnx::MatMul_1657/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -13721,7 +13647,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -13732,7 +13658,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/query/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -13745,20 +13671,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/query/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="945" name="/blocks.6/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.6/attn/Constant_2_output_0">
 					<dim>4</dim>
@@ -13770,7 +13696,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -13780,14 +13706,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/Reshape_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="947" name="Constant_746" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -13798,7 +13724,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -13810,19 +13736,19 @@
 				<port id="2" precision="FP32" names="/blocks.6/attn/Transpose_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="949" name="/blocks.6/attn/Constant" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353109" size="4" />
+			<data element_type="f32" shape="" offset="6889109" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.6/attn/Constant_output_0" />
 			</output>
 		</layer>
 		<layer id="950" name="/blocks.6/attn/Constant_1" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353113" size="4" />
+			<data element_type="f32" shape="" offset="6889113" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.6/attn/Constant_1_output_0" />
 			</output>
@@ -13843,7 +13769,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -13852,13 +13778,13 @@
 				<port id="2" precision="FP32" names="/blocks.6/attn/Mul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="953" name="onnx::MatMul_1659" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="27471581" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="29007588" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -13868,7 +13794,7 @@
 			</output>
 		</layer>
 		<layer id="954" name="Convert_1653" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -13877,7 +13803,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -13885,7 +13811,7 @@
 			</output>
 		</layer>
 		<layer id="955" name="onnx::MatMul_1659/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="12692193" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -13893,32 +13819,32 @@
 			</output>
 		</layer>
 		<layer id="956" name="Convert_1655" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="957" name="Subtract_1656" type="Subtract" version="opset1">
+		<layer id="957" name="onnx::MatMul_1659/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -13926,7 +13852,7 @@
 			</output>
 		</layer>
 		<layer id="958" name="onnx::MatMul_1659/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="27766493" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="29302500" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -13935,57 +13861,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="959" name="onnx::MatMul_1659/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="960" name="onnx::MatMul_1659/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="959" name="onnx::MatMul_1659/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="961" name="Constant_1661" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="960" name="Constant_1660" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="962" name="Reshape_1662" type="Reshape" version="opset1">
+		<layer id="961" name="Reshape_1661" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -13995,7 +13904,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="962" name="onnx::MatMul_1659/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -14006,7 +13930,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -14017,13 +13941,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/key/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="964" name="/blocks.6/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.6/attn/Constant_3_output_0">
 					<dim>4</dim>
@@ -14035,7 +13959,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -14045,14 +13969,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/Reshape_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="966" name="Constant_751" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5657245" size="32" />
+			<data element_type="i64" shape="4" offset="7193246" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -14063,7 +13987,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -14076,7 +14000,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -14087,7 +14011,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -14096,7 +14020,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -14106,22 +14030,22 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/MatMul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -14131,16 +14055,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.6/attn/Cast_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -14150,16 +14074,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.6/attn/Softmax_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -14169,21 +14093,21 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.6/attn/Cast_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="973" name="onnx::Add_1661" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="27775709" size="3072" />
+			<data element_type="f32" shape="768" offset="29311716" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1661">
 					<dim>768</dim>
@@ -14191,7 +14115,7 @@
 			</output>
 		</layer>
 		<layer id="974" name="onnx::MatMul_1662" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="27778781" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="29314788" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -14201,7 +14125,7 @@
 			</output>
 		</layer>
 		<layer id="975" name="Convert_1642" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -14210,7 +14134,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -14218,7 +14142,7 @@
 			</output>
 		</layer>
 		<layer id="976" name="onnx::MatMul_1662/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -14226,32 +14150,32 @@
 			</output>
 		</layer>
 		<layer id="977" name="Convert_1644" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="978" name="Subtract_1645" type="Subtract" version="opset1">
+		<layer id="978" name="onnx::MatMul_1662/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -14259,7 +14183,7 @@
 			</output>
 		</layer>
 		<layer id="979" name="onnx::MatMul_1662/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="28073693" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="29609700" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -14268,57 +14192,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="980" name="onnx::MatMul_1662/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="981" name="onnx::MatMul_1662/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="980" name="onnx::MatMul_1662/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="982" name="Constant_1650" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="981" name="Constant_1649" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="983" name="Reshape_1651" type="Reshape" version="opset1">
+		<layer id="982" name="Reshape_1650" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -14328,7 +14235,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="983" name="onnx::MatMul_1662/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -14339,7 +14261,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -14350,7 +14272,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/value/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -14363,20 +14285,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/value/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="986" name="/blocks.6/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.6/attn/Constant_4_output_0">
 					<dim>4</dim>
@@ -14388,7 +14310,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -14398,14 +14320,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/Reshape_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="988" name="Constant_756" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -14416,7 +14338,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -14428,7 +14350,7 @@
 				<port id="2" precision="FP32" names="/blocks.6/attn/Transpose_2_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
@@ -14439,13 +14361,13 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</input>
@@ -14453,13 +14375,13 @@
 				<port id="2" precision="FP32" names="/blocks.6/attn/MatMul_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="991" name="Constant_763" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -14471,7 +14393,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -14481,7 +14403,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/Transpose_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -14492,7 +14414,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -14504,7 +14426,7 @@
 			</output>
 		</layer>
 		<layer id="994" name="/blocks.6/attn/Constant_6" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.6/attn/Constant_6_output_0">
 					<dim>1</dim>
@@ -14557,7 +14479,7 @@
 			</output>
 		</layer>
 		<layer id="999" name="/blocks.6/attn/Constant_5" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.6/attn/Constant_5_output_0">
 					<dim>1</dim>
@@ -14589,7 +14511,7 @@
 			</output>
 		</layer>
 		<layer id="1001" name="/blocks.6/attn/Constant_8" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042740" size="8" />
+			<data element_type="i64" shape="1" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.6/attn/Constant_8_output_0">
 					<dim>1</dim>
@@ -14617,7 +14539,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -14628,13 +14550,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/Reshape_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1004" name="onnx::MatMul_1680" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="28082909" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="29618916" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -14644,7 +14566,7 @@
 			</output>
 		</layer>
 		<layer id="1005" name="Convert_1675" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -14653,7 +14575,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -14661,7 +14583,7 @@
 			</output>
 		</layer>
 		<layer id="1006" name="onnx::MatMul_1680/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -14669,32 +14591,32 @@
 			</output>
 		</layer>
 		<layer id="1007" name="Convert_1677" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1008" name="Subtract_1678" type="Subtract" version="opset1">
+		<layer id="1008" name="onnx::MatMul_1680/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -14702,7 +14624,7 @@
 			</output>
 		</layer>
 		<layer id="1009" name="onnx::MatMul_1680/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="28377821" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="29913828" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -14711,57 +14633,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1010" name="onnx::MatMul_1680/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1011" name="onnx::MatMul_1680/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1010" name="onnx::MatMul_1680/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1012" name="Constant_1683" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1011" name="Constant_1682" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1013" name="Reshape_1684" type="Reshape" version="opset1">
+		<layer id="1012" name="Reshape_1683" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -14771,7 +14676,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1013" name="onnx::MatMul_1680/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -14782,7 +14702,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -14793,7 +14713,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/out/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -14806,14 +14726,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/attn/out/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -14823,19 +14743,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -14845,20 +14765,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.6/mlp_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1018" name="onnx::Add_1685" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="28387037" size="3072" />
+			<data element_type="f32" shape="768" offset="29923044" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1685">
 					<dim>768</dim>
@@ -14866,7 +14786,7 @@
 			</output>
 		</layer>
 		<layer id="1019" name="onnx::Add_1682" type="Const" version="opset1">
-			<data element_type="f32" shape="3072" offset="28390109" size="12288" />
+			<data element_type="f32" shape="3072" offset="29926116" size="12288" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1682">
 					<dim>3072</dim>
@@ -14874,13 +14794,13 @@
 			</output>
 		</layer>
 		<layer id="1020" name="Constant_789" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="1021" name="Constant_788" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -14909,7 +14829,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -14919,13 +14839,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1025" name="blocks.6.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="28402397" size="3072" />
+			<data element_type="f32" shape="768" offset="29938404" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.6.mlp_ln.weight">
 					<dim>768</dim>
@@ -14937,7 +14857,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -14947,13 +14867,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1027" name="blocks.6.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="28405469" size="3072" />
+			<data element_type="f32" shape="768" offset="29941476" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.6.mlp_ln.bias">
 					<dim>768</dim>
@@ -14965,7 +14885,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -14975,7 +14895,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/mlp_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -14985,20 +14905,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.6/mlp_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1030" name="onnx::MatMul_1683" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 3072" offset="28408541" size="1179648" />
+			<data element_type="u4" shape="6, 128, 3072" offset="29944548" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -15008,7 +14928,7 @@
 			</output>
 		</layer>
 		<layer id="1031" name="Convert_1686" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -15017,7 +14937,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -15025,7 +14945,7 @@
 			</output>
 		</layer>
 		<layer id="1032" name="onnx::MatMul_1683/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7491262" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -15033,32 +14953,32 @@
 			</output>
 		</layer>
 		<layer id="1033" name="Convert_1688" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1034" name="Subtract_1689" type="Subtract" version="opset1">
+		<layer id="1034" name="onnx::MatMul_1683/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -15066,7 +14986,7 @@
 			</output>
 		</layer>
 		<layer id="1035" name="onnx::MatMul_1683/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 3072" offset="29588189" size="36864" />
+			<data element_type="f16" shape="6, 1, 3072" offset="31124196" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -15075,57 +14995,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1036" name="onnx::MatMul_1683/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1037" name="onnx::MatMul_1683/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1036" name="onnx::MatMul_1683/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1038" name="Constant_1694" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="7506621" size="16" />
+		<layer id="1037" name="Constant_1693" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="9042623" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1039" name="Reshape_1695" type="Reshape" version="opset1">
+		<layer id="1038" name="Reshape_1694" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -15135,7 +15038,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1039" name="onnx::MatMul_1683/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
@@ -15146,7 +15064,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -15157,7 +15075,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/mlp/mlp.0/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -15170,14 +15088,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/mlp/mlp.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -15193,7 +15111,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -15201,7 +15119,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/mlp/mlp.1/Div_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -15210,14 +15128,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.6/mlp/mlp.1/Erf_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -15233,7 +15151,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -15241,7 +15159,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/mlp/mlp.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -15251,19 +15169,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/mlp/mlp.1/Mul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -15279,7 +15197,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -15287,13 +15205,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/mlp/mlp.1/Mul_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1050" name="onnx::MatMul_1686" type="Const" version="opset1">
-			<data element_type="u4" shape="24, 128, 768" offset="29625053" size="1179648" />
+			<data element_type="u4" shape="24, 128, 768" offset="31161060" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -15303,7 +15221,7 @@
 			</output>
 		</layer>
 		<layer id="1051" name="Convert_1697" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -15312,7 +15230,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -15320,7 +15238,7 @@
 			</output>
 		</layer>
 		<layer id="1052" name="onnx::MatMul_1686/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -15328,32 +15246,32 @@
 			</output>
 		</layer>
 		<layer id="1053" name="Convert_1699" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1054" name="Subtract_1700" type="Subtract" version="opset1">
+		<layer id="1054" name="onnx::MatMul_1686/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -15361,7 +15279,7 @@
 			</output>
 		</layer>
 		<layer id="1055" name="onnx::MatMul_1686/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="24, 1, 768" offset="30804701" size="36864" />
+			<data element_type="f16" shape="24, 1, 768" offset="32340708" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>24</dim>
@@ -15370,57 +15288,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1056" name="onnx::MatMul_1686/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1057" name="onnx::MatMul_1686/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1056" name="onnx::MatMul_1686/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1058" name="Constant_1705" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="8723149" size="16" />
+		<layer id="1057" name="Constant_1704" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="10259151" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1059" name="Reshape_1706" type="Reshape" version="opset1">
+		<layer id="1058" name="Reshape_1705" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -15430,7 +15331,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1059" name="onnx::MatMul_1686/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
@@ -15441,7 +15357,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -15452,7 +15368,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/mlp/mlp.2/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -15465,14 +15381,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/mlp/mlp.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -15482,19 +15398,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.6/Add_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -15504,20 +15420,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.7/attn_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1064" name="onnx::Add_1711" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="30841565" size="3072" />
+			<data element_type="f32" shape="768" offset="32377572" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1711">
 					<dim>768</dim>
@@ -15525,7 +15441,7 @@
 			</output>
 		</layer>
 		<layer id="1065" name="onnx::Add_1688" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="30844637" size="3072" />
+			<data element_type="f32" shape="768" offset="32380644" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1688">
 					<dim>768</dim>
@@ -15533,13 +15449,13 @@
 			</output>
 		</layer>
 		<layer id="1066" name="Constant_811" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="1067" name="Constant_810" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -15568,7 +15484,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -15578,13 +15494,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1071" name="blocks.7.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="30847709" size="3072" />
+			<data element_type="f32" shape="768" offset="32383716" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.7.attn_ln.weight">
 					<dim>768</dim>
@@ -15596,7 +15512,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -15606,13 +15522,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1073" name="blocks.7.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="30850781" size="3072" />
+			<data element_type="f32" shape="768" offset="32386788" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.7.attn_ln.bias">
 					<dim>768</dim>
@@ -15624,7 +15540,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -15634,7 +15550,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -15644,20 +15560,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.7/attn_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1076" name="onnx::MatMul_1689" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="30853853" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="32389860" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -15667,7 +15583,7 @@
 			</output>
 		</layer>
 		<layer id="1077" name="Convert_1730" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -15676,7 +15592,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -15684,7 +15600,7 @@
 			</output>
 		</layer>
 		<layer id="1078" name="onnx::MatMul_1689/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="32684772" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -15692,32 +15608,32 @@
 			</output>
 		</layer>
 		<layer id="1079" name="Convert_1732" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1080" name="Subtract_1733" type="Subtract" version="opset1">
+		<layer id="1080" name="onnx::MatMul_1689/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -15725,7 +15641,7 @@
 			</output>
 		</layer>
 		<layer id="1081" name="onnx::MatMul_1689/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="31148765" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="32684773" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -15734,57 +15650,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1082" name="onnx::MatMul_1689/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1083" name="onnx::MatMul_1689/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1082" name="onnx::MatMul_1689/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1084" name="Constant_1738" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1083" name="Constant_1737" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1085" name="Reshape_1739" type="Reshape" version="opset1">
+		<layer id="1084" name="Reshape_1738" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -15794,7 +15693,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1085" name="onnx::MatMul_1689/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -15805,7 +15719,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -15816,7 +15730,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/query/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -15829,20 +15743,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/query/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1088" name="/blocks.7/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.7/attn/Constant_2_output_0">
 					<dim>4</dim>
@@ -15854,7 +15768,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -15864,14 +15778,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/Reshape_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1090" name="Constant_827" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -15882,7 +15796,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -15894,19 +15808,19 @@
 				<port id="2" precision="FP32" names="/blocks.7/attn/Transpose_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1092" name="/blocks.7/attn/Constant" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353109" size="4" />
+			<data element_type="f32" shape="" offset="6889109" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.7/attn/Constant_output_0" />
 			</output>
 		</layer>
 		<layer id="1093" name="/blocks.7/attn/Constant_1" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353113" size="4" />
+			<data element_type="f32" shape="" offset="6889113" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.7/attn/Constant_1_output_0" />
 			</output>
@@ -15927,7 +15841,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -15936,13 +15850,13 @@
 				<port id="2" precision="FP32" names="/blocks.7/attn/Mul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1096" name="onnx::MatMul_1691" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="31157981" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="32693989" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -15952,7 +15866,7 @@
 			</output>
 		</layer>
 		<layer id="1097" name="Convert_1719" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -15961,7 +15875,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -15969,7 +15883,7 @@
 			</output>
 		</layer>
 		<layer id="1098" name="onnx::MatMul_1691/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -15977,32 +15891,32 @@
 			</output>
 		</layer>
 		<layer id="1099" name="Convert_1721" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1100" name="Subtract_1722" type="Subtract" version="opset1">
+		<layer id="1100" name="onnx::MatMul_1691/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -16010,7 +15924,7 @@
 			</output>
 		</layer>
 		<layer id="1101" name="onnx::MatMul_1691/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="31452893" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="32988901" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -16019,57 +15933,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1102" name="onnx::MatMul_1691/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1103" name="onnx::MatMul_1691/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1102" name="onnx::MatMul_1691/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1104" name="Constant_1727" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1103" name="Constant_1726" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1105" name="Reshape_1728" type="Reshape" version="opset1">
+		<layer id="1104" name="Reshape_1727" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -16079,7 +15976,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1105" name="onnx::MatMul_1691/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -16090,7 +16002,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -16101,13 +16013,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/key/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1107" name="/blocks.7/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.7/attn/Constant_3_output_0">
 					<dim>4</dim>
@@ -16119,7 +16031,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -16129,14 +16041,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/Reshape_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1109" name="Constant_832" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5657245" size="32" />
+			<data element_type="i64" shape="4" offset="7193246" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -16147,7 +16059,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -16160,7 +16072,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -16171,7 +16083,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -16180,7 +16092,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -16190,22 +16102,22 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/MatMul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -16215,16 +16127,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.7/attn/Cast_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -16234,16 +16146,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.7/attn/Softmax_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -16253,21 +16165,21 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.7/attn/Cast_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1116" name="onnx::Add_1693" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="31462109" size="3072" />
+			<data element_type="f32" shape="768" offset="32998117" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1693">
 					<dim>768</dim>
@@ -16275,7 +16187,7 @@
 			</output>
 		</layer>
 		<layer id="1117" name="onnx::MatMul_1694" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="31465181" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="33001189" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -16285,7 +16197,7 @@
 			</output>
 		</layer>
 		<layer id="1118" name="Convert_1708" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -16294,7 +16206,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -16302,7 +16214,7 @@
 			</output>
 		</layer>
 		<layer id="1119" name="onnx::MatMul_1694/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7491262" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -16310,99 +16222,82 @@
 			</output>
 		</layer>
 		<layer id="1120" name="Convert_1710" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1121" name="Subtract_1711" type="Subtract" version="opset1">
+		<layer id="1121" name="onnx::MatMul_1694/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
-			<output>
-				<port id="2" precision="FP32">
-					<dim>6</dim>
-					<dim>128</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1122" name="onnx::MatMul_1694/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="31760093" size="9216" />
-			<output>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1123" name="onnx::MatMul_1694/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
+			<output>
+				<port id="2" precision="FP16">
 					<dim>6</dim>
-					<dim>1</dim>
+					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-			</input>
+			</output>
+		</layer>
+		<layer id="1122" name="onnx::MatMul_1694/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="6, 1, 768" offset="33296101" size="9216" />
 			<output>
-				<port id="1" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1124" name="onnx::MatMul_1694/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1123" name="onnx::MatMul_1694/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1125" name="Constant_1716" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1124" name="Constant_1715" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1126" name="Reshape_1717" type="Reshape" version="opset1">
+		<layer id="1125" name="Reshape_1716" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -16412,7 +16307,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1126" name="onnx::MatMul_1694/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -16423,7 +16333,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -16434,7 +16344,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/value/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -16447,20 +16357,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/value/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1129" name="/blocks.7/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.7/attn/Constant_4_output_0">
 					<dim>4</dim>
@@ -16472,7 +16382,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -16482,14 +16392,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/Reshape_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1131" name="Constant_837" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -16500,7 +16410,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -16512,7 +16422,7 @@
 				<port id="2" precision="FP32" names="/blocks.7/attn/Transpose_2_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
@@ -16523,13 +16433,13 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</input>
@@ -16537,13 +16447,13 @@
 				<port id="2" precision="FP32" names="/blocks.7/attn/MatMul_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1134" name="Constant_844" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -16555,7 +16465,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -16565,7 +16475,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/Transpose_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -16576,7 +16486,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -16588,7 +16498,7 @@
 			</output>
 		</layer>
 		<layer id="1137" name="/blocks.7/attn/Constant_6" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.7/attn/Constant_6_output_0">
 					<dim>1</dim>
@@ -16641,7 +16551,7 @@
 			</output>
 		</layer>
 		<layer id="1142" name="/blocks.7/attn/Constant_5" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.7/attn/Constant_5_output_0">
 					<dim>1</dim>
@@ -16673,7 +16583,7 @@
 			</output>
 		</layer>
 		<layer id="1144" name="/blocks.7/attn/Constant_8" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042740" size="8" />
+			<data element_type="i64" shape="1" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.7/attn/Constant_8_output_0">
 					<dim>1</dim>
@@ -16701,7 +16611,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -16712,13 +16622,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/Reshape_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1147" name="onnx::MatMul_1712" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="31769309" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="33305317" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -16728,7 +16638,7 @@
 			</output>
 		</layer>
 		<layer id="1148" name="Convert_1741" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -16737,7 +16647,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -16745,7 +16655,7 @@
 			</output>
 		</layer>
 		<layer id="1149" name="onnx::MatMul_1712/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="14864098" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -16753,32 +16663,32 @@
 			</output>
 		</layer>
 		<layer id="1150" name="Convert_1743" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1151" name="Subtract_1744" type="Subtract" version="opset1">
+		<layer id="1151" name="onnx::MatMul_1712/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -16786,7 +16696,7 @@
 			</output>
 		</layer>
 		<layer id="1152" name="onnx::MatMul_1712/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="32064221" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="33600229" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -16795,57 +16705,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1153" name="onnx::MatMul_1712/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1154" name="onnx::MatMul_1712/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1153" name="onnx::MatMul_1712/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1155" name="Constant_1749" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1154" name="Constant_1748" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1156" name="Reshape_1750" type="Reshape" version="opset1">
+		<layer id="1155" name="Reshape_1749" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -16855,7 +16748,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1156" name="onnx::MatMul_1712/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -16866,7 +16774,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -16877,7 +16785,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/out/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -16890,14 +16798,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/attn/out/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -16907,19 +16815,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -16929,20 +16837,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.7/mlp_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1161" name="onnx::Add_1717" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="32073437" size="3072" />
+			<data element_type="f32" shape="768" offset="33609445" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1717">
 					<dim>768</dim>
@@ -16950,7 +16858,7 @@
 			</output>
 		</layer>
 		<layer id="1162" name="onnx::Add_1714" type="Const" version="opset1">
-			<data element_type="f32" shape="3072" offset="32076509" size="12288" />
+			<data element_type="f32" shape="3072" offset="33612517" size="12288" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1714">
 					<dim>3072</dim>
@@ -16958,13 +16866,13 @@
 			</output>
 		</layer>
 		<layer id="1163" name="Constant_870" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="1164" name="Constant_869" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -16993,7 +16901,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -17003,13 +16911,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1168" name="blocks.7.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="32088797" size="3072" />
+			<data element_type="f32" shape="768" offset="33624805" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.7.mlp_ln.weight">
 					<dim>768</dim>
@@ -17021,7 +16929,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -17031,13 +16939,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1170" name="blocks.7.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="32091869" size="3072" />
+			<data element_type="f32" shape="768" offset="33627877" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.7.mlp_ln.bias">
 					<dim>768</dim>
@@ -17049,7 +16957,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -17059,7 +16967,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/mlp_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -17069,20 +16977,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.7/mlp_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1173" name="onnx::MatMul_1715" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 3072" offset="32094941" size="1179648" />
+			<data element_type="u4" shape="6, 128, 3072" offset="33630949" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -17092,7 +17000,7 @@
 			</output>
 		</layer>
 		<layer id="1174" name="Convert_1752" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -17101,7 +17009,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -17109,7 +17017,7 @@
 			</output>
 		</layer>
 		<layer id="1175" name="onnx::MatMul_1715/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -17117,32 +17025,32 @@
 			</output>
 		</layer>
 		<layer id="1176" name="Convert_1754" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1177" name="Subtract_1755" type="Subtract" version="opset1">
+		<layer id="1177" name="onnx::MatMul_1715/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -17150,7 +17058,7 @@
 			</output>
 		</layer>
 		<layer id="1178" name="onnx::MatMul_1715/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 3072" offset="33274589" size="36864" />
+			<data element_type="f16" shape="6, 1, 3072" offset="34810597" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -17159,57 +17067,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1179" name="onnx::MatMul_1715/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1180" name="onnx::MatMul_1715/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1179" name="onnx::MatMul_1715/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1181" name="Constant_1760" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="7506621" size="16" />
+		<layer id="1180" name="Constant_1759" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="9042623" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1182" name="Reshape_1761" type="Reshape" version="opset1">
+		<layer id="1181" name="Reshape_1760" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -17219,7 +17110,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1182" name="onnx::MatMul_1715/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
@@ -17230,7 +17136,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -17241,7 +17147,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/mlp/mlp.0/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -17254,14 +17160,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/mlp/mlp.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -17277,7 +17183,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -17285,7 +17191,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/mlp/mlp.1/Div_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -17294,14 +17200,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.7/mlp/mlp.1/Erf_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -17317,7 +17223,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -17325,7 +17231,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/mlp/mlp.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -17335,19 +17241,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/mlp/mlp.1/Mul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -17363,7 +17269,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -17371,13 +17277,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/mlp/mlp.1/Mul_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1193" name="onnx::MatMul_1718" type="Const" version="opset1">
-			<data element_type="u4" shape="24, 128, 768" offset="33311453" size="1179648" />
+			<data element_type="u4" shape="24, 128, 768" offset="34847461" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -17387,7 +17293,7 @@
 			</output>
 		</layer>
 		<layer id="1194" name="Convert_1763" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -17396,7 +17302,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -17404,7 +17310,7 @@
 			</output>
 		</layer>
 		<layer id="1195" name="onnx::MatMul_1718/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="12692193" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -17412,32 +17318,32 @@
 			</output>
 		</layer>
 		<layer id="1196" name="Convert_1765" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1197" name="Subtract_1766" type="Subtract" version="opset1">
+		<layer id="1197" name="onnx::MatMul_1718/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -17445,7 +17351,7 @@
 			</output>
 		</layer>
 		<layer id="1198" name="onnx::MatMul_1718/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="24, 1, 768" offset="34491101" size="36864" />
+			<data element_type="f16" shape="24, 1, 768" offset="36027109" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>24</dim>
@@ -17454,57 +17360,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1199" name="onnx::MatMul_1718/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1200" name="onnx::MatMul_1718/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1199" name="onnx::MatMul_1718/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1201" name="Constant_1771" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="8723149" size="16" />
+		<layer id="1200" name="Constant_1770" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="10259151" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1202" name="Reshape_1772" type="Reshape" version="opset1">
+		<layer id="1201" name="Reshape_1771" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -17514,7 +17403,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1202" name="onnx::MatMul_1718/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
@@ -17525,7 +17429,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -17536,7 +17440,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/mlp/mlp.2/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -17549,14 +17453,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/mlp/mlp.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -17566,19 +17470,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.7/Add_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -17588,20 +17492,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.8/attn_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1207" name="onnx::Add_1743" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="34527965" size="3072" />
+			<data element_type="f32" shape="768" offset="36063973" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1743">
 					<dim>768</dim>
@@ -17609,7 +17513,7 @@
 			</output>
 		</layer>
 		<layer id="1208" name="onnx::Add_1720" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="34531037" size="3072" />
+			<data element_type="f32" shape="768" offset="36067045" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1720">
 					<dim>768</dim>
@@ -17617,13 +17521,13 @@
 			</output>
 		</layer>
 		<layer id="1209" name="Constant_892" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="1210" name="Constant_891" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -17652,7 +17556,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -17662,13 +17566,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1214" name="blocks.8.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="34534109" size="3072" />
+			<data element_type="f32" shape="768" offset="36070117" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.8.attn_ln.weight">
 					<dim>768</dim>
@@ -17680,7 +17584,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -17690,13 +17594,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1216" name="blocks.8.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="34537181" size="3072" />
+			<data element_type="f32" shape="768" offset="36073189" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.8.attn_ln.bias">
 					<dim>768</dim>
@@ -17708,7 +17612,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -17718,7 +17622,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -17728,20 +17632,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.8/attn_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1219" name="onnx::MatMul_1721" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="34540253" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="36076261" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -17751,7 +17655,7 @@
 			</output>
 		</layer>
 		<layer id="1220" name="Convert_1796" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -17760,7 +17664,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -17768,7 +17672,7 @@
 			</output>
 		</layer>
 		<layer id="1221" name="onnx::MatMul_1721/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7491262" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -17776,32 +17680,32 @@
 			</output>
 		</layer>
 		<layer id="1222" name="Convert_1798" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1223" name="Subtract_1799" type="Subtract" version="opset1">
+		<layer id="1223" name="onnx::MatMul_1721/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -17809,7 +17713,7 @@
 			</output>
 		</layer>
 		<layer id="1224" name="onnx::MatMul_1721/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="34835165" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="36371173" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -17818,57 +17722,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1225" name="onnx::MatMul_1721/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1226" name="onnx::MatMul_1721/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1225" name="onnx::MatMul_1721/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1227" name="Constant_1804" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1226" name="Constant_1803" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1228" name="Reshape_1805" type="Reshape" version="opset1">
+		<layer id="1227" name="Reshape_1804" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -17878,7 +17765,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1228" name="onnx::MatMul_1721/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -17889,7 +17791,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -17900,7 +17802,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/query/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -17913,20 +17815,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/query/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1231" name="/blocks.8/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.8/attn/Constant_2_output_0">
 					<dim>4</dim>
@@ -17938,7 +17840,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -17948,14 +17850,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/Reshape_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1233" name="Constant_908" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -17966,7 +17868,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -17978,19 +17880,19 @@
 				<port id="2" precision="FP32" names="/blocks.8/attn/Transpose_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1235" name="/blocks.8/attn/Constant" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353109" size="4" />
+			<data element_type="f32" shape="" offset="6889109" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.8/attn/Constant_output_0" />
 			</output>
 		</layer>
 		<layer id="1236" name="/blocks.8/attn/Constant_1" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353113" size="4" />
+			<data element_type="f32" shape="" offset="6889113" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.8/attn/Constant_1_output_0" />
 			</output>
@@ -18011,7 +17913,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -18020,13 +17922,13 @@
 				<port id="2" precision="FP32" names="/blocks.8/attn/Mul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1239" name="onnx::MatMul_1723" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="34844381" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="36380389" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -18036,7 +17938,7 @@
 			</output>
 		</layer>
 		<layer id="1240" name="Convert_1785" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -18045,7 +17947,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -18053,7 +17955,7 @@
 			</output>
 		</layer>
 		<layer id="1241" name="onnx::MatMul_1723/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7491262" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -18061,32 +17963,32 @@
 			</output>
 		</layer>
 		<layer id="1242" name="Convert_1787" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1243" name="Subtract_1788" type="Subtract" version="opset1">
+		<layer id="1243" name="onnx::MatMul_1723/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -18094,7 +17996,7 @@
 			</output>
 		</layer>
 		<layer id="1244" name="onnx::MatMul_1723/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="35139293" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="36675301" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -18103,67 +18005,65 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1245" name="onnx::MatMul_1723/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1246" name="onnx::MatMul_1723/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1245" name="onnx::MatMul_1723/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1247" name="Constant_1793" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1246" name="Constant_1792" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1248" name="Reshape_1794" type="Reshape" version="opset1">
+		<layer id="1247" name="Reshape_1793" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="I64">
-					<dim>2</dim>
-				</port>
+				<port id="1" precision="I64">
+					<dim>2</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1248" name="onnx::MatMul_1723/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -18174,7 +18074,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -18185,13 +18085,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/key/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1250" name="/blocks.8/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.8/attn/Constant_3_output_0">
 					<dim>4</dim>
@@ -18203,7 +18103,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -18213,14 +18113,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/Reshape_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1252" name="Constant_913" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5657245" size="32" />
+			<data element_type="i64" shape="4" offset="7193246" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -18231,7 +18131,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -18244,7 +18144,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -18255,7 +18155,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -18264,7 +18164,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -18274,22 +18174,22 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/MatMul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -18299,16 +18199,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.8/attn/Cast_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -18318,16 +18218,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.8/attn/Softmax_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -18337,21 +18237,21 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.8/attn/Cast_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1259" name="onnx::Add_1725" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="35148509" size="3072" />
+			<data element_type="f32" shape="768" offset="36684517" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1725">
 					<dim>768</dim>
@@ -18359,7 +18259,7 @@
 			</output>
 		</layer>
 		<layer id="1260" name="onnx::MatMul_1726" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="35151581" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="36687589" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -18369,7 +18269,7 @@
 			</output>
 		</layer>
 		<layer id="1261" name="Convert_1774" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -18378,7 +18278,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -18386,7 +18286,7 @@
 			</output>
 		</layer>
 		<layer id="1262" name="onnx::MatMul_1726/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -18394,32 +18294,32 @@
 			</output>
 		</layer>
 		<layer id="1263" name="Convert_1776" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1264" name="Subtract_1777" type="Subtract" version="opset1">
+		<layer id="1264" name="onnx::MatMul_1726/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -18427,7 +18327,7 @@
 			</output>
 		</layer>
 		<layer id="1265" name="onnx::MatMul_1726/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="35446493" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="36982501" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -18436,57 +18336,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1266" name="onnx::MatMul_1726/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1267" name="onnx::MatMul_1726/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1266" name="onnx::MatMul_1726/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1268" name="Constant_1782" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1267" name="Constant_1781" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1269" name="Reshape_1783" type="Reshape" version="opset1">
+		<layer id="1268" name="Reshape_1782" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -18496,7 +18379,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1269" name="onnx::MatMul_1726/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -18507,7 +18405,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -18518,7 +18416,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/value/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -18531,20 +18429,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/value/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1272" name="/blocks.8/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.8/attn/Constant_4_output_0">
 					<dim>4</dim>
@@ -18556,7 +18454,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -18566,14 +18464,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/Reshape_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1274" name="Constant_918" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -18584,7 +18482,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -18596,7 +18494,7 @@
 				<port id="2" precision="FP32" names="/blocks.8/attn/Transpose_2_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
@@ -18607,13 +18505,13 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</input>
@@ -18621,13 +18519,13 @@
 				<port id="2" precision="FP32" names="/blocks.8/attn/MatMul_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1277" name="Constant_925" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -18639,7 +18537,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -18649,7 +18547,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/Transpose_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -18660,7 +18558,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -18672,7 +18570,7 @@
 			</output>
 		</layer>
 		<layer id="1280" name="/blocks.8/attn/Constant_6" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.8/attn/Constant_6_output_0">
 					<dim>1</dim>
@@ -18725,7 +18623,7 @@
 			</output>
 		</layer>
 		<layer id="1285" name="/blocks.8/attn/Constant_5" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.8/attn/Constant_5_output_0">
 					<dim>1</dim>
@@ -18757,7 +18655,7 @@
 			</output>
 		</layer>
 		<layer id="1287" name="/blocks.8/attn/Constant_8" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042740" size="8" />
+			<data element_type="i64" shape="1" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.8/attn/Constant_8_output_0">
 					<dim>1</dim>
@@ -18785,7 +18683,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -18796,13 +18694,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/Reshape_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1290" name="onnx::MatMul_1744" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="35455709" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="36991717" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -18812,7 +18710,7 @@
 			</output>
 		</layer>
 		<layer id="1291" name="Convert_1807" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -18821,7 +18719,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -18829,7 +18727,7 @@
 			</output>
 		</layer>
 		<layer id="1292" name="onnx::MatMul_1744/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -18837,32 +18735,32 @@
 			</output>
 		</layer>
 		<layer id="1293" name="Convert_1809" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1294" name="Subtract_1810" type="Subtract" version="opset1">
+		<layer id="1294" name="onnx::MatMul_1744/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -18870,7 +18768,7 @@
 			</output>
 		</layer>
 		<layer id="1295" name="onnx::MatMul_1744/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="35750621" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="37286629" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -18879,57 +18777,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1296" name="onnx::MatMul_1744/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1297" name="onnx::MatMul_1744/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1296" name="onnx::MatMul_1744/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1298" name="Constant_1815" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1297" name="Constant_1814" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1299" name="Reshape_1816" type="Reshape" version="opset1">
+		<layer id="1298" name="Reshape_1815" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -18939,7 +18820,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1299" name="onnx::MatMul_1744/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -18950,7 +18846,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -18961,7 +18857,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/out/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -18974,14 +18870,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/attn/out/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -18991,19 +18887,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -19013,20 +18909,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.8/mlp_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1304" name="onnx::Add_1749" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="35759837" size="3072" />
+			<data element_type="f32" shape="768" offset="37295845" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1749">
 					<dim>768</dim>
@@ -19034,7 +18930,7 @@
 			</output>
 		</layer>
 		<layer id="1305" name="onnx::Add_1746" type="Const" version="opset1">
-			<data element_type="f32" shape="3072" offset="35762909" size="12288" />
+			<data element_type="f32" shape="3072" offset="37298917" size="12288" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1746">
 					<dim>3072</dim>
@@ -19042,13 +18938,13 @@
 			</output>
 		</layer>
 		<layer id="1306" name="Constant_951" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="1307" name="Constant_950" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -19077,7 +18973,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -19087,13 +18983,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1311" name="blocks.8.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="35775197" size="3072" />
+			<data element_type="f32" shape="768" offset="37311205" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.8.mlp_ln.weight">
 					<dim>768</dim>
@@ -19105,7 +19001,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -19115,13 +19011,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1313" name="blocks.8.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="35778269" size="3072" />
+			<data element_type="f32" shape="768" offset="37314277" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.8.mlp_ln.bias">
 					<dim>768</dim>
@@ -19133,7 +19029,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -19143,7 +19039,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/mlp_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -19153,20 +19049,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.8/mlp_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1316" name="onnx::MatMul_1747" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 3072" offset="35781341" size="1179648" />
+			<data element_type="u4" shape="6, 128, 3072" offset="37317349" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -19176,7 +19072,7 @@
 			</output>
 		</layer>
 		<layer id="1317" name="Convert_1818" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -19185,7 +19081,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -19193,7 +19089,7 @@
 			</output>
 		</layer>
 		<layer id="1318" name="onnx::MatMul_1747/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -19201,32 +19097,32 @@
 			</output>
 		</layer>
 		<layer id="1319" name="Convert_1820" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1320" name="Subtract_1821" type="Subtract" version="opset1">
+		<layer id="1320" name="onnx::MatMul_1747/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -19234,7 +19130,7 @@
 			</output>
 		</layer>
 		<layer id="1321" name="onnx::MatMul_1747/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 3072" offset="36960989" size="36864" />
+			<data element_type="f16" shape="6, 1, 3072" offset="38496997" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -19243,57 +19139,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1322" name="onnx::MatMul_1747/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1323" name="onnx::MatMul_1747/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1322" name="onnx::MatMul_1747/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1324" name="Constant_1826" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="7506621" size="16" />
+		<layer id="1323" name="Constant_1825" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="9042623" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1325" name="Reshape_1827" type="Reshape" version="opset1">
+		<layer id="1324" name="Reshape_1826" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -19303,7 +19182,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1325" name="onnx::MatMul_1747/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
@@ -19314,7 +19208,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -19325,7 +19219,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/mlp/mlp.0/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -19338,14 +19232,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/mlp/mlp.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -19361,7 +19255,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -19369,7 +19263,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/mlp/mlp.1/Div_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -19378,14 +19272,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.8/mlp/mlp.1/Erf_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -19401,7 +19295,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -19409,7 +19303,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/mlp/mlp.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -19419,19 +19313,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/mlp/mlp.1/Mul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -19447,7 +19341,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -19455,13 +19349,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/mlp/mlp.1/Mul_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1336" name="onnx::MatMul_1750" type="Const" version="opset1">
-			<data element_type="u4" shape="24, 128, 768" offset="36997853" size="1179648" />
+			<data element_type="u4" shape="24, 128, 768" offset="38533861" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -19471,7 +19365,7 @@
 			</output>
 		</layer>
 		<layer id="1337" name="Convert_1829" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -19480,7 +19374,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -19488,7 +19382,7 @@
 			</output>
 		</layer>
 		<layer id="1338" name="onnx::MatMul_1750/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="39713509" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -19496,32 +19390,32 @@
 			</output>
 		</layer>
 		<layer id="1339" name="Convert_1831" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1340" name="Subtract_1832" type="Subtract" version="opset1">
+		<layer id="1340" name="onnx::MatMul_1750/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -19529,7 +19423,7 @@
 			</output>
 		</layer>
 		<layer id="1341" name="onnx::MatMul_1750/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="24, 1, 768" offset="38177501" size="36864" />
+			<data element_type="f16" shape="24, 1, 768" offset="39713510" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>24</dim>
@@ -19538,57 +19432,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1342" name="onnx::MatMul_1750/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1343" name="onnx::MatMul_1750/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1342" name="onnx::MatMul_1750/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1344" name="Constant_1837" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="8723149" size="16" />
+		<layer id="1343" name="Constant_1836" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="10259151" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1345" name="Reshape_1838" type="Reshape" version="opset1">
+		<layer id="1344" name="Reshape_1837" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -19598,7 +19475,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1345" name="onnx::MatMul_1750/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
@@ -19609,7 +19501,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -19620,7 +19512,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/mlp/mlp.2/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -19633,14 +19525,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/mlp/mlp.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -19650,19 +19542,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.8/Add_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -19672,20 +19564,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.9/attn_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1350" name="onnx::Add_1775" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="38214365" size="3072" />
+			<data element_type="f32" shape="768" offset="39750374" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1775">
 					<dim>768</dim>
@@ -19693,7 +19585,7 @@
 			</output>
 		</layer>
 		<layer id="1351" name="onnx::Add_1752" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="38217437" size="3072" />
+			<data element_type="f32" shape="768" offset="39753446" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1752">
 					<dim>768</dim>
@@ -19701,13 +19593,13 @@
 			</output>
 		</layer>
 		<layer id="1352" name="Constant_973" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="1353" name="Constant_972" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -19736,7 +19628,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -19746,13 +19638,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1357" name="blocks.9.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="38220509" size="3072" />
+			<data element_type="f32" shape="768" offset="39756518" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.9.attn_ln.weight">
 					<dim>768</dim>
@@ -19764,7 +19656,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -19774,13 +19666,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1359" name="blocks.9.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="38223581" size="3072" />
+			<data element_type="f32" shape="768" offset="39759590" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.9.attn_ln.bias">
 					<dim>768</dim>
@@ -19792,7 +19684,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -19802,7 +19694,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -19812,20 +19704,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.9/attn_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1362" name="onnx::MatMul_1753" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="38226653" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="39762662" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -19835,7 +19727,7 @@
 			</output>
 		</layer>
 		<layer id="1363" name="Convert_1862" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -19844,7 +19736,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -19852,7 +19744,7 @@
 			</output>
 		</layer>
 		<layer id="1364" name="onnx::MatMul_1753/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -19860,32 +19752,32 @@
 			</output>
 		</layer>
 		<layer id="1365" name="Convert_1864" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1366" name="Subtract_1865" type="Subtract" version="opset1">
+		<layer id="1366" name="onnx::MatMul_1753/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -19893,7 +19785,7 @@
 			</output>
 		</layer>
 		<layer id="1367" name="onnx::MatMul_1753/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="38521565" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="40057574" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -19902,57 +19794,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1368" name="onnx::MatMul_1753/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1369" name="onnx::MatMul_1753/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1368" name="onnx::MatMul_1753/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1370" name="Constant_1870" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1369" name="Constant_1869" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1371" name="Reshape_1871" type="Reshape" version="opset1">
+		<layer id="1370" name="Reshape_1870" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -19962,7 +19837,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1371" name="onnx::MatMul_1753/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -19973,7 +19863,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -19984,7 +19874,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/query/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -19997,20 +19887,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/query/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1374" name="/blocks.9/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.9/attn/Constant_2_output_0">
 					<dim>4</dim>
@@ -20022,7 +19912,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -20032,14 +19922,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/Reshape_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1376" name="Constant_989" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -20050,7 +19940,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -20062,19 +19952,19 @@
 				<port id="2" precision="FP32" names="/blocks.9/attn/Transpose_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1378" name="/blocks.9/attn/Constant" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353109" size="4" />
+			<data element_type="f32" shape="" offset="6889109" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.9/attn/Constant_output_0" />
 			</output>
 		</layer>
 		<layer id="1379" name="/blocks.9/attn/Constant_1" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353113" size="4" />
+			<data element_type="f32" shape="" offset="6889113" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.9/attn/Constant_1_output_0" />
 			</output>
@@ -20095,7 +19985,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -20104,13 +19994,13 @@
 				<port id="2" precision="FP32" names="/blocks.9/attn/Mul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1382" name="onnx::MatMul_1755" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="38530781" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="40066790" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -20120,7 +20010,7 @@
 			</output>
 		</layer>
 		<layer id="1383" name="Convert_1851" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -20129,7 +20019,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -20137,7 +20027,7 @@
 			</output>
 		</layer>
 		<layer id="1384" name="onnx::MatMul_1755/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -20145,99 +20035,82 @@
 			</output>
 		</layer>
 		<layer id="1385" name="Convert_1853" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1386" name="Subtract_1854" type="Subtract" version="opset1">
+		<layer id="1386" name="onnx::MatMul_1755/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>6</dim>
-					<dim>128</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1387" name="onnx::MatMul_1755/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="38825693" size="9216" />
-			<output>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1388" name="onnx::MatMul_1755/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
-					<dim>1</dim>
+					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-			</input>
+			</output>
+		</layer>
+		<layer id="1387" name="onnx::MatMul_1755/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="6, 1, 768" offset="40361702" size="9216" />
 			<output>
-				<port id="1" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1389" name="onnx::MatMul_1755/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1388" name="onnx::MatMul_1755/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1390" name="Constant_1859" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1389" name="Constant_1858" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1391" name="Reshape_1860" type="Reshape" version="opset1">
+		<layer id="1390" name="Reshape_1859" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -20247,7 +20120,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1391" name="onnx::MatMul_1755/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -20258,7 +20146,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -20269,13 +20157,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/key/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1393" name="/blocks.9/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.9/attn/Constant_3_output_0">
 					<dim>4</dim>
@@ -20287,7 +20175,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -20297,14 +20185,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/Reshape_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1395" name="Constant_994" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5657245" size="32" />
+			<data element_type="i64" shape="4" offset="7193246" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -20315,7 +20203,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -20328,7 +20216,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -20339,7 +20227,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -20348,7 +20236,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -20358,22 +20246,22 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/MatMul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -20383,16 +20271,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.9/attn/Cast_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -20402,16 +20290,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.9/attn/Softmax_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -20421,21 +20309,21 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.9/attn/Cast_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1402" name="onnx::Add_1757" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="38834909" size="3072" />
+			<data element_type="f32" shape="768" offset="40370918" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1757">
 					<dim>768</dim>
@@ -20443,7 +20331,7 @@
 			</output>
 		</layer>
 		<layer id="1403" name="onnx::MatMul_1758" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="38837981" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="40373990" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -20453,7 +20341,7 @@
 			</output>
 		</layer>
 		<layer id="1404" name="Convert_1840" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -20462,7 +20350,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -20470,7 +20358,7 @@
 			</output>
 		</layer>
 		<layer id="1405" name="onnx::MatMul_1758/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -20478,32 +20366,32 @@
 			</output>
 		</layer>
 		<layer id="1406" name="Convert_1842" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1407" name="Subtract_1843" type="Subtract" version="opset1">
+		<layer id="1407" name="onnx::MatMul_1758/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -20511,7 +20399,7 @@
 			</output>
 		</layer>
 		<layer id="1408" name="onnx::MatMul_1758/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="39132893" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="40668902" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -20520,57 +20408,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1409" name="onnx::MatMul_1758/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1410" name="onnx::MatMul_1758/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1409" name="onnx::MatMul_1758/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1411" name="Constant_1848" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1410" name="Constant_1847" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1412" name="Reshape_1849" type="Reshape" version="opset1">
+		<layer id="1411" name="Reshape_1848" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -20580,7 +20451,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1412" name="onnx::MatMul_1758/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -20591,7 +20477,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -20602,7 +20488,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/value/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -20615,20 +20501,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/value/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1415" name="/blocks.9/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.9/attn/Constant_4_output_0">
 					<dim>4</dim>
@@ -20640,7 +20526,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -20650,14 +20536,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/Reshape_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1417" name="Constant_999" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -20668,7 +20554,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -20680,7 +20566,7 @@
 				<port id="2" precision="FP32" names="/blocks.9/attn/Transpose_2_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
@@ -20691,13 +20577,13 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</input>
@@ -20705,13 +20591,13 @@
 				<port id="2" precision="FP32" names="/blocks.9/attn/MatMul_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1420" name="Constant_1006" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -20723,7 +20609,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -20733,7 +20619,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/Transpose_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -20744,7 +20630,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -20756,7 +20642,7 @@
 			</output>
 		</layer>
 		<layer id="1423" name="/blocks.9/attn/Constant_6" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.9/attn/Constant_6_output_0">
 					<dim>1</dim>
@@ -20809,7 +20695,7 @@
 			</output>
 		</layer>
 		<layer id="1428" name="/blocks.9/attn/Constant_5" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.9/attn/Constant_5_output_0">
 					<dim>1</dim>
@@ -20841,7 +20727,7 @@
 			</output>
 		</layer>
 		<layer id="1430" name="/blocks.9/attn/Constant_8" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042740" size="8" />
+			<data element_type="i64" shape="1" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.9/attn/Constant_8_output_0">
 					<dim>1</dim>
@@ -20869,7 +20755,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -20880,13 +20766,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/Reshape_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1433" name="onnx::MatMul_1776" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="39142109" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="40678118" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -20896,7 +20782,7 @@
 			</output>
 		</layer>
 		<layer id="1434" name="Convert_1873" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -20905,7 +20791,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -20913,7 +20799,7 @@
 			</output>
 		</layer>
 		<layer id="1435" name="onnx::MatMul_1776/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -20921,32 +20807,32 @@
 			</output>
 		</layer>
 		<layer id="1436" name="Convert_1875" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1437" name="Subtract_1876" type="Subtract" version="opset1">
+		<layer id="1437" name="onnx::MatMul_1776/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -20954,7 +20840,7 @@
 			</output>
 		</layer>
 		<layer id="1438" name="onnx::MatMul_1776/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="39437021" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="40973030" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -20963,57 +20849,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1439" name="onnx::MatMul_1776/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1440" name="onnx::MatMul_1776/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1439" name="onnx::MatMul_1776/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1441" name="Constant_1881" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1440" name="Constant_1880" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1442" name="Reshape_1882" type="Reshape" version="opset1">
+		<layer id="1441" name="Reshape_1881" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -21023,7 +20892,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1442" name="onnx::MatMul_1776/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -21034,7 +20918,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -21045,7 +20929,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/out/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -21058,14 +20942,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/attn/out/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -21075,19 +20959,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -21097,20 +20981,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.9/mlp_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1447" name="onnx::Add_1781" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="39446237" size="3072" />
+			<data element_type="f32" shape="768" offset="40982246" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1781">
 					<dim>768</dim>
@@ -21118,7 +21002,7 @@
 			</output>
 		</layer>
 		<layer id="1448" name="onnx::Add_1778" type="Const" version="opset1">
-			<data element_type="f32" shape="3072" offset="39449309" size="12288" />
+			<data element_type="f32" shape="3072" offset="40985318" size="12288" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1778">
 					<dim>3072</dim>
@@ -21126,13 +21010,13 @@
 			</output>
 		</layer>
 		<layer id="1449" name="Constant_1032" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="1450" name="Constant_1031" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -21161,7 +21045,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -21171,13 +21055,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1454" name="blocks.9.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="39461597" size="3072" />
+			<data element_type="f32" shape="768" offset="40997606" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.9.mlp_ln.weight">
 					<dim>768</dim>
@@ -21189,7 +21073,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -21199,13 +21083,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1456" name="blocks.9.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="39464669" size="3072" />
+			<data element_type="f32" shape="768" offset="41000678" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.9.mlp_ln.bias">
 					<dim>768</dim>
@@ -21217,7 +21101,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -21227,7 +21111,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/mlp_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -21237,20 +21121,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.9/mlp_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1459" name="onnx::MatMul_1779" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 3072" offset="39467741" size="1179648" />
+			<data element_type="u4" shape="6, 128, 3072" offset="41003750" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -21260,7 +21144,7 @@
 			</output>
 		</layer>
 		<layer id="1460" name="Convert_1884" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -21269,7 +21153,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -21277,7 +21161,7 @@
 			</output>
 		</layer>
 		<layer id="1461" name="onnx::MatMul_1779/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="20064995" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -21285,32 +21169,32 @@
 			</output>
 		</layer>
 		<layer id="1462" name="Convert_1886" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1463" name="Subtract_1887" type="Subtract" version="opset1">
+		<layer id="1463" name="onnx::MatMul_1779/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -21318,7 +21202,7 @@
 			</output>
 		</layer>
 		<layer id="1464" name="onnx::MatMul_1779/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 3072" offset="40647389" size="36864" />
+			<data element_type="f16" shape="6, 1, 3072" offset="42183398" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -21327,57 +21211,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1465" name="onnx::MatMul_1779/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1466" name="onnx::MatMul_1779/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1465" name="onnx::MatMul_1779/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1467" name="Constant_1892" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="7506621" size="16" />
+		<layer id="1466" name="Constant_1891" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="9042623" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1468" name="Reshape_1893" type="Reshape" version="opset1">
+		<layer id="1467" name="Reshape_1892" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -21387,7 +21254,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1468" name="onnx::MatMul_1779/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
@@ -21398,7 +21280,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -21409,7 +21291,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/mlp/mlp.0/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -21422,14 +21304,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/mlp/mlp.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -21445,7 +21327,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -21453,7 +21335,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/mlp/mlp.1/Div_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -21462,14 +21344,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.9/mlp/mlp.1/Erf_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -21485,7 +21367,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -21493,7 +21375,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/mlp/mlp.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -21503,19 +21385,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/mlp/mlp.1/Mul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -21531,7 +21413,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -21539,13 +21421,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/mlp/mlp.1/Mul_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1479" name="onnx::MatMul_1782" type="Const" version="opset1">
-			<data element_type="u4" shape="24, 128, 768" offset="40684253" size="1179648" />
+			<data element_type="u4" shape="24, 128, 768" offset="42220262" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -21555,7 +21437,7 @@
 			</output>
 		</layer>
 		<layer id="1480" name="Convert_1895" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -21564,7 +21446,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -21572,7 +21454,7 @@
 			</output>
 		</layer>
 		<layer id="1481" name="onnx::MatMul_1782/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -21580,32 +21462,32 @@
 			</output>
 		</layer>
 		<layer id="1482" name="Convert_1897" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1483" name="Subtract_1898" type="Subtract" version="opset1">
+		<layer id="1483" name="onnx::MatMul_1782/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -21613,7 +21495,7 @@
 			</output>
 		</layer>
 		<layer id="1484" name="onnx::MatMul_1782/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="24, 1, 768" offset="41863901" size="36864" />
+			<data element_type="f16" shape="24, 1, 768" offset="43399910" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>24</dim>
@@ -21622,57 +21504,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1485" name="onnx::MatMul_1782/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1486" name="onnx::MatMul_1782/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1485" name="onnx::MatMul_1782/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1487" name="Constant_1903" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="8723149" size="16" />
+		<layer id="1486" name="Constant_1902" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="10259151" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1488" name="Reshape_1904" type="Reshape" version="opset1">
+		<layer id="1487" name="Reshape_1903" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -21682,7 +21547,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1488" name="onnx::MatMul_1782/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
@@ -21693,7 +21573,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -21704,7 +21584,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/mlp/mlp.2/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -21717,14 +21597,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/mlp/mlp.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -21734,19 +21614,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.9/Add_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -21756,20 +21636,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.10/attn_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1493" name="onnx::Add_1807" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="41900765" size="3072" />
+			<data element_type="f32" shape="768" offset="43436774" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1807">
 					<dim>768</dim>
@@ -21777,7 +21657,7 @@
 			</output>
 		</layer>
 		<layer id="1494" name="onnx::Add_1784" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="41903837" size="3072" />
+			<data element_type="f32" shape="768" offset="43439846" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1784">
 					<dim>768</dim>
@@ -21785,13 +21665,13 @@
 			</output>
 		</layer>
 		<layer id="1495" name="Constant_1054" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="1496" name="Constant_1053" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -21820,7 +21700,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -21830,13 +21710,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1500" name="blocks.10.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="41906909" size="3072" />
+			<data element_type="f32" shape="768" offset="43442918" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.10.attn_ln.weight">
 					<dim>768</dim>
@@ -21848,7 +21728,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -21858,13 +21738,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1502" name="blocks.10.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="41909981" size="3072" />
+			<data element_type="f32" shape="768" offset="43445990" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.10.attn_ln.bias">
 					<dim>768</dim>
@@ -21876,7 +21756,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -21886,7 +21766,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -21896,20 +21776,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.10/attn_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1505" name="onnx::MatMul_1785" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="41913053" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="43449062" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -21919,7 +21799,7 @@
 			</output>
 		</layer>
 		<layer id="1506" name="Convert_1928" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -21928,7 +21808,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -21936,7 +21816,7 @@
 			</output>
 		</layer>
 		<layer id="1507" name="onnx::MatMul_1785/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -21944,32 +21824,32 @@
 			</output>
 		</layer>
 		<layer id="1508" name="Convert_1930" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1509" name="Subtract_1931" type="Subtract" version="opset1">
+		<layer id="1509" name="onnx::MatMul_1785/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -21977,7 +21857,7 @@
 			</output>
 		</layer>
 		<layer id="1510" name="onnx::MatMul_1785/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="42207965" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="43743974" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -21986,57 +21866,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1511" name="onnx::MatMul_1785/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1512" name="onnx::MatMul_1785/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1511" name="onnx::MatMul_1785/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1513" name="Constant_1936" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1512" name="Constant_1935" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1514" name="Reshape_1937" type="Reshape" version="opset1">
+		<layer id="1513" name="Reshape_1936" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -22046,7 +21909,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1514" name="onnx::MatMul_1785/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -22057,7 +21935,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -22068,7 +21946,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/query/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -22081,20 +21959,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/query/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1517" name="/blocks.10/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.10/attn/Constant_2_output_0">
 					<dim>4</dim>
@@ -22106,7 +21984,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -22116,14 +21994,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/Reshape_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1519" name="Constant_1070" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -22134,7 +22012,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -22146,19 +22024,19 @@
 				<port id="2" precision="FP32" names="/blocks.10/attn/Transpose_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1521" name="/blocks.10/attn/Constant" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353109" size="4" />
+			<data element_type="f32" shape="" offset="6889109" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.10/attn/Constant_output_0" />
 			</output>
 		</layer>
 		<layer id="1522" name="/blocks.10/attn/Constant_1" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353113" size="4" />
+			<data element_type="f32" shape="" offset="6889113" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.10/attn/Constant_1_output_0" />
 			</output>
@@ -22179,7 +22057,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -22188,13 +22066,13 @@
 				<port id="2" precision="FP32" names="/blocks.10/attn/Mul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1525" name="onnx::MatMul_1787" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="42217181" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="43753190" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -22204,7 +22082,7 @@
 			</output>
 		</layer>
 		<layer id="1526" name="Convert_1917" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -22213,7 +22091,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -22221,7 +22099,7 @@
 			</output>
 		</layer>
 		<layer id="1527" name="onnx::MatMul_1787/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -22229,32 +22107,32 @@
 			</output>
 		</layer>
 		<layer id="1528" name="Convert_1919" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1529" name="Subtract_1920" type="Subtract" version="opset1">
+		<layer id="1529" name="onnx::MatMul_1787/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -22262,7 +22140,7 @@
 			</output>
 		</layer>
 		<layer id="1530" name="onnx::MatMul_1787/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="42512093" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="44048102" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -22271,57 +22149,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1531" name="onnx::MatMul_1787/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1532" name="onnx::MatMul_1787/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1531" name="onnx::MatMul_1787/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1533" name="Constant_1925" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1532" name="Constant_1924" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1534" name="Reshape_1926" type="Reshape" version="opset1">
+		<layer id="1533" name="Reshape_1925" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -22331,7 +22192,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1534" name="onnx::MatMul_1787/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -22342,7 +22218,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -22353,13 +22229,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/key/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1536" name="/blocks.10/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.10/attn/Constant_3_output_0">
 					<dim>4</dim>
@@ -22371,7 +22247,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -22381,14 +22257,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/Reshape_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1538" name="Constant_1075" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5657245" size="32" />
+			<data element_type="i64" shape="4" offset="7193246" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -22399,7 +22275,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -22412,7 +22288,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -22423,7 +22299,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -22432,7 +22308,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -22442,22 +22318,22 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/MatMul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -22467,16 +22343,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.10/attn/Cast_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -22486,16 +22362,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.10/attn/Softmax_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -22505,21 +22381,21 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.10/attn/Cast_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1545" name="onnx::Add_1789" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="42521309" size="3072" />
+			<data element_type="f32" shape="768" offset="44057318" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1789">
 					<dim>768</dim>
@@ -22527,7 +22403,7 @@
 			</output>
 		</layer>
 		<layer id="1546" name="onnx::MatMul_1790" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="42524381" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="44060390" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -22537,7 +22413,7 @@
 			</output>
 		</layer>
 		<layer id="1547" name="Convert_1906" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -22546,7 +22422,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -22554,7 +22430,7 @@
 			</output>
 		</layer>
 		<layer id="1548" name="onnx::MatMul_1790/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="44355302" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -22562,32 +22438,32 @@
 			</output>
 		</layer>
 		<layer id="1549" name="Convert_1908" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1550" name="Subtract_1909" type="Subtract" version="opset1">
+		<layer id="1550" name="onnx::MatMul_1790/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -22595,7 +22471,7 @@
 			</output>
 		</layer>
 		<layer id="1551" name="onnx::MatMul_1790/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="42819293" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="44355303" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -22604,57 +22480,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1552" name="onnx::MatMul_1790/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1553" name="onnx::MatMul_1790/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1552" name="onnx::MatMul_1790/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1554" name="Constant_1914" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1553" name="Constant_1913" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1555" name="Reshape_1915" type="Reshape" version="opset1">
+		<layer id="1554" name="Reshape_1914" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -22664,7 +22523,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1555" name="onnx::MatMul_1790/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -22675,7 +22549,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -22686,7 +22560,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/value/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -22699,20 +22573,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/value/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1558" name="/blocks.10/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.10/attn/Constant_4_output_0">
 					<dim>4</dim>
@@ -22724,7 +22598,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -22734,14 +22608,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/Reshape_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1560" name="Constant_1080" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -22752,7 +22626,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -22764,7 +22638,7 @@
 				<port id="2" precision="FP32" names="/blocks.10/attn/Transpose_2_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
@@ -22775,13 +22649,13 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</input>
@@ -22789,13 +22663,13 @@
 				<port id="2" precision="FP32" names="/blocks.10/attn/MatMul_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1563" name="Constant_1087" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -22807,7 +22681,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -22817,7 +22691,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/Transpose_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -22828,7 +22702,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -22840,7 +22714,7 @@
 			</output>
 		</layer>
 		<layer id="1566" name="/blocks.10/attn/Constant_6" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.10/attn/Constant_6_output_0">
 					<dim>1</dim>
@@ -22893,7 +22767,7 @@
 			</output>
 		</layer>
 		<layer id="1571" name="/blocks.10/attn/Constant_5" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.10/attn/Constant_5_output_0">
 					<dim>1</dim>
@@ -22925,7 +22799,7 @@
 			</output>
 		</layer>
 		<layer id="1573" name="/blocks.10/attn/Constant_8" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042740" size="8" />
+			<data element_type="i64" shape="1" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.10/attn/Constant_8_output_0">
 					<dim>1</dim>
@@ -22953,7 +22827,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -22964,13 +22838,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/Reshape_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1576" name="onnx::MatMul_1808" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="42828509" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="44364519" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -22980,7 +22854,7 @@
 			</output>
 		</layer>
 		<layer id="1577" name="Convert_1939" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -22989,7 +22863,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -22997,7 +22871,7 @@
 			</output>
 		</layer>
 		<layer id="1578" name="onnx::MatMul_1808/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -23005,32 +22879,32 @@
 			</output>
 		</layer>
 		<layer id="1579" name="Convert_1941" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1580" name="Subtract_1942" type="Subtract" version="opset1">
+		<layer id="1580" name="onnx::MatMul_1808/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -23038,7 +22912,7 @@
 			</output>
 		</layer>
 		<layer id="1581" name="onnx::MatMul_1808/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="43123421" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="44659431" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -23047,57 +22921,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1582" name="onnx::MatMul_1808/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1583" name="onnx::MatMul_1808/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1582" name="onnx::MatMul_1808/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1584" name="Constant_1947" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1583" name="Constant_1946" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1585" name="Reshape_1948" type="Reshape" version="opset1">
+		<layer id="1584" name="Reshape_1947" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -23107,7 +22964,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1585" name="onnx::MatMul_1808/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -23118,7 +22990,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -23129,7 +23001,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/out/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -23142,14 +23014,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/attn/out/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -23159,19 +23031,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -23181,20 +23053,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.10/mlp_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1590" name="onnx::Add_1813" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="43132637" size="3072" />
+			<data element_type="f32" shape="768" offset="44668647" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1813">
 					<dim>768</dim>
@@ -23202,7 +23074,7 @@
 			</output>
 		</layer>
 		<layer id="1591" name="onnx::Add_1810" type="Const" version="opset1">
-			<data element_type="f32" shape="3072" offset="43135709" size="12288" />
+			<data element_type="f32" shape="3072" offset="44671719" size="12288" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1810">
 					<dim>3072</dim>
@@ -23210,13 +23082,13 @@
 			</output>
 		</layer>
 		<layer id="1592" name="Constant_1113" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="1593" name="Constant_1112" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -23245,7 +23117,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -23255,13 +23127,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1597" name="blocks.10.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="43147997" size="3072" />
+			<data element_type="f32" shape="768" offset="44684007" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.10.mlp_ln.weight">
 					<dim>768</dim>
@@ -23273,7 +23145,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -23283,13 +23155,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1599" name="blocks.10.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="43151069" size="3072" />
+			<data element_type="f32" shape="768" offset="44687079" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.10.mlp_ln.bias">
 					<dim>768</dim>
@@ -23301,7 +23173,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -23311,7 +23183,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/mlp_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -23321,20 +23193,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.10/mlp_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1602" name="onnx::MatMul_1811" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 3072" offset="43154141" size="1179648" />
+			<data element_type="u4" shape="6, 128, 3072" offset="44690151" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -23344,7 +23216,7 @@
 			</output>
 		</layer>
 		<layer id="1603" name="Convert_1950" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -23353,7 +23225,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -23361,7 +23233,7 @@
 			</output>
 		</layer>
 		<layer id="1604" name="onnx::MatMul_1811/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -23369,32 +23241,32 @@
 			</output>
 		</layer>
 		<layer id="1605" name="Convert_1952" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1606" name="Subtract_1953" type="Subtract" version="opset1">
+		<layer id="1606" name="onnx::MatMul_1811/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -23402,7 +23274,7 @@
 			</output>
 		</layer>
 		<layer id="1607" name="onnx::MatMul_1811/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 3072" offset="44333789" size="36864" />
+			<data element_type="f16" shape="6, 1, 3072" offset="45869799" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -23411,57 +23283,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1608" name="onnx::MatMul_1811/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1609" name="onnx::MatMul_1811/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1608" name="onnx::MatMul_1811/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1610" name="Constant_1958" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="7506621" size="16" />
+		<layer id="1609" name="Constant_1957" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="9042623" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1611" name="Reshape_1959" type="Reshape" version="opset1">
+		<layer id="1610" name="Reshape_1958" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -23471,7 +23326,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1611" name="onnx::MatMul_1811/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
@@ -23482,7 +23352,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -23493,7 +23363,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/mlp/mlp.0/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -23506,14 +23376,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/mlp/mlp.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -23529,7 +23399,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -23537,7 +23407,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/mlp/mlp.1/Div_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -23546,14 +23416,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.10/mlp/mlp.1/Erf_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -23569,7 +23439,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -23577,7 +23447,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/mlp/mlp.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -23587,19 +23457,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/mlp/mlp.1/Mul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -23615,7 +23485,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -23623,13 +23493,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/mlp/mlp.1/Mul_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1622" name="onnx::MatMul_1814" type="Const" version="opset1">
-			<data element_type="u4" shape="24, 128, 768" offset="44370653" size="1179648" />
+			<data element_type="u4" shape="24, 128, 768" offset="45906663" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -23639,7 +23509,7 @@
 			</output>
 		</layer>
 		<layer id="1623" name="Convert_1961" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>24</dim>
@@ -23648,7 +23518,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -23656,7 +23526,7 @@
 			</output>
 		</layer>
 		<layer id="1624" name="onnx::MatMul_1814/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -23664,32 +23534,32 @@
 			</output>
 		</layer>
 		<layer id="1625" name="Convert_1963" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1626" name="Subtract_1964" type="Subtract" version="opset1">
+		<layer id="1626" name="onnx::MatMul_1814/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -23697,7 +23567,7 @@
 			</output>
 		</layer>
 		<layer id="1627" name="onnx::MatMul_1814/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="24, 1, 768" offset="45550301" size="36864" />
+			<data element_type="f16" shape="24, 1, 768" offset="47086311" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>24</dim>
@@ -23706,57 +23576,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1628" name="onnx::MatMul_1814/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>24</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1629" name="onnx::MatMul_1814/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1628" name="onnx::MatMul_1814/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>24</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1630" name="Constant_1969" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="8723149" size="16" />
+		<layer id="1629" name="Constant_1968" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="10259151" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1631" name="Reshape_1970" type="Reshape" version="opset1">
+		<layer id="1630" name="Reshape_1969" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>24</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -23766,7 +23619,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1631" name="onnx::MatMul_1814/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
@@ -23777,7 +23645,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -23788,7 +23656,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/mlp/mlp.2/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -23801,14 +23669,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/mlp/mlp.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -23818,19 +23686,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.10/Add_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -23840,20 +23708,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.11/attn_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1636" name="onnx::Add_1839" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="45587165" size="3072" />
+			<data element_type="f32" shape="768" offset="47123175" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1839">
 					<dim>768</dim>
@@ -23861,7 +23729,7 @@
 			</output>
 		</layer>
 		<layer id="1637" name="onnx::Add_1816" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="45590237" size="3072" />
+			<data element_type="f32" shape="768" offset="47126247" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1816">
 					<dim>768</dim>
@@ -23869,13 +23737,13 @@
 			</output>
 		</layer>
 		<layer id="1638" name="Constant_1135" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="1639" name="Constant_1134" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -23904,7 +23772,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -23914,13 +23782,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1643" name="blocks.11.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="45593309" size="3072" />
+			<data element_type="f32" shape="768" offset="47129319" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.11.attn_ln.weight">
 					<dim>768</dim>
@@ -23932,7 +23800,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -23942,13 +23810,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1645" name="blocks.11.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="45596381" size="3072" />
+			<data element_type="f32" shape="768" offset="47132391" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.11.attn_ln.bias">
 					<dim>768</dim>
@@ -23960,7 +23828,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -23970,7 +23838,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -23980,20 +23848,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.11/attn_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1648" name="onnx::MatMul_1817" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="45599453" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="47135463" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -24003,7 +23871,7 @@
 			</output>
 		</layer>
 		<layer id="1649" name="Convert_1994" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -24012,7 +23880,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -24020,7 +23888,7 @@
 			</output>
 		</layer>
 		<layer id="1650" name="onnx::MatMul_1817/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -24028,99 +23896,82 @@
 			</output>
 		</layer>
 		<layer id="1651" name="Convert_1996" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1652" name="Subtract_1997" type="Subtract" version="opset1">
-			<data auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>6</dim>
-					<dim>128</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-				</port>
-			</input>
-			<output>
-				<port id="2" precision="FP32">
-					<dim>6</dim>
-					<dim>128</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1653" name="onnx::MatMul_1817/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="45894365" size="9216" />
-			<output>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
+				<port id="1" precision="FP16">
 					<dim>1</dim>
-					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1654" name="onnx::MatMul_1817/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1652" name="onnx::MatMul_1817/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
-					<dim>1</dim>
+					<dim>128</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>6</dim>
+					<dim>128</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1653" name="onnx::MatMul_1817/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="6, 1, 768" offset="47430375" size="9216" />
+			<output>
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1655" name="onnx::MatMul_1817/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1654" name="onnx::MatMul_1817/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1656" name="Constant_2002" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1655" name="Constant_2001" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1657" name="Reshape_2003" type="Reshape" version="opset1">
+		<layer id="1656" name="Reshape_2002" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -24130,7 +23981,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1657" name="onnx::MatMul_1817/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -24141,7 +24007,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -24152,7 +24018,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/query/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -24165,20 +24031,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/query/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1660" name="/blocks.11/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.11/attn/Constant_2_output_0">
 					<dim>4</dim>
@@ -24190,7 +24056,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -24200,14 +24066,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/Reshape_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1662" name="Constant_1151" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -24218,7 +24084,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -24230,19 +24096,19 @@
 				<port id="2" precision="FP32" names="/blocks.11/attn/Transpose_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1664" name="/blocks.11/attn/Constant" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353109" size="4" />
+			<data element_type="f32" shape="" offset="6889109" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.11/attn/Constant_output_0" />
 			</output>
 		</layer>
 		<layer id="1665" name="/blocks.11/attn/Constant_1" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="5353113" size="4" />
+			<data element_type="f32" shape="" offset="6889113" size="4" />
 			<output>
 				<port id="0" precision="FP32" names="/blocks.11/attn/Constant_1_output_0" />
 			</output>
@@ -24263,7 +24129,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -24272,13 +24138,13 @@
 				<port id="2" precision="FP32" names="/blocks.11/attn/Mul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1668" name="onnx::MatMul_1819" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="45903581" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="47439591" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -24288,7 +24154,7 @@
 			</output>
 		</layer>
 		<layer id="1669" name="Convert_1983" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -24297,7 +24163,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -24305,7 +24171,7 @@
 			</output>
 		</layer>
 		<layer id="1670" name="onnx::MatMul_1819/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -24313,32 +24179,32 @@
 			</output>
 		</layer>
 		<layer id="1671" name="Convert_1985" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1672" name="Subtract_1986" type="Subtract" version="opset1">
+		<layer id="1672" name="onnx::MatMul_1819/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -24346,7 +24212,7 @@
 			</output>
 		</layer>
 		<layer id="1673" name="onnx::MatMul_1819/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="46198493" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="47734503" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -24355,57 +24221,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1674" name="onnx::MatMul_1819/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1675" name="onnx::MatMul_1819/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1674" name="onnx::MatMul_1819/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1676" name="Constant_1991" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1675" name="Constant_1990" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1677" name="Reshape_1992" type="Reshape" version="opset1">
+		<layer id="1676" name="Reshape_1991" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -24415,7 +24264,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1677" name="onnx::MatMul_1819/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -24426,7 +24290,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -24437,13 +24301,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/key/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1679" name="/blocks.11/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.11/attn/Constant_3_output_0">
 					<dim>4</dim>
@@ -24455,7 +24319,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -24465,14 +24329,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/Reshape_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1681" name="Constant_1156" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5657245" size="32" />
+			<data element_type="i64" shape="4" offset="7193246" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -24483,7 +24347,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -24496,7 +24360,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -24507,7 +24371,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32" />
 			</input>
@@ -24516,7 +24380,7 @@
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -24526,22 +24390,22 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/MatMul_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -24551,16 +24415,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.11/attn/Cast_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -24570,16 +24434,16 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.11/attn/Softmax_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -24589,21 +24453,21 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.11/attn/Cast_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1688" name="onnx::Add_1821" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="46207709" size="3072" />
+			<data element_type="f32" shape="768" offset="47743719" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1821">
 					<dim>768</dim>
@@ -24611,7 +24475,7 @@
 			</output>
 		</layer>
 		<layer id="1689" name="onnx::MatMul_1822" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="46210781" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="47746791" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -24621,7 +24485,7 @@
 			</output>
 		</layer>
 		<layer id="1690" name="Convert_1972" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -24630,7 +24494,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -24638,7 +24502,7 @@
 			</output>
 		</layer>
 		<layer id="1691" name="onnx::MatMul_1822/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -24646,32 +24510,32 @@
 			</output>
 		</layer>
 		<layer id="1692" name="Convert_1974" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1693" name="Subtract_1975" type="Subtract" version="opset1">
+		<layer id="1693" name="onnx::MatMul_1822/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -24679,7 +24543,7 @@
 			</output>
 		</layer>
 		<layer id="1694" name="onnx::MatMul_1822/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="46505693" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="48041703" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -24688,57 +24552,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1695" name="onnx::MatMul_1822/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1696" name="onnx::MatMul_1822/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1695" name="onnx::MatMul_1822/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1697" name="Constant_1980" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1696" name="Constant_1979" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1698" name="Reshape_1981" type="Reshape" version="opset1">
+		<layer id="1697" name="Reshape_1980" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -24748,7 +24595,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1698" name="onnx::MatMul_1822/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -24759,7 +24621,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -24770,7 +24632,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/value/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -24783,20 +24645,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/value/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1701" name="/blocks.11/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353045" size="32" />
+			<data element_type="i64" shape="4" offset="6889045" size="32" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.11/attn/Constant_4_output_0">
 					<dim>4</dim>
@@ -24808,7 +24670,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -24818,14 +24680,14 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/Reshape_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1703" name="Constant_1161" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -24836,7 +24698,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -24848,7 +24710,7 @@
 				<port id="2" precision="FP32" names="/blocks.11/attn/Transpose_2_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
@@ -24859,13 +24721,13 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</input>
@@ -24873,13 +24735,13 @@
 				<port id="2" precision="FP32" names="/blocks.11/attn/MatMul_1_output_0">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1706" name="Constant_1168" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="5353077" size="32" />
+			<data element_type="i64" shape="4" offset="6889077" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -24891,7 +24753,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>12</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -24901,7 +24763,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/Transpose_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -24912,7 +24774,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -24924,7 +24786,7 @@
 			</output>
 		</layer>
 		<layer id="1709" name="/blocks.11/attn/Constant_6" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.11/attn/Constant_6_output_0">
 					<dim>1</dim>
@@ -24977,7 +24839,7 @@
 			</output>
 		</layer>
 		<layer id="1714" name="/blocks.11/attn/Constant_5" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042748" size="8" />
+			<data element_type="i64" shape="1" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.11/attn/Constant_5_output_0">
 					<dim>1</dim>
@@ -25009,7 +24871,7 @@
 			</output>
 		</layer>
 		<layer id="1716" name="/blocks.11/attn/Constant_8" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="5042740" size="8" />
+			<data element_type="i64" shape="1" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" names="/blocks.11/attn/Constant_8_output_0">
 					<dim>1</dim>
@@ -25037,7 +24899,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
@@ -25048,13 +24910,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/Reshape_3_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1719" name="onnx::MatMul_1840" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 768" offset="46514909" size="294912" />
+			<data element_type="u4" shape="6, 128, 768" offset="48050919" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -25064,7 +24926,7 @@
 			</output>
 		</layer>
 		<layer id="1720" name="Convert_2005" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -25073,7 +24935,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -25081,7 +24943,7 @@
 			</output>
 		</layer>
 		<layer id="1721" name="onnx::MatMul_1840/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="6879812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -25089,32 +24951,32 @@
 			</output>
 		</layer>
 		<layer id="1722" name="Convert_2007" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1723" name="Subtract_2008" type="Subtract" version="opset1">
+		<layer id="1723" name="onnx::MatMul_1840/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -25122,7 +24984,7 @@
 			</output>
 		</layer>
 		<layer id="1724" name="onnx::MatMul_1840/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 768" offset="46809821" size="9216" />
+			<data element_type="f16" shape="6, 1, 768" offset="48345831" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -25131,57 +24993,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1725" name="onnx::MatMul_1840/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1726" name="onnx::MatMul_1840/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1725" name="onnx::MatMul_1840/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1727" name="Constant_2013" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="5353029" size="16" />
+		<layer id="1726" name="Constant_2012" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="6889029" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1728" name="Reshape_2014" type="Reshape" version="opset1">
+		<layer id="1727" name="Reshape_2013" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>768</dim>
@@ -25191,7 +25036,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1728" name="onnx::MatMul_1840/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
@@ -25202,7 +25062,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -25213,7 +25073,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/out/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -25226,14 +25086,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/attn/out/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -25243,19 +25103,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -25265,20 +25125,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.11/mlp_ln/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1733" name="onnx::Add_1845" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="46819037" size="3072" />
+			<data element_type="f32" shape="768" offset="48355047" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1845">
 					<dim>768</dim>
@@ -25286,7 +25146,7 @@
 			</output>
 		</layer>
 		<layer id="1734" name="onnx::Add_1842" type="Const" version="opset1">
-			<data element_type="f32" shape="3072" offset="46822109" size="12288" />
+			<data element_type="f32" shape="3072" offset="48358119" size="12288" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_1842">
 					<dim>3072</dim>
@@ -25294,13 +25154,13 @@
 			</output>
 		</layer>
 		<layer id="1735" name="Constant_1194" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="1736" name="Constant_1193" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -25329,7 +25189,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -25339,13 +25199,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1740" name="blocks.11.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="46834397" size="3072" />
+			<data element_type="f32" shape="768" offset="48370407" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.11.mlp_ln.weight">
 					<dim>768</dim>
@@ -25357,7 +25217,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -25367,13 +25227,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1742" name="blocks.11.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="46837469" size="3072" />
+			<data element_type="f32" shape="768" offset="48373479" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.11.mlp_ln.bias">
 					<dim>768</dim>
@@ -25385,7 +25245,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -25395,7 +25255,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/mlp_ln/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -25405,20 +25265,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.11/mlp_ln/Cast_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1745" name="onnx::MatMul_1843" type="Const" version="opset1">
-			<data element_type="u4" shape="6, 128, 3072" offset="46840541" size="1179648" />
+			<data element_type="u4" shape="6, 128, 3072" offset="48376551" size="1179648" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -25428,7 +25288,7 @@
 			</output>
 		</layer>
 		<layer id="1746" name="Convert_2016" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>6</dim>
@@ -25437,7 +25297,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -25445,7 +25305,7 @@
 			</output>
 		</layer>
 		<layer id="1747" name="onnx::MatMul_1843/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="5343812" size="1" />
+			<data element_type="u4" shape="1" offset="7184029" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -25453,32 +25313,32 @@
 			</output>
 		</layer>
 		<layer id="1748" name="Convert_2018" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1749" name="Subtract_2019" type="Subtract" version="opset1">
+		<layer id="1749" name="onnx::MatMul_1843/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -25486,7 +25346,7 @@
 			</output>
 		</layer>
 		<layer id="1750" name="onnx::MatMul_1843/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="6, 1, 3072" offset="48020189" size="36864" />
+			<data element_type="f16" shape="6, 1, 3072" offset="49556199" size="36864" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>6</dim>
@@ -25495,57 +25355,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1751" name="onnx::MatMul_1843/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>6</dim>
-					<dim>1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1752" name="onnx::MatMul_1843/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1751" name="onnx::MatMul_1843/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>6</dim>
 					<dim>1</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1753" name="Constant_2024" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="7506621" size="16" />
+		<layer id="1752" name="Constant_2023" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="9042623" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1754" name="Reshape_2025" type="Reshape" version="opset1">
+		<layer id="1753" name="Reshape_2024" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>6</dim>
 					<dim>128</dim>
 					<dim>3072</dim>
@@ -25555,7 +25398,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1754" name="onnx::MatMul_1843/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
@@ -25566,7 +25424,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -25577,7 +25435,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/mlp/mlp.0/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -25590,14 +25448,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/mlp/mlp.0/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -25613,7 +25471,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -25621,7 +25479,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/mlp/mlp.1/Div_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -25630,14 +25488,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/blocks.11/mlp/mlp.1/Erf_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -25653,7 +25511,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -25661,7 +25519,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/mlp/mlp.1/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -25671,19 +25529,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/mlp/mlp.1/Mul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
@@ -25699,7 +25557,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32" />
@@ -25707,13 +25565,13 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/mlp/mlp.1/Mul_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1765" name="onnx::MatMul_1846" type="Const" version="opset1">
-			<data element_type="u8" shape="3072, 768" offset="48057053" size="2359296" />
+			<data element_type="u8" shape="3072, 768" offset="49593063" size="2359296" />
 			<output>
 				<port id="0" precision="U8">
 					<dim>3072</dim>
@@ -25722,7 +25580,7 @@
 			</output>
 		</layer>
 		<layer id="1766" name="Convert_2028" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>3072</dim>
@@ -25730,14 +25588,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1767" name="onnx::MatMul_1846/zero_point" type="Const" version="opset1">
-			<data element_type="u8" shape="1, 768" offset="50416349" size="768" />
+			<data element_type="u8" shape="1, 768" offset="51952359" size="768" />
 			<output>
 				<port id="0" precision="U8">
 					<dim>1</dim>
@@ -25746,7 +25604,7 @@
 			</output>
 		</layer>
 		<layer id="1768" name="Convert_2031" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>1</dim>
@@ -25754,33 +25612,33 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1769" name="Subtract_2032" type="Subtract" version="opset1">
+		<layer id="1769" name="onnx::MatMul_1846/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1770" name="onnx::MatMul_1846/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="1, 768" offset="50417117" size="1536" />
+			<data element_type="f16" shape="1, 768" offset="51953127" size="1536" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>1</dim>
@@ -25788,35 +25646,35 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1771" name="onnx::MatMul_1846/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1771" name="onnx::MatMul_1846/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1772" name="onnx::MatMul_1846/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1772" name="onnx::MatMul_1846/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>768</dim>
-				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
@@ -25827,7 +25685,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>3072</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -25838,7 +25696,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/mlp/mlp.2/MatMul_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -25851,14 +25709,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/mlp/mlp.2/Add_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -25868,19 +25726,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/blocks.11/Add_1_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -25890,26 +25748,26 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/ln_post/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1777" name="Constant_1216" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042740" size="8" />
+			<data element_type="i64" shape="" offset="6578740" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="1778" name="Constant_1215" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="5042748" size="8" />
+			<data element_type="i64" shape="" offset="6578748" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -25938,7 +25796,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -25948,13 +25806,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1782" name="ln_post.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="50418653" size="3072" />
+			<data element_type="f32" shape="768" offset="51954663" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="ln_post.weight">
 					<dim>768</dim>
@@ -25966,7 +25824,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -25976,13 +25834,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="1784" name="ln_post.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="768" offset="50421725" size="3072" />
+			<data element_type="f32" shape="768" offset="51957735" size="3072" />
 			<output>
 				<port id="0" precision="FP32" names="ln_post.bias">
 					<dim>768</dim>
@@ -25994,7 +25852,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -26004,7 +25862,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/ln_post/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -26014,14 +25872,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="output_features">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</output>
@@ -26030,7 +25888,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>768</dim>
 				</port>
 			</input>
@@ -26042,10 +25900,10 @@
 		<edge from-layer="2" from-port="1" to-layer="5" to-port="0" />
 		<edge from-layer="3" from-port="0" to-layer="4" to-port="0" />
 		<edge from-layer="4" from-port="1" to-layer="5" to-port="1" />
-		<edge from-layer="5" from-port="2" to-layer="8" to-port="0" />
-		<edge from-layer="6" from-port="0" to-layer="7" to-port="0" />
-		<edge from-layer="7" from-port="1" to-layer="8" to-port="1" />
-		<edge from-layer="8" from-port="2" to-layer="9" to-port="1" />
+		<edge from-layer="5" from-port="2" to-layer="7" to-port="0" />
+		<edge from-layer="6" from-port="0" to-layer="7" to-port="1" />
+		<edge from-layer="7" from-port="2" to-layer="8" to-port="0" />
+		<edge from-layer="8" from-port="1" to-layer="9" to-port="1" />
 		<edge from-layer="9" from-port="2" to-layer="13" to-port="0" />
 		<edge from-layer="9" from-port="2" to-layer="20" to-port="0" />
 		<edge from-layer="10" from-port="0" to-layer="12" to-port="0" />
@@ -26074,10 +25932,10 @@
 		<edge from-layer="30" from-port="1" to-layer="33" to-port="0" />
 		<edge from-layer="31" from-port="0" to-layer="32" to-port="0" />
 		<edge from-layer="32" from-port="1" to-layer="33" to-port="1" />
-		<edge from-layer="33" from-port="2" to-layer="36" to-port="0" />
-		<edge from-layer="34" from-port="0" to-layer="35" to-port="0" />
-		<edge from-layer="35" from-port="1" to-layer="36" to-port="1" />
-		<edge from-layer="36" from-port="2" to-layer="37" to-port="1" />
+		<edge from-layer="33" from-port="2" to-layer="35" to-port="0" />
+		<edge from-layer="34" from-port="0" to-layer="35" to-port="1" />
+		<edge from-layer="35" from-port="2" to-layer="36" to-port="0" />
+		<edge from-layer="36" from-port="1" to-layer="37" to-port="1" />
 		<edge from-layer="37" from-port="2" to-layer="48" to-port="0" />
 		<edge from-layer="37" from-port="2" to-layer="41" to-port="0" />
 		<edge from-layer="38" from-port="0" to-layer="47" to-port="0" />
@@ -26127,12 +25985,12 @@
 		<edge from-layer="76" from-port="1" to-layer="79" to-port="0" />
 		<edge from-layer="77" from-port="0" to-layer="78" to-port="0" />
 		<edge from-layer="78" from-port="1" to-layer="79" to-port="1" />
-		<edge from-layer="79" from-port="2" to-layer="82" to-port="0" />
-		<edge from-layer="80" from-port="0" to-layer="81" to-port="0" />
-		<edge from-layer="81" from-port="1" to-layer="82" to-port="1" />
-		<edge from-layer="82" from-port="2" to-layer="84" to-port="0" />
-		<edge from-layer="83" from-port="0" to-layer="84" to-port="1" />
-		<edge from-layer="84" from-port="2" to-layer="85" to-port="1" />
+		<edge from-layer="79" from-port="2" to-layer="81" to-port="0" />
+		<edge from-layer="80" from-port="0" to-layer="81" to-port="1" />
+		<edge from-layer="81" from-port="2" to-layer="83" to-port="0" />
+		<edge from-layer="82" from-port="0" to-layer="83" to-port="1" />
+		<edge from-layer="83" from-port="2" to-layer="84" to-port="0" />
+		<edge from-layer="84" from-port="1" to-layer="85" to-port="1" />
 		<edge from-layer="85" from-port="2" to-layer="86" to-port="1" />
 		<edge from-layer="86" from-port="2" to-layer="88" to-port="0" />
 		<edge from-layer="87" from-port="0" to-layer="88" to-port="1" />
@@ -26148,12 +26006,12 @@
 		<edge from-layer="96" from-port="1" to-layer="99" to-port="0" />
 		<edge from-layer="97" from-port="0" to-layer="98" to-port="0" />
 		<edge from-layer="98" from-port="1" to-layer="99" to-port="1" />
-		<edge from-layer="99" from-port="2" to-layer="102" to-port="0" />
-		<edge from-layer="100" from-port="0" to-layer="101" to-port="0" />
-		<edge from-layer="101" from-port="1" to-layer="102" to-port="1" />
-		<edge from-layer="102" from-port="2" to-layer="104" to-port="0" />
-		<edge from-layer="103" from-port="0" to-layer="104" to-port="1" />
-		<edge from-layer="104" from-port="2" to-layer="105" to-port="1" />
+		<edge from-layer="99" from-port="2" to-layer="101" to-port="0" />
+		<edge from-layer="100" from-port="0" to-layer="101" to-port="1" />
+		<edge from-layer="101" from-port="2" to-layer="103" to-port="0" />
+		<edge from-layer="102" from-port="0" to-layer="103" to-port="1" />
+		<edge from-layer="103" from-port="2" to-layer="104" to-port="0" />
+		<edge from-layer="104" from-port="1" to-layer="105" to-port="1" />
 		<edge from-layer="105" from-port="2" to-layer="107" to-port="0" />
 		<edge from-layer="106" from-port="0" to-layer="107" to-port="1" />
 		<edge from-layer="107" from-port="2" to-layer="109" to-port="0" />
@@ -26169,12 +26027,12 @@
 		<edge from-layer="117" from-port="1" to-layer="120" to-port="0" />
 		<edge from-layer="118" from-port="0" to-layer="119" to-port="0" />
 		<edge from-layer="119" from-port="1" to-layer="120" to-port="1" />
-		<edge from-layer="120" from-port="2" to-layer="123" to-port="0" />
-		<edge from-layer="121" from-port="0" to-layer="122" to-port="0" />
-		<edge from-layer="122" from-port="1" to-layer="123" to-port="1" />
-		<edge from-layer="123" from-port="2" to-layer="125" to-port="0" />
-		<edge from-layer="124" from-port="0" to-layer="125" to-port="1" />
-		<edge from-layer="125" from-port="2" to-layer="126" to-port="1" />
+		<edge from-layer="120" from-port="2" to-layer="122" to-port="0" />
+		<edge from-layer="121" from-port="0" to-layer="122" to-port="1" />
+		<edge from-layer="122" from-port="2" to-layer="124" to-port="0" />
+		<edge from-layer="123" from-port="0" to-layer="124" to-port="1" />
+		<edge from-layer="124" from-port="2" to-layer="125" to-port="0" />
+		<edge from-layer="125" from-port="1" to-layer="126" to-port="1" />
 		<edge from-layer="126" from-port="2" to-layer="127" to-port="1" />
 		<edge from-layer="127" from-port="2" to-layer="129" to-port="0" />
 		<edge from-layer="128" from-port="0" to-layer="129" to-port="1" />
@@ -26201,12 +26059,12 @@
 		<edge from-layer="147" from-port="1" to-layer="150" to-port="0" />
 		<edge from-layer="148" from-port="0" to-layer="149" to-port="0" />
 		<edge from-layer="149" from-port="1" to-layer="150" to-port="1" />
-		<edge from-layer="150" from-port="2" to-layer="153" to-port="0" />
-		<edge from-layer="151" from-port="0" to-layer="152" to-port="0" />
-		<edge from-layer="152" from-port="1" to-layer="153" to-port="1" />
-		<edge from-layer="153" from-port="2" to-layer="155" to-port="0" />
-		<edge from-layer="154" from-port="0" to-layer="155" to-port="1" />
-		<edge from-layer="155" from-port="2" to-layer="156" to-port="1" />
+		<edge from-layer="150" from-port="2" to-layer="152" to-port="0" />
+		<edge from-layer="151" from-port="0" to-layer="152" to-port="1" />
+		<edge from-layer="152" from-port="2" to-layer="154" to-port="0" />
+		<edge from-layer="153" from-port="0" to-layer="154" to-port="1" />
+		<edge from-layer="154" from-port="2" to-layer="155" to-port="0" />
+		<edge from-layer="155" from-port="1" to-layer="156" to-port="1" />
 		<edge from-layer="156" from-port="2" to-layer="157" to-port="1" />
 		<edge from-layer="157" from-port="2" to-layer="158" to-port="1" />
 		<edge from-layer="158" from-port="2" to-layer="159" to-port="0" />
@@ -26228,12 +26086,12 @@
 		<edge from-layer="173" from-port="1" to-layer="176" to-port="0" />
 		<edge from-layer="174" from-port="0" to-layer="175" to-port="0" />
 		<edge from-layer="175" from-port="1" to-layer="176" to-port="1" />
-		<edge from-layer="176" from-port="2" to-layer="179" to-port="0" />
-		<edge from-layer="177" from-port="0" to-layer="178" to-port="0" />
-		<edge from-layer="178" from-port="1" to-layer="179" to-port="1" />
-		<edge from-layer="179" from-port="2" to-layer="181" to-port="0" />
-		<edge from-layer="180" from-port="0" to-layer="181" to-port="1" />
-		<edge from-layer="181" from-port="2" to-layer="182" to-port="1" />
+		<edge from-layer="176" from-port="2" to-layer="178" to-port="0" />
+		<edge from-layer="177" from-port="0" to-layer="178" to-port="1" />
+		<edge from-layer="178" from-port="2" to-layer="180" to-port="0" />
+		<edge from-layer="179" from-port="0" to-layer="180" to-port="1" />
+		<edge from-layer="180" from-port="2" to-layer="181" to-port="0" />
+		<edge from-layer="181" from-port="1" to-layer="182" to-port="1" />
 		<edge from-layer="182" from-port="2" to-layer="183" to-port="1" />
 		<edge from-layer="183" from-port="2" to-layer="185" to-port="0" />
 		<edge from-layer="183" from-port="2" to-layer="189" to-port="0" />
@@ -26249,12 +26107,12 @@
 		<edge from-layer="193" from-port="1" to-layer="196" to-port="0" />
 		<edge from-layer="194" from-port="0" to-layer="195" to-port="0" />
 		<edge from-layer="195" from-port="1" to-layer="196" to-port="1" />
-		<edge from-layer="196" from-port="2" to-layer="199" to-port="0" />
-		<edge from-layer="197" from-port="0" to-layer="198" to-port="0" />
-		<edge from-layer="198" from-port="1" to-layer="199" to-port="1" />
-		<edge from-layer="199" from-port="2" to-layer="201" to-port="0" />
-		<edge from-layer="200" from-port="0" to-layer="201" to-port="1" />
-		<edge from-layer="201" from-port="2" to-layer="202" to-port="1" />
+		<edge from-layer="196" from-port="2" to-layer="198" to-port="0" />
+		<edge from-layer="197" from-port="0" to-layer="198" to-port="1" />
+		<edge from-layer="198" from-port="2" to-layer="200" to-port="0" />
+		<edge from-layer="199" from-port="0" to-layer="200" to-port="1" />
+		<edge from-layer="200" from-port="2" to-layer="201" to-port="0" />
+		<edge from-layer="201" from-port="1" to-layer="202" to-port="1" />
 		<edge from-layer="202" from-port="2" to-layer="203" to-port="1" />
 		<edge from-layer="203" from-port="2" to-layer="204" to-port="1" />
 		<edge from-layer="204" from-port="2" to-layer="205" to-port="0" />
@@ -26278,12 +26136,12 @@
 		<edge from-layer="219" from-port="1" to-layer="222" to-port="0" />
 		<edge from-layer="220" from-port="0" to-layer="221" to-port="0" />
 		<edge from-layer="221" from-port="1" to-layer="222" to-port="1" />
-		<edge from-layer="222" from-port="2" to-layer="225" to-port="0" />
-		<edge from-layer="223" from-port="0" to-layer="224" to-port="0" />
-		<edge from-layer="224" from-port="1" to-layer="225" to-port="1" />
-		<edge from-layer="225" from-port="2" to-layer="227" to-port="0" />
-		<edge from-layer="226" from-port="0" to-layer="227" to-port="1" />
-		<edge from-layer="227" from-port="2" to-layer="228" to-port="1" />
+		<edge from-layer="222" from-port="2" to-layer="224" to-port="0" />
+		<edge from-layer="223" from-port="0" to-layer="224" to-port="1" />
+		<edge from-layer="224" from-port="2" to-layer="226" to-port="0" />
+		<edge from-layer="225" from-port="0" to-layer="226" to-port="1" />
+		<edge from-layer="226" from-port="2" to-layer="227" to-port="0" />
+		<edge from-layer="227" from-port="1" to-layer="228" to-port="1" />
 		<edge from-layer="228" from-port="2" to-layer="229" to-port="1" />
 		<edge from-layer="229" from-port="2" to-layer="231" to-port="0" />
 		<edge from-layer="230" from-port="0" to-layer="231" to-port="1" />
@@ -26299,12 +26157,12 @@
 		<edge from-layer="239" from-port="1" to-layer="242" to-port="0" />
 		<edge from-layer="240" from-port="0" to-layer="241" to-port="0" />
 		<edge from-layer="241" from-port="1" to-layer="242" to-port="1" />
-		<edge from-layer="242" from-port="2" to-layer="245" to-port="0" />
-		<edge from-layer="243" from-port="0" to-layer="244" to-port="0" />
-		<edge from-layer="244" from-port="1" to-layer="245" to-port="1" />
-		<edge from-layer="245" from-port="2" to-layer="247" to-port="0" />
-		<edge from-layer="246" from-port="0" to-layer="247" to-port="1" />
-		<edge from-layer="247" from-port="2" to-layer="248" to-port="1" />
+		<edge from-layer="242" from-port="2" to-layer="244" to-port="0" />
+		<edge from-layer="243" from-port="0" to-layer="244" to-port="1" />
+		<edge from-layer="244" from-port="2" to-layer="246" to-port="0" />
+		<edge from-layer="245" from-port="0" to-layer="246" to-port="1" />
+		<edge from-layer="246" from-port="2" to-layer="247" to-port="0" />
+		<edge from-layer="247" from-port="1" to-layer="248" to-port="1" />
 		<edge from-layer="248" from-port="2" to-layer="250" to-port="0" />
 		<edge from-layer="249" from-port="0" to-layer="250" to-port="1" />
 		<edge from-layer="250" from-port="2" to-layer="252" to-port="0" />
@@ -26320,12 +26178,12 @@
 		<edge from-layer="260" from-port="1" to-layer="263" to-port="0" />
 		<edge from-layer="261" from-port="0" to-layer="262" to-port="0" />
 		<edge from-layer="262" from-port="1" to-layer="263" to-port="1" />
-		<edge from-layer="263" from-port="2" to-layer="266" to-port="0" />
-		<edge from-layer="264" from-port="0" to-layer="265" to-port="0" />
-		<edge from-layer="265" from-port="1" to-layer="266" to-port="1" />
-		<edge from-layer="266" from-port="2" to-layer="268" to-port="0" />
-		<edge from-layer="267" from-port="0" to-layer="268" to-port="1" />
-		<edge from-layer="268" from-port="2" to-layer="269" to-port="1" />
+		<edge from-layer="263" from-port="2" to-layer="265" to-port="0" />
+		<edge from-layer="264" from-port="0" to-layer="265" to-port="1" />
+		<edge from-layer="265" from-port="2" to-layer="267" to-port="0" />
+		<edge from-layer="266" from-port="0" to-layer="267" to-port="1" />
+		<edge from-layer="267" from-port="2" to-layer="268" to-port="0" />
+		<edge from-layer="268" from-port="1" to-layer="269" to-port="1" />
 		<edge from-layer="269" from-port="2" to-layer="270" to-port="1" />
 		<edge from-layer="270" from-port="2" to-layer="272" to-port="0" />
 		<edge from-layer="271" from-port="0" to-layer="272" to-port="1" />
@@ -26352,12 +26210,12 @@
 		<edge from-layer="290" from-port="1" to-layer="293" to-port="0" />
 		<edge from-layer="291" from-port="0" to-layer="292" to-port="0" />
 		<edge from-layer="292" from-port="1" to-layer="293" to-port="1" />
-		<edge from-layer="293" from-port="2" to-layer="296" to-port="0" />
-		<edge from-layer="294" from-port="0" to-layer="295" to-port="0" />
-		<edge from-layer="295" from-port="1" to-layer="296" to-port="1" />
-		<edge from-layer="296" from-port="2" to-layer="298" to-port="0" />
-		<edge from-layer="297" from-port="0" to-layer="298" to-port="1" />
-		<edge from-layer="298" from-port="2" to-layer="299" to-port="1" />
+		<edge from-layer="293" from-port="2" to-layer="295" to-port="0" />
+		<edge from-layer="294" from-port="0" to-layer="295" to-port="1" />
+		<edge from-layer="295" from-port="2" to-layer="297" to-port="0" />
+		<edge from-layer="296" from-port="0" to-layer="297" to-port="1" />
+		<edge from-layer="297" from-port="2" to-layer="298" to-port="0" />
+		<edge from-layer="298" from-port="1" to-layer="299" to-port="1" />
 		<edge from-layer="299" from-port="2" to-layer="300" to-port="1" />
 		<edge from-layer="300" from-port="2" to-layer="301" to-port="1" />
 		<edge from-layer="301" from-port="2" to-layer="302" to-port="0" />
@@ -26379,12 +26237,12 @@
 		<edge from-layer="316" from-port="1" to-layer="319" to-port="0" />
 		<edge from-layer="317" from-port="0" to-layer="318" to-port="0" />
 		<edge from-layer="318" from-port="1" to-layer="319" to-port="1" />
-		<edge from-layer="319" from-port="2" to-layer="322" to-port="0" />
-		<edge from-layer="320" from-port="0" to-layer="321" to-port="0" />
-		<edge from-layer="321" from-port="1" to-layer="322" to-port="1" />
-		<edge from-layer="322" from-port="2" to-layer="324" to-port="0" />
-		<edge from-layer="323" from-port="0" to-layer="324" to-port="1" />
-		<edge from-layer="324" from-port="2" to-layer="325" to-port="1" />
+		<edge from-layer="319" from-port="2" to-layer="321" to-port="0" />
+		<edge from-layer="320" from-port="0" to-layer="321" to-port="1" />
+		<edge from-layer="321" from-port="2" to-layer="323" to-port="0" />
+		<edge from-layer="322" from-port="0" to-layer="323" to-port="1" />
+		<edge from-layer="323" from-port="2" to-layer="324" to-port="0" />
+		<edge from-layer="324" from-port="1" to-layer="325" to-port="1" />
 		<edge from-layer="325" from-port="2" to-layer="326" to-port="1" />
 		<edge from-layer="326" from-port="2" to-layer="328" to-port="0" />
 		<edge from-layer="326" from-port="2" to-layer="332" to-port="0" />
@@ -26400,12 +26258,12 @@
 		<edge from-layer="336" from-port="1" to-layer="339" to-port="0" />
 		<edge from-layer="337" from-port="0" to-layer="338" to-port="0" />
 		<edge from-layer="338" from-port="1" to-layer="339" to-port="1" />
-		<edge from-layer="339" from-port="2" to-layer="342" to-port="0" />
-		<edge from-layer="340" from-port="0" to-layer="341" to-port="0" />
-		<edge from-layer="341" from-port="1" to-layer="342" to-port="1" />
-		<edge from-layer="342" from-port="2" to-layer="344" to-port="0" />
-		<edge from-layer="343" from-port="0" to-layer="344" to-port="1" />
-		<edge from-layer="344" from-port="2" to-layer="345" to-port="1" />
+		<edge from-layer="339" from-port="2" to-layer="341" to-port="0" />
+		<edge from-layer="340" from-port="0" to-layer="341" to-port="1" />
+		<edge from-layer="341" from-port="2" to-layer="343" to-port="0" />
+		<edge from-layer="342" from-port="0" to-layer="343" to-port="1" />
+		<edge from-layer="343" from-port="2" to-layer="344" to-port="0" />
+		<edge from-layer="344" from-port="1" to-layer="345" to-port="1" />
 		<edge from-layer="345" from-port="2" to-layer="346" to-port="1" />
 		<edge from-layer="346" from-port="2" to-layer="347" to-port="1" />
 		<edge from-layer="347" from-port="2" to-layer="348" to-port="0" />
@@ -26429,12 +26287,12 @@
 		<edge from-layer="362" from-port="1" to-layer="365" to-port="0" />
 		<edge from-layer="363" from-port="0" to-layer="364" to-port="0" />
 		<edge from-layer="364" from-port="1" to-layer="365" to-port="1" />
-		<edge from-layer="365" from-port="2" to-layer="368" to-port="0" />
-		<edge from-layer="366" from-port="0" to-layer="367" to-port="0" />
-		<edge from-layer="367" from-port="1" to-layer="368" to-port="1" />
-		<edge from-layer="368" from-port="2" to-layer="370" to-port="0" />
-		<edge from-layer="369" from-port="0" to-layer="370" to-port="1" />
-		<edge from-layer="370" from-port="2" to-layer="371" to-port="1" />
+		<edge from-layer="365" from-port="2" to-layer="367" to-port="0" />
+		<edge from-layer="366" from-port="0" to-layer="367" to-port="1" />
+		<edge from-layer="367" from-port="2" to-layer="369" to-port="0" />
+		<edge from-layer="368" from-port="0" to-layer="369" to-port="1" />
+		<edge from-layer="369" from-port="2" to-layer="370" to-port="0" />
+		<edge from-layer="370" from-port="1" to-layer="371" to-port="1" />
 		<edge from-layer="371" from-port="2" to-layer="372" to-port="1" />
 		<edge from-layer="372" from-port="2" to-layer="374" to-port="0" />
 		<edge from-layer="373" from-port="0" to-layer="374" to-port="1" />
@@ -26450,12 +26308,12 @@
 		<edge from-layer="382" from-port="1" to-layer="385" to-port="0" />
 		<edge from-layer="383" from-port="0" to-layer="384" to-port="0" />
 		<edge from-layer="384" from-port="1" to-layer="385" to-port="1" />
-		<edge from-layer="385" from-port="2" to-layer="388" to-port="0" />
-		<edge from-layer="386" from-port="0" to-layer="387" to-port="0" />
-		<edge from-layer="387" from-port="1" to-layer="388" to-port="1" />
-		<edge from-layer="388" from-port="2" to-layer="390" to-port="0" />
-		<edge from-layer="389" from-port="0" to-layer="390" to-port="1" />
-		<edge from-layer="390" from-port="2" to-layer="391" to-port="1" />
+		<edge from-layer="385" from-port="2" to-layer="387" to-port="0" />
+		<edge from-layer="386" from-port="0" to-layer="387" to-port="1" />
+		<edge from-layer="387" from-port="2" to-layer="389" to-port="0" />
+		<edge from-layer="388" from-port="0" to-layer="389" to-port="1" />
+		<edge from-layer="389" from-port="2" to-layer="390" to-port="0" />
+		<edge from-layer="390" from-port="1" to-layer="391" to-port="1" />
 		<edge from-layer="391" from-port="2" to-layer="393" to-port="0" />
 		<edge from-layer="392" from-port="0" to-layer="393" to-port="1" />
 		<edge from-layer="393" from-port="2" to-layer="395" to-port="0" />
@@ -26471,12 +26329,12 @@
 		<edge from-layer="403" from-port="1" to-layer="406" to-port="0" />
 		<edge from-layer="404" from-port="0" to-layer="405" to-port="0" />
 		<edge from-layer="405" from-port="1" to-layer="406" to-port="1" />
-		<edge from-layer="406" from-port="2" to-layer="409" to-port="0" />
-		<edge from-layer="407" from-port="0" to-layer="408" to-port="0" />
-		<edge from-layer="408" from-port="1" to-layer="409" to-port="1" />
-		<edge from-layer="409" from-port="2" to-layer="411" to-port="0" />
-		<edge from-layer="410" from-port="0" to-layer="411" to-port="1" />
-		<edge from-layer="411" from-port="2" to-layer="412" to-port="1" />
+		<edge from-layer="406" from-port="2" to-layer="408" to-port="0" />
+		<edge from-layer="407" from-port="0" to-layer="408" to-port="1" />
+		<edge from-layer="408" from-port="2" to-layer="410" to-port="0" />
+		<edge from-layer="409" from-port="0" to-layer="410" to-port="1" />
+		<edge from-layer="410" from-port="2" to-layer="411" to-port="0" />
+		<edge from-layer="411" from-port="1" to-layer="412" to-port="1" />
 		<edge from-layer="412" from-port="2" to-layer="413" to-port="1" />
 		<edge from-layer="413" from-port="2" to-layer="415" to-port="0" />
 		<edge from-layer="414" from-port="0" to-layer="415" to-port="1" />
@@ -26503,12 +26361,12 @@
 		<edge from-layer="433" from-port="1" to-layer="436" to-port="0" />
 		<edge from-layer="434" from-port="0" to-layer="435" to-port="0" />
 		<edge from-layer="435" from-port="1" to-layer="436" to-port="1" />
-		<edge from-layer="436" from-port="2" to-layer="439" to-port="0" />
-		<edge from-layer="437" from-port="0" to-layer="438" to-port="0" />
-		<edge from-layer="438" from-port="1" to-layer="439" to-port="1" />
-		<edge from-layer="439" from-port="2" to-layer="441" to-port="0" />
-		<edge from-layer="440" from-port="0" to-layer="441" to-port="1" />
-		<edge from-layer="441" from-port="2" to-layer="442" to-port="1" />
+		<edge from-layer="436" from-port="2" to-layer="438" to-port="0" />
+		<edge from-layer="437" from-port="0" to-layer="438" to-port="1" />
+		<edge from-layer="438" from-port="2" to-layer="440" to-port="0" />
+		<edge from-layer="439" from-port="0" to-layer="440" to-port="1" />
+		<edge from-layer="440" from-port="2" to-layer="441" to-port="0" />
+		<edge from-layer="441" from-port="1" to-layer="442" to-port="1" />
 		<edge from-layer="442" from-port="2" to-layer="443" to-port="1" />
 		<edge from-layer="443" from-port="2" to-layer="444" to-port="1" />
 		<edge from-layer="444" from-port="2" to-layer="445" to-port="0" />
@@ -26530,12 +26388,12 @@
 		<edge from-layer="459" from-port="1" to-layer="462" to-port="0" />
 		<edge from-layer="460" from-port="0" to-layer="461" to-port="0" />
 		<edge from-layer="461" from-port="1" to-layer="462" to-port="1" />
-		<edge from-layer="462" from-port="2" to-layer="465" to-port="0" />
-		<edge from-layer="463" from-port="0" to-layer="464" to-port="0" />
-		<edge from-layer="464" from-port="1" to-layer="465" to-port="1" />
-		<edge from-layer="465" from-port="2" to-layer="467" to-port="0" />
-		<edge from-layer="466" from-port="0" to-layer="467" to-port="1" />
-		<edge from-layer="467" from-port="2" to-layer="468" to-port="1" />
+		<edge from-layer="462" from-port="2" to-layer="464" to-port="0" />
+		<edge from-layer="463" from-port="0" to-layer="464" to-port="1" />
+		<edge from-layer="464" from-port="2" to-layer="466" to-port="0" />
+		<edge from-layer="465" from-port="0" to-layer="466" to-port="1" />
+		<edge from-layer="466" from-port="2" to-layer="467" to-port="0" />
+		<edge from-layer="467" from-port="1" to-layer="468" to-port="1" />
 		<edge from-layer="468" from-port="2" to-layer="469" to-port="1" />
 		<edge from-layer="469" from-port="2" to-layer="471" to-port="0" />
 		<edge from-layer="469" from-port="2" to-layer="475" to-port="0" />
@@ -26551,12 +26409,12 @@
 		<edge from-layer="479" from-port="1" to-layer="482" to-port="0" />
 		<edge from-layer="480" from-port="0" to-layer="481" to-port="0" />
 		<edge from-layer="481" from-port="1" to-layer="482" to-port="1" />
-		<edge from-layer="482" from-port="2" to-layer="485" to-port="0" />
-		<edge from-layer="483" from-port="0" to-layer="484" to-port="0" />
-		<edge from-layer="484" from-port="1" to-layer="485" to-port="1" />
-		<edge from-layer="485" from-port="2" to-layer="487" to-port="0" />
-		<edge from-layer="486" from-port="0" to-layer="487" to-port="1" />
-		<edge from-layer="487" from-port="2" to-layer="488" to-port="1" />
+		<edge from-layer="482" from-port="2" to-layer="484" to-port="0" />
+		<edge from-layer="483" from-port="0" to-layer="484" to-port="1" />
+		<edge from-layer="484" from-port="2" to-layer="486" to-port="0" />
+		<edge from-layer="485" from-port="0" to-layer="486" to-port="1" />
+		<edge from-layer="486" from-port="2" to-layer="487" to-port="0" />
+		<edge from-layer="487" from-port="1" to-layer="488" to-port="1" />
 		<edge from-layer="488" from-port="2" to-layer="489" to-port="1" />
 		<edge from-layer="489" from-port="2" to-layer="490" to-port="1" />
 		<edge from-layer="490" from-port="2" to-layer="491" to-port="0" />
@@ -26580,12 +26438,12 @@
 		<edge from-layer="505" from-port="1" to-layer="508" to-port="0" />
 		<edge from-layer="506" from-port="0" to-layer="507" to-port="0" />
 		<edge from-layer="507" from-port="1" to-layer="508" to-port="1" />
-		<edge from-layer="508" from-port="2" to-layer="511" to-port="0" />
-		<edge from-layer="509" from-port="0" to-layer="510" to-port="0" />
-		<edge from-layer="510" from-port="1" to-layer="511" to-port="1" />
-		<edge from-layer="511" from-port="2" to-layer="513" to-port="0" />
-		<edge from-layer="512" from-port="0" to-layer="513" to-port="1" />
-		<edge from-layer="513" from-port="2" to-layer="514" to-port="1" />
+		<edge from-layer="508" from-port="2" to-layer="510" to-port="0" />
+		<edge from-layer="509" from-port="0" to-layer="510" to-port="1" />
+		<edge from-layer="510" from-port="2" to-layer="512" to-port="0" />
+		<edge from-layer="511" from-port="0" to-layer="512" to-port="1" />
+		<edge from-layer="512" from-port="2" to-layer="513" to-port="0" />
+		<edge from-layer="513" from-port="1" to-layer="514" to-port="1" />
 		<edge from-layer="514" from-port="2" to-layer="515" to-port="1" />
 		<edge from-layer="515" from-port="2" to-layer="517" to-port="0" />
 		<edge from-layer="516" from-port="0" to-layer="517" to-port="1" />
@@ -26601,12 +26459,12 @@
 		<edge from-layer="525" from-port="1" to-layer="528" to-port="0" />
 		<edge from-layer="526" from-port="0" to-layer="527" to-port="0" />
 		<edge from-layer="527" from-port="1" to-layer="528" to-port="1" />
-		<edge from-layer="528" from-port="2" to-layer="531" to-port="0" />
-		<edge from-layer="529" from-port="0" to-layer="530" to-port="0" />
-		<edge from-layer="530" from-port="1" to-layer="531" to-port="1" />
-		<edge from-layer="531" from-port="2" to-layer="533" to-port="0" />
-		<edge from-layer="532" from-port="0" to-layer="533" to-port="1" />
-		<edge from-layer="533" from-port="2" to-layer="534" to-port="1" />
+		<edge from-layer="528" from-port="2" to-layer="530" to-port="0" />
+		<edge from-layer="529" from-port="0" to-layer="530" to-port="1" />
+		<edge from-layer="530" from-port="2" to-layer="532" to-port="0" />
+		<edge from-layer="531" from-port="0" to-layer="532" to-port="1" />
+		<edge from-layer="532" from-port="2" to-layer="533" to-port="0" />
+		<edge from-layer="533" from-port="1" to-layer="534" to-port="1" />
 		<edge from-layer="534" from-port="2" to-layer="536" to-port="0" />
 		<edge from-layer="535" from-port="0" to-layer="536" to-port="1" />
 		<edge from-layer="536" from-port="2" to-layer="538" to-port="0" />
@@ -26622,12 +26480,12 @@
 		<edge from-layer="546" from-port="1" to-layer="549" to-port="0" />
 		<edge from-layer="547" from-port="0" to-layer="548" to-port="0" />
 		<edge from-layer="548" from-port="1" to-layer="549" to-port="1" />
-		<edge from-layer="549" from-port="2" to-layer="552" to-port="0" />
-		<edge from-layer="550" from-port="0" to-layer="551" to-port="0" />
-		<edge from-layer="551" from-port="1" to-layer="552" to-port="1" />
-		<edge from-layer="552" from-port="2" to-layer="554" to-port="0" />
-		<edge from-layer="553" from-port="0" to-layer="554" to-port="1" />
-		<edge from-layer="554" from-port="2" to-layer="555" to-port="1" />
+		<edge from-layer="549" from-port="2" to-layer="551" to-port="0" />
+		<edge from-layer="550" from-port="0" to-layer="551" to-port="1" />
+		<edge from-layer="551" from-port="2" to-layer="553" to-port="0" />
+		<edge from-layer="552" from-port="0" to-layer="553" to-port="1" />
+		<edge from-layer="553" from-port="2" to-layer="554" to-port="0" />
+		<edge from-layer="554" from-port="1" to-layer="555" to-port="1" />
 		<edge from-layer="555" from-port="2" to-layer="556" to-port="1" />
 		<edge from-layer="556" from-port="2" to-layer="558" to-port="0" />
 		<edge from-layer="557" from-port="0" to-layer="558" to-port="1" />
@@ -26654,12 +26512,12 @@
 		<edge from-layer="576" from-port="1" to-layer="579" to-port="0" />
 		<edge from-layer="577" from-port="0" to-layer="578" to-port="0" />
 		<edge from-layer="578" from-port="1" to-layer="579" to-port="1" />
-		<edge from-layer="579" from-port="2" to-layer="582" to-port="0" />
-		<edge from-layer="580" from-port="0" to-layer="581" to-port="0" />
-		<edge from-layer="581" from-port="1" to-layer="582" to-port="1" />
-		<edge from-layer="582" from-port="2" to-layer="584" to-port="0" />
-		<edge from-layer="583" from-port="0" to-layer="584" to-port="1" />
-		<edge from-layer="584" from-port="2" to-layer="585" to-port="1" />
+		<edge from-layer="579" from-port="2" to-layer="581" to-port="0" />
+		<edge from-layer="580" from-port="0" to-layer="581" to-port="1" />
+		<edge from-layer="581" from-port="2" to-layer="583" to-port="0" />
+		<edge from-layer="582" from-port="0" to-layer="583" to-port="1" />
+		<edge from-layer="583" from-port="2" to-layer="584" to-port="0" />
+		<edge from-layer="584" from-port="1" to-layer="585" to-port="1" />
 		<edge from-layer="585" from-port="2" to-layer="586" to-port="1" />
 		<edge from-layer="586" from-port="2" to-layer="587" to-port="1" />
 		<edge from-layer="587" from-port="2" to-layer="588" to-port="0" />
@@ -26681,12 +26539,12 @@
 		<edge from-layer="602" from-port="1" to-layer="605" to-port="0" />
 		<edge from-layer="603" from-port="0" to-layer="604" to-port="0" />
 		<edge from-layer="604" from-port="1" to-layer="605" to-port="1" />
-		<edge from-layer="605" from-port="2" to-layer="608" to-port="0" />
-		<edge from-layer="606" from-port="0" to-layer="607" to-port="0" />
-		<edge from-layer="607" from-port="1" to-layer="608" to-port="1" />
-		<edge from-layer="608" from-port="2" to-layer="610" to-port="0" />
-		<edge from-layer="609" from-port="0" to-layer="610" to-port="1" />
-		<edge from-layer="610" from-port="2" to-layer="611" to-port="1" />
+		<edge from-layer="605" from-port="2" to-layer="607" to-port="0" />
+		<edge from-layer="606" from-port="0" to-layer="607" to-port="1" />
+		<edge from-layer="607" from-port="2" to-layer="609" to-port="0" />
+		<edge from-layer="608" from-port="0" to-layer="609" to-port="1" />
+		<edge from-layer="609" from-port="2" to-layer="610" to-port="0" />
+		<edge from-layer="610" from-port="1" to-layer="611" to-port="1" />
 		<edge from-layer="611" from-port="2" to-layer="612" to-port="1" />
 		<edge from-layer="612" from-port="2" to-layer="614" to-port="0" />
 		<edge from-layer="612" from-port="2" to-layer="618" to-port="0" />
@@ -26702,12 +26560,12 @@
 		<edge from-layer="622" from-port="1" to-layer="625" to-port="0" />
 		<edge from-layer="623" from-port="0" to-layer="624" to-port="0" />
 		<edge from-layer="624" from-port="1" to-layer="625" to-port="1" />
-		<edge from-layer="625" from-port="2" to-layer="628" to-port="0" />
-		<edge from-layer="626" from-port="0" to-layer="627" to-port="0" />
-		<edge from-layer="627" from-port="1" to-layer="628" to-port="1" />
-		<edge from-layer="628" from-port="2" to-layer="630" to-port="0" />
-		<edge from-layer="629" from-port="0" to-layer="630" to-port="1" />
-		<edge from-layer="630" from-port="2" to-layer="631" to-port="1" />
+		<edge from-layer="625" from-port="2" to-layer="627" to-port="0" />
+		<edge from-layer="626" from-port="0" to-layer="627" to-port="1" />
+		<edge from-layer="627" from-port="2" to-layer="629" to-port="0" />
+		<edge from-layer="628" from-port="0" to-layer="629" to-port="1" />
+		<edge from-layer="629" from-port="2" to-layer="630" to-port="0" />
+		<edge from-layer="630" from-port="1" to-layer="631" to-port="1" />
 		<edge from-layer="631" from-port="2" to-layer="632" to-port="1" />
 		<edge from-layer="632" from-port="2" to-layer="633" to-port="1" />
 		<edge from-layer="633" from-port="2" to-layer="634" to-port="0" />
@@ -26731,12 +26589,12 @@
 		<edge from-layer="648" from-port="1" to-layer="651" to-port="0" />
 		<edge from-layer="649" from-port="0" to-layer="650" to-port="0" />
 		<edge from-layer="650" from-port="1" to-layer="651" to-port="1" />
-		<edge from-layer="651" from-port="2" to-layer="654" to-port="0" />
-		<edge from-layer="652" from-port="0" to-layer="653" to-port="0" />
-		<edge from-layer="653" from-port="1" to-layer="654" to-port="1" />
-		<edge from-layer="654" from-port="2" to-layer="656" to-port="0" />
-		<edge from-layer="655" from-port="0" to-layer="656" to-port="1" />
-		<edge from-layer="656" from-port="2" to-layer="657" to-port="1" />
+		<edge from-layer="651" from-port="2" to-layer="653" to-port="0" />
+		<edge from-layer="652" from-port="0" to-layer="653" to-port="1" />
+		<edge from-layer="653" from-port="2" to-layer="655" to-port="0" />
+		<edge from-layer="654" from-port="0" to-layer="655" to-port="1" />
+		<edge from-layer="655" from-port="2" to-layer="656" to-port="0" />
+		<edge from-layer="656" from-port="1" to-layer="657" to-port="1" />
 		<edge from-layer="657" from-port="2" to-layer="658" to-port="1" />
 		<edge from-layer="658" from-port="2" to-layer="660" to-port="0" />
 		<edge from-layer="659" from-port="0" to-layer="660" to-port="1" />
@@ -26752,12 +26610,12 @@
 		<edge from-layer="668" from-port="1" to-layer="671" to-port="0" />
 		<edge from-layer="669" from-port="0" to-layer="670" to-port="0" />
 		<edge from-layer="670" from-port="1" to-layer="671" to-port="1" />
-		<edge from-layer="671" from-port="2" to-layer="674" to-port="0" />
-		<edge from-layer="672" from-port="0" to-layer="673" to-port="0" />
-		<edge from-layer="673" from-port="1" to-layer="674" to-port="1" />
-		<edge from-layer="674" from-port="2" to-layer="676" to-port="0" />
-		<edge from-layer="675" from-port="0" to-layer="676" to-port="1" />
-		<edge from-layer="676" from-port="2" to-layer="677" to-port="1" />
+		<edge from-layer="671" from-port="2" to-layer="673" to-port="0" />
+		<edge from-layer="672" from-port="0" to-layer="673" to-port="1" />
+		<edge from-layer="673" from-port="2" to-layer="675" to-port="0" />
+		<edge from-layer="674" from-port="0" to-layer="675" to-port="1" />
+		<edge from-layer="675" from-port="2" to-layer="676" to-port="0" />
+		<edge from-layer="676" from-port="1" to-layer="677" to-port="1" />
 		<edge from-layer="677" from-port="2" to-layer="679" to-port="0" />
 		<edge from-layer="678" from-port="0" to-layer="679" to-port="1" />
 		<edge from-layer="679" from-port="2" to-layer="681" to-port="0" />
@@ -26773,12 +26631,12 @@
 		<edge from-layer="689" from-port="1" to-layer="692" to-port="0" />
 		<edge from-layer="690" from-port="0" to-layer="691" to-port="0" />
 		<edge from-layer="691" from-port="1" to-layer="692" to-port="1" />
-		<edge from-layer="692" from-port="2" to-layer="695" to-port="0" />
-		<edge from-layer="693" from-port="0" to-layer="694" to-port="0" />
-		<edge from-layer="694" from-port="1" to-layer="695" to-port="1" />
-		<edge from-layer="695" from-port="2" to-layer="697" to-port="0" />
-		<edge from-layer="696" from-port="0" to-layer="697" to-port="1" />
-		<edge from-layer="697" from-port="2" to-layer="698" to-port="1" />
+		<edge from-layer="692" from-port="2" to-layer="694" to-port="0" />
+		<edge from-layer="693" from-port="0" to-layer="694" to-port="1" />
+		<edge from-layer="694" from-port="2" to-layer="696" to-port="0" />
+		<edge from-layer="695" from-port="0" to-layer="696" to-port="1" />
+		<edge from-layer="696" from-port="2" to-layer="697" to-port="0" />
+		<edge from-layer="697" from-port="1" to-layer="698" to-port="1" />
 		<edge from-layer="698" from-port="2" to-layer="699" to-port="1" />
 		<edge from-layer="699" from-port="2" to-layer="701" to-port="0" />
 		<edge from-layer="700" from-port="0" to-layer="701" to-port="1" />
@@ -26805,12 +26663,12 @@
 		<edge from-layer="719" from-port="1" to-layer="722" to-port="0" />
 		<edge from-layer="720" from-port="0" to-layer="721" to-port="0" />
 		<edge from-layer="721" from-port="1" to-layer="722" to-port="1" />
-		<edge from-layer="722" from-port="2" to-layer="725" to-port="0" />
-		<edge from-layer="723" from-port="0" to-layer="724" to-port="0" />
-		<edge from-layer="724" from-port="1" to-layer="725" to-port="1" />
-		<edge from-layer="725" from-port="2" to-layer="727" to-port="0" />
-		<edge from-layer="726" from-port="0" to-layer="727" to-port="1" />
-		<edge from-layer="727" from-port="2" to-layer="728" to-port="1" />
+		<edge from-layer="722" from-port="2" to-layer="724" to-port="0" />
+		<edge from-layer="723" from-port="0" to-layer="724" to-port="1" />
+		<edge from-layer="724" from-port="2" to-layer="726" to-port="0" />
+		<edge from-layer="725" from-port="0" to-layer="726" to-port="1" />
+		<edge from-layer="726" from-port="2" to-layer="727" to-port="0" />
+		<edge from-layer="727" from-port="1" to-layer="728" to-port="1" />
 		<edge from-layer="728" from-port="2" to-layer="729" to-port="1" />
 		<edge from-layer="729" from-port="2" to-layer="730" to-port="1" />
 		<edge from-layer="730" from-port="2" to-layer="731" to-port="0" />
@@ -26832,12 +26690,12 @@
 		<edge from-layer="745" from-port="1" to-layer="748" to-port="0" />
 		<edge from-layer="746" from-port="0" to-layer="747" to-port="0" />
 		<edge from-layer="747" from-port="1" to-layer="748" to-port="1" />
-		<edge from-layer="748" from-port="2" to-layer="751" to-port="0" />
-		<edge from-layer="749" from-port="0" to-layer="750" to-port="0" />
-		<edge from-layer="750" from-port="1" to-layer="751" to-port="1" />
-		<edge from-layer="751" from-port="2" to-layer="753" to-port="0" />
-		<edge from-layer="752" from-port="0" to-layer="753" to-port="1" />
-		<edge from-layer="753" from-port="2" to-layer="754" to-port="1" />
+		<edge from-layer="748" from-port="2" to-layer="750" to-port="0" />
+		<edge from-layer="749" from-port="0" to-layer="750" to-port="1" />
+		<edge from-layer="750" from-port="2" to-layer="752" to-port="0" />
+		<edge from-layer="751" from-port="0" to-layer="752" to-port="1" />
+		<edge from-layer="752" from-port="2" to-layer="753" to-port="0" />
+		<edge from-layer="753" from-port="1" to-layer="754" to-port="1" />
 		<edge from-layer="754" from-port="2" to-layer="755" to-port="1" />
 		<edge from-layer="755" from-port="2" to-layer="757" to-port="0" />
 		<edge from-layer="755" from-port="2" to-layer="761" to-port="0" />
@@ -26853,12 +26711,12 @@
 		<edge from-layer="765" from-port="1" to-layer="768" to-port="0" />
 		<edge from-layer="766" from-port="0" to-layer="767" to-port="0" />
 		<edge from-layer="767" from-port="1" to-layer="768" to-port="1" />
-		<edge from-layer="768" from-port="2" to-layer="771" to-port="0" />
-		<edge from-layer="769" from-port="0" to-layer="770" to-port="0" />
-		<edge from-layer="770" from-port="1" to-layer="771" to-port="1" />
-		<edge from-layer="771" from-port="2" to-layer="773" to-port="0" />
-		<edge from-layer="772" from-port="0" to-layer="773" to-port="1" />
-		<edge from-layer="773" from-port="2" to-layer="774" to-port="1" />
+		<edge from-layer="768" from-port="2" to-layer="770" to-port="0" />
+		<edge from-layer="769" from-port="0" to-layer="770" to-port="1" />
+		<edge from-layer="770" from-port="2" to-layer="772" to-port="0" />
+		<edge from-layer="771" from-port="0" to-layer="772" to-port="1" />
+		<edge from-layer="772" from-port="2" to-layer="773" to-port="0" />
+		<edge from-layer="773" from-port="1" to-layer="774" to-port="1" />
 		<edge from-layer="774" from-port="2" to-layer="775" to-port="1" />
 		<edge from-layer="775" from-port="2" to-layer="776" to-port="1" />
 		<edge from-layer="776" from-port="2" to-layer="777" to-port="0" />
@@ -26882,12 +26740,12 @@
 		<edge from-layer="791" from-port="1" to-layer="794" to-port="0" />
 		<edge from-layer="792" from-port="0" to-layer="793" to-port="0" />
 		<edge from-layer="793" from-port="1" to-layer="794" to-port="1" />
-		<edge from-layer="794" from-port="2" to-layer="797" to-port="0" />
-		<edge from-layer="795" from-port="0" to-layer="796" to-port="0" />
-		<edge from-layer="796" from-port="1" to-layer="797" to-port="1" />
-		<edge from-layer="797" from-port="2" to-layer="799" to-port="0" />
-		<edge from-layer="798" from-port="0" to-layer="799" to-port="1" />
-		<edge from-layer="799" from-port="2" to-layer="800" to-port="1" />
+		<edge from-layer="794" from-port="2" to-layer="796" to-port="0" />
+		<edge from-layer="795" from-port="0" to-layer="796" to-port="1" />
+		<edge from-layer="796" from-port="2" to-layer="798" to-port="0" />
+		<edge from-layer="797" from-port="0" to-layer="798" to-port="1" />
+		<edge from-layer="798" from-port="2" to-layer="799" to-port="0" />
+		<edge from-layer="799" from-port="1" to-layer="800" to-port="1" />
 		<edge from-layer="800" from-port="2" to-layer="801" to-port="1" />
 		<edge from-layer="801" from-port="2" to-layer="803" to-port="0" />
 		<edge from-layer="802" from-port="0" to-layer="803" to-port="1" />
@@ -26903,12 +26761,12 @@
 		<edge from-layer="811" from-port="1" to-layer="814" to-port="0" />
 		<edge from-layer="812" from-port="0" to-layer="813" to-port="0" />
 		<edge from-layer="813" from-port="1" to-layer="814" to-port="1" />
-		<edge from-layer="814" from-port="2" to-layer="817" to-port="0" />
-		<edge from-layer="815" from-port="0" to-layer="816" to-port="0" />
-		<edge from-layer="816" from-port="1" to-layer="817" to-port="1" />
-		<edge from-layer="817" from-port="2" to-layer="819" to-port="0" />
-		<edge from-layer="818" from-port="0" to-layer="819" to-port="1" />
-		<edge from-layer="819" from-port="2" to-layer="820" to-port="1" />
+		<edge from-layer="814" from-port="2" to-layer="816" to-port="0" />
+		<edge from-layer="815" from-port="0" to-layer="816" to-port="1" />
+		<edge from-layer="816" from-port="2" to-layer="818" to-port="0" />
+		<edge from-layer="817" from-port="0" to-layer="818" to-port="1" />
+		<edge from-layer="818" from-port="2" to-layer="819" to-port="0" />
+		<edge from-layer="819" from-port="1" to-layer="820" to-port="1" />
 		<edge from-layer="820" from-port="2" to-layer="822" to-port="0" />
 		<edge from-layer="821" from-port="0" to-layer="822" to-port="1" />
 		<edge from-layer="822" from-port="2" to-layer="824" to-port="0" />
@@ -26924,12 +26782,12 @@
 		<edge from-layer="832" from-port="1" to-layer="835" to-port="0" />
 		<edge from-layer="833" from-port="0" to-layer="834" to-port="0" />
 		<edge from-layer="834" from-port="1" to-layer="835" to-port="1" />
-		<edge from-layer="835" from-port="2" to-layer="838" to-port="0" />
-		<edge from-layer="836" from-port="0" to-layer="837" to-port="0" />
-		<edge from-layer="837" from-port="1" to-layer="838" to-port="1" />
-		<edge from-layer="838" from-port="2" to-layer="840" to-port="0" />
-		<edge from-layer="839" from-port="0" to-layer="840" to-port="1" />
-		<edge from-layer="840" from-port="2" to-layer="841" to-port="1" />
+		<edge from-layer="835" from-port="2" to-layer="837" to-port="0" />
+		<edge from-layer="836" from-port="0" to-layer="837" to-port="1" />
+		<edge from-layer="837" from-port="2" to-layer="839" to-port="0" />
+		<edge from-layer="838" from-port="0" to-layer="839" to-port="1" />
+		<edge from-layer="839" from-port="2" to-layer="840" to-port="0" />
+		<edge from-layer="840" from-port="1" to-layer="841" to-port="1" />
 		<edge from-layer="841" from-port="2" to-layer="842" to-port="1" />
 		<edge from-layer="842" from-port="2" to-layer="844" to-port="0" />
 		<edge from-layer="843" from-port="0" to-layer="844" to-port="1" />
@@ -26956,12 +26814,12 @@
 		<edge from-layer="862" from-port="1" to-layer="865" to-port="0" />
 		<edge from-layer="863" from-port="0" to-layer="864" to-port="0" />
 		<edge from-layer="864" from-port="1" to-layer="865" to-port="1" />
-		<edge from-layer="865" from-port="2" to-layer="868" to-port="0" />
-		<edge from-layer="866" from-port="0" to-layer="867" to-port="0" />
-		<edge from-layer="867" from-port="1" to-layer="868" to-port="1" />
-		<edge from-layer="868" from-port="2" to-layer="870" to-port="0" />
-		<edge from-layer="869" from-port="0" to-layer="870" to-port="1" />
-		<edge from-layer="870" from-port="2" to-layer="871" to-port="1" />
+		<edge from-layer="865" from-port="2" to-layer="867" to-port="0" />
+		<edge from-layer="866" from-port="0" to-layer="867" to-port="1" />
+		<edge from-layer="867" from-port="2" to-layer="869" to-port="0" />
+		<edge from-layer="868" from-port="0" to-layer="869" to-port="1" />
+		<edge from-layer="869" from-port="2" to-layer="870" to-port="0" />
+		<edge from-layer="870" from-port="1" to-layer="871" to-port="1" />
 		<edge from-layer="871" from-port="2" to-layer="872" to-port="1" />
 		<edge from-layer="872" from-port="2" to-layer="873" to-port="1" />
 		<edge from-layer="873" from-port="2" to-layer="874" to-port="0" />
@@ -26983,12 +26841,12 @@
 		<edge from-layer="888" from-port="1" to-layer="891" to-port="0" />
 		<edge from-layer="889" from-port="0" to-layer="890" to-port="0" />
 		<edge from-layer="890" from-port="1" to-layer="891" to-port="1" />
-		<edge from-layer="891" from-port="2" to-layer="894" to-port="0" />
-		<edge from-layer="892" from-port="0" to-layer="893" to-port="0" />
-		<edge from-layer="893" from-port="1" to-layer="894" to-port="1" />
-		<edge from-layer="894" from-port="2" to-layer="896" to-port="0" />
-		<edge from-layer="895" from-port="0" to-layer="896" to-port="1" />
-		<edge from-layer="896" from-port="2" to-layer="897" to-port="1" />
+		<edge from-layer="891" from-port="2" to-layer="893" to-port="0" />
+		<edge from-layer="892" from-port="0" to-layer="893" to-port="1" />
+		<edge from-layer="893" from-port="2" to-layer="895" to-port="0" />
+		<edge from-layer="894" from-port="0" to-layer="895" to-port="1" />
+		<edge from-layer="895" from-port="2" to-layer="896" to-port="0" />
+		<edge from-layer="896" from-port="1" to-layer="897" to-port="1" />
 		<edge from-layer="897" from-port="2" to-layer="898" to-port="1" />
 		<edge from-layer="898" from-port="2" to-layer="900" to-port="0" />
 		<edge from-layer="898" from-port="2" to-layer="904" to-port="0" />
@@ -27004,12 +26862,12 @@
 		<edge from-layer="908" from-port="1" to-layer="911" to-port="0" />
 		<edge from-layer="909" from-port="0" to-layer="910" to-port="0" />
 		<edge from-layer="910" from-port="1" to-layer="911" to-port="1" />
-		<edge from-layer="911" from-port="2" to-layer="914" to-port="0" />
-		<edge from-layer="912" from-port="0" to-layer="913" to-port="0" />
-		<edge from-layer="913" from-port="1" to-layer="914" to-port="1" />
-		<edge from-layer="914" from-port="2" to-layer="916" to-port="0" />
-		<edge from-layer="915" from-port="0" to-layer="916" to-port="1" />
-		<edge from-layer="916" from-port="2" to-layer="917" to-port="1" />
+		<edge from-layer="911" from-port="2" to-layer="913" to-port="0" />
+		<edge from-layer="912" from-port="0" to-layer="913" to-port="1" />
+		<edge from-layer="913" from-port="2" to-layer="915" to-port="0" />
+		<edge from-layer="914" from-port="0" to-layer="915" to-port="1" />
+		<edge from-layer="915" from-port="2" to-layer="916" to-port="0" />
+		<edge from-layer="916" from-port="1" to-layer="917" to-port="1" />
 		<edge from-layer="917" from-port="2" to-layer="918" to-port="1" />
 		<edge from-layer="918" from-port="2" to-layer="919" to-port="1" />
 		<edge from-layer="919" from-port="2" to-layer="920" to-port="0" />
@@ -27033,12 +26891,12 @@
 		<edge from-layer="934" from-port="1" to-layer="937" to-port="0" />
 		<edge from-layer="935" from-port="0" to-layer="936" to-port="0" />
 		<edge from-layer="936" from-port="1" to-layer="937" to-port="1" />
-		<edge from-layer="937" from-port="2" to-layer="940" to-port="0" />
-		<edge from-layer="938" from-port="0" to-layer="939" to-port="0" />
-		<edge from-layer="939" from-port="1" to-layer="940" to-port="1" />
-		<edge from-layer="940" from-port="2" to-layer="942" to-port="0" />
-		<edge from-layer="941" from-port="0" to-layer="942" to-port="1" />
-		<edge from-layer="942" from-port="2" to-layer="943" to-port="1" />
+		<edge from-layer="937" from-port="2" to-layer="939" to-port="0" />
+		<edge from-layer="938" from-port="0" to-layer="939" to-port="1" />
+		<edge from-layer="939" from-port="2" to-layer="941" to-port="0" />
+		<edge from-layer="940" from-port="0" to-layer="941" to-port="1" />
+		<edge from-layer="941" from-port="2" to-layer="942" to-port="0" />
+		<edge from-layer="942" from-port="1" to-layer="943" to-port="1" />
 		<edge from-layer="943" from-port="2" to-layer="944" to-port="1" />
 		<edge from-layer="944" from-port="2" to-layer="946" to-port="0" />
 		<edge from-layer="945" from-port="0" to-layer="946" to-port="1" />
@@ -27054,12 +26912,12 @@
 		<edge from-layer="954" from-port="1" to-layer="957" to-port="0" />
 		<edge from-layer="955" from-port="0" to-layer="956" to-port="0" />
 		<edge from-layer="956" from-port="1" to-layer="957" to-port="1" />
-		<edge from-layer="957" from-port="2" to-layer="960" to-port="0" />
-		<edge from-layer="958" from-port="0" to-layer="959" to-port="0" />
-		<edge from-layer="959" from-port="1" to-layer="960" to-port="1" />
-		<edge from-layer="960" from-port="2" to-layer="962" to-port="0" />
-		<edge from-layer="961" from-port="0" to-layer="962" to-port="1" />
-		<edge from-layer="962" from-port="2" to-layer="963" to-port="1" />
+		<edge from-layer="957" from-port="2" to-layer="959" to-port="0" />
+		<edge from-layer="958" from-port="0" to-layer="959" to-port="1" />
+		<edge from-layer="959" from-port="2" to-layer="961" to-port="0" />
+		<edge from-layer="960" from-port="0" to-layer="961" to-port="1" />
+		<edge from-layer="961" from-port="2" to-layer="962" to-port="0" />
+		<edge from-layer="962" from-port="1" to-layer="963" to-port="1" />
 		<edge from-layer="963" from-port="2" to-layer="965" to-port="0" />
 		<edge from-layer="964" from-port="0" to-layer="965" to-port="1" />
 		<edge from-layer="965" from-port="2" to-layer="967" to-port="0" />
@@ -27075,12 +26933,12 @@
 		<edge from-layer="975" from-port="1" to-layer="978" to-port="0" />
 		<edge from-layer="976" from-port="0" to-layer="977" to-port="0" />
 		<edge from-layer="977" from-port="1" to-layer="978" to-port="1" />
-		<edge from-layer="978" from-port="2" to-layer="981" to-port="0" />
-		<edge from-layer="979" from-port="0" to-layer="980" to-port="0" />
-		<edge from-layer="980" from-port="1" to-layer="981" to-port="1" />
-		<edge from-layer="981" from-port="2" to-layer="983" to-port="0" />
-		<edge from-layer="982" from-port="0" to-layer="983" to-port="1" />
-		<edge from-layer="983" from-port="2" to-layer="984" to-port="1" />
+		<edge from-layer="978" from-port="2" to-layer="980" to-port="0" />
+		<edge from-layer="979" from-port="0" to-layer="980" to-port="1" />
+		<edge from-layer="980" from-port="2" to-layer="982" to-port="0" />
+		<edge from-layer="981" from-port="0" to-layer="982" to-port="1" />
+		<edge from-layer="982" from-port="2" to-layer="983" to-port="0" />
+		<edge from-layer="983" from-port="1" to-layer="984" to-port="1" />
 		<edge from-layer="984" from-port="2" to-layer="985" to-port="1" />
 		<edge from-layer="985" from-port="2" to-layer="987" to-port="0" />
 		<edge from-layer="986" from-port="0" to-layer="987" to-port="1" />
@@ -27107,12 +26965,12 @@
 		<edge from-layer="1005" from-port="1" to-layer="1008" to-port="0" />
 		<edge from-layer="1006" from-port="0" to-layer="1007" to-port="0" />
 		<edge from-layer="1007" from-port="1" to-layer="1008" to-port="1" />
-		<edge from-layer="1008" from-port="2" to-layer="1011" to-port="0" />
-		<edge from-layer="1009" from-port="0" to-layer="1010" to-port="0" />
-		<edge from-layer="1010" from-port="1" to-layer="1011" to-port="1" />
-		<edge from-layer="1011" from-port="2" to-layer="1013" to-port="0" />
-		<edge from-layer="1012" from-port="0" to-layer="1013" to-port="1" />
-		<edge from-layer="1013" from-port="2" to-layer="1014" to-port="1" />
+		<edge from-layer="1008" from-port="2" to-layer="1010" to-port="0" />
+		<edge from-layer="1009" from-port="0" to-layer="1010" to-port="1" />
+		<edge from-layer="1010" from-port="2" to-layer="1012" to-port="0" />
+		<edge from-layer="1011" from-port="0" to-layer="1012" to-port="1" />
+		<edge from-layer="1012" from-port="2" to-layer="1013" to-port="0" />
+		<edge from-layer="1013" from-port="1" to-layer="1014" to-port="1" />
 		<edge from-layer="1014" from-port="2" to-layer="1015" to-port="1" />
 		<edge from-layer="1015" from-port="2" to-layer="1016" to-port="1" />
 		<edge from-layer="1016" from-port="2" to-layer="1017" to-port="0" />
@@ -27134,12 +26992,12 @@
 		<edge from-layer="1031" from-port="1" to-layer="1034" to-port="0" />
 		<edge from-layer="1032" from-port="0" to-layer="1033" to-port="0" />
 		<edge from-layer="1033" from-port="1" to-layer="1034" to-port="1" />
-		<edge from-layer="1034" from-port="2" to-layer="1037" to-port="0" />
-		<edge from-layer="1035" from-port="0" to-layer="1036" to-port="0" />
-		<edge from-layer="1036" from-port="1" to-layer="1037" to-port="1" />
-		<edge from-layer="1037" from-port="2" to-layer="1039" to-port="0" />
-		<edge from-layer="1038" from-port="0" to-layer="1039" to-port="1" />
-		<edge from-layer="1039" from-port="2" to-layer="1040" to-port="1" />
+		<edge from-layer="1034" from-port="2" to-layer="1036" to-port="0" />
+		<edge from-layer="1035" from-port="0" to-layer="1036" to-port="1" />
+		<edge from-layer="1036" from-port="2" to-layer="1038" to-port="0" />
+		<edge from-layer="1037" from-port="0" to-layer="1038" to-port="1" />
+		<edge from-layer="1038" from-port="2" to-layer="1039" to-port="0" />
+		<edge from-layer="1039" from-port="1" to-layer="1040" to-port="1" />
 		<edge from-layer="1040" from-port="2" to-layer="1041" to-port="1" />
 		<edge from-layer="1041" from-port="2" to-layer="1043" to-port="0" />
 		<edge from-layer="1041" from-port="2" to-layer="1047" to-port="0" />
@@ -27155,12 +27013,12 @@
 		<edge from-layer="1051" from-port="1" to-layer="1054" to-port="0" />
 		<edge from-layer="1052" from-port="0" to-layer="1053" to-port="0" />
 		<edge from-layer="1053" from-port="1" to-layer="1054" to-port="1" />
-		<edge from-layer="1054" from-port="2" to-layer="1057" to-port="0" />
-		<edge from-layer="1055" from-port="0" to-layer="1056" to-port="0" />
-		<edge from-layer="1056" from-port="1" to-layer="1057" to-port="1" />
-		<edge from-layer="1057" from-port="2" to-layer="1059" to-port="0" />
-		<edge from-layer="1058" from-port="0" to-layer="1059" to-port="1" />
-		<edge from-layer="1059" from-port="2" to-layer="1060" to-port="1" />
+		<edge from-layer="1054" from-port="2" to-layer="1056" to-port="0" />
+		<edge from-layer="1055" from-port="0" to-layer="1056" to-port="1" />
+		<edge from-layer="1056" from-port="2" to-layer="1058" to-port="0" />
+		<edge from-layer="1057" from-port="0" to-layer="1058" to-port="1" />
+		<edge from-layer="1058" from-port="2" to-layer="1059" to-port="0" />
+		<edge from-layer="1059" from-port="1" to-layer="1060" to-port="1" />
 		<edge from-layer="1060" from-port="2" to-layer="1061" to-port="1" />
 		<edge from-layer="1061" from-port="2" to-layer="1062" to-port="1" />
 		<edge from-layer="1062" from-port="2" to-layer="1063" to-port="0" />
@@ -27184,12 +27042,12 @@
 		<edge from-layer="1077" from-port="1" to-layer="1080" to-port="0" />
 		<edge from-layer="1078" from-port="0" to-layer="1079" to-port="0" />
 		<edge from-layer="1079" from-port="1" to-layer="1080" to-port="1" />
-		<edge from-layer="1080" from-port="2" to-layer="1083" to-port="0" />
-		<edge from-layer="1081" from-port="0" to-layer="1082" to-port="0" />
-		<edge from-layer="1082" from-port="1" to-layer="1083" to-port="1" />
-		<edge from-layer="1083" from-port="2" to-layer="1085" to-port="0" />
-		<edge from-layer="1084" from-port="0" to-layer="1085" to-port="1" />
-		<edge from-layer="1085" from-port="2" to-layer="1086" to-port="1" />
+		<edge from-layer="1080" from-port="2" to-layer="1082" to-port="0" />
+		<edge from-layer="1081" from-port="0" to-layer="1082" to-port="1" />
+		<edge from-layer="1082" from-port="2" to-layer="1084" to-port="0" />
+		<edge from-layer="1083" from-port="0" to-layer="1084" to-port="1" />
+		<edge from-layer="1084" from-port="2" to-layer="1085" to-port="0" />
+		<edge from-layer="1085" from-port="1" to-layer="1086" to-port="1" />
 		<edge from-layer="1086" from-port="2" to-layer="1087" to-port="1" />
 		<edge from-layer="1087" from-port="2" to-layer="1089" to-port="0" />
 		<edge from-layer="1088" from-port="0" to-layer="1089" to-port="1" />
@@ -27205,12 +27063,12 @@
 		<edge from-layer="1097" from-port="1" to-layer="1100" to-port="0" />
 		<edge from-layer="1098" from-port="0" to-layer="1099" to-port="0" />
 		<edge from-layer="1099" from-port="1" to-layer="1100" to-port="1" />
-		<edge from-layer="1100" from-port="2" to-layer="1103" to-port="0" />
-		<edge from-layer="1101" from-port="0" to-layer="1102" to-port="0" />
-		<edge from-layer="1102" from-port="1" to-layer="1103" to-port="1" />
-		<edge from-layer="1103" from-port="2" to-layer="1105" to-port="0" />
-		<edge from-layer="1104" from-port="0" to-layer="1105" to-port="1" />
-		<edge from-layer="1105" from-port="2" to-layer="1106" to-port="1" />
+		<edge from-layer="1100" from-port="2" to-layer="1102" to-port="0" />
+		<edge from-layer="1101" from-port="0" to-layer="1102" to-port="1" />
+		<edge from-layer="1102" from-port="2" to-layer="1104" to-port="0" />
+		<edge from-layer="1103" from-port="0" to-layer="1104" to-port="1" />
+		<edge from-layer="1104" from-port="2" to-layer="1105" to-port="0" />
+		<edge from-layer="1105" from-port="1" to-layer="1106" to-port="1" />
 		<edge from-layer="1106" from-port="2" to-layer="1108" to-port="0" />
 		<edge from-layer="1107" from-port="0" to-layer="1108" to-port="1" />
 		<edge from-layer="1108" from-port="2" to-layer="1110" to-port="0" />
@@ -27226,12 +27084,12 @@
 		<edge from-layer="1118" from-port="1" to-layer="1121" to-port="0" />
 		<edge from-layer="1119" from-port="0" to-layer="1120" to-port="0" />
 		<edge from-layer="1120" from-port="1" to-layer="1121" to-port="1" />
-		<edge from-layer="1121" from-port="2" to-layer="1124" to-port="0" />
-		<edge from-layer="1122" from-port="0" to-layer="1123" to-port="0" />
-		<edge from-layer="1123" from-port="1" to-layer="1124" to-port="1" />
-		<edge from-layer="1124" from-port="2" to-layer="1126" to-port="0" />
-		<edge from-layer="1125" from-port="0" to-layer="1126" to-port="1" />
-		<edge from-layer="1126" from-port="2" to-layer="1127" to-port="1" />
+		<edge from-layer="1121" from-port="2" to-layer="1123" to-port="0" />
+		<edge from-layer="1122" from-port="0" to-layer="1123" to-port="1" />
+		<edge from-layer="1123" from-port="2" to-layer="1125" to-port="0" />
+		<edge from-layer="1124" from-port="0" to-layer="1125" to-port="1" />
+		<edge from-layer="1125" from-port="2" to-layer="1126" to-port="0" />
+		<edge from-layer="1126" from-port="1" to-layer="1127" to-port="1" />
 		<edge from-layer="1127" from-port="2" to-layer="1128" to-port="1" />
 		<edge from-layer="1128" from-port="2" to-layer="1130" to-port="0" />
 		<edge from-layer="1129" from-port="0" to-layer="1130" to-port="1" />
@@ -27258,12 +27116,12 @@
 		<edge from-layer="1148" from-port="1" to-layer="1151" to-port="0" />
 		<edge from-layer="1149" from-port="0" to-layer="1150" to-port="0" />
 		<edge from-layer="1150" from-port="1" to-layer="1151" to-port="1" />
-		<edge from-layer="1151" from-port="2" to-layer="1154" to-port="0" />
-		<edge from-layer="1152" from-port="0" to-layer="1153" to-port="0" />
-		<edge from-layer="1153" from-port="1" to-layer="1154" to-port="1" />
-		<edge from-layer="1154" from-port="2" to-layer="1156" to-port="0" />
-		<edge from-layer="1155" from-port="0" to-layer="1156" to-port="1" />
-		<edge from-layer="1156" from-port="2" to-layer="1157" to-port="1" />
+		<edge from-layer="1151" from-port="2" to-layer="1153" to-port="0" />
+		<edge from-layer="1152" from-port="0" to-layer="1153" to-port="1" />
+		<edge from-layer="1153" from-port="2" to-layer="1155" to-port="0" />
+		<edge from-layer="1154" from-port="0" to-layer="1155" to-port="1" />
+		<edge from-layer="1155" from-port="2" to-layer="1156" to-port="0" />
+		<edge from-layer="1156" from-port="1" to-layer="1157" to-port="1" />
 		<edge from-layer="1157" from-port="2" to-layer="1158" to-port="1" />
 		<edge from-layer="1158" from-port="2" to-layer="1159" to-port="1" />
 		<edge from-layer="1159" from-port="2" to-layer="1160" to-port="0" />
@@ -27285,12 +27143,12 @@
 		<edge from-layer="1174" from-port="1" to-layer="1177" to-port="0" />
 		<edge from-layer="1175" from-port="0" to-layer="1176" to-port="0" />
 		<edge from-layer="1176" from-port="1" to-layer="1177" to-port="1" />
-		<edge from-layer="1177" from-port="2" to-layer="1180" to-port="0" />
-		<edge from-layer="1178" from-port="0" to-layer="1179" to-port="0" />
-		<edge from-layer="1179" from-port="1" to-layer="1180" to-port="1" />
-		<edge from-layer="1180" from-port="2" to-layer="1182" to-port="0" />
-		<edge from-layer="1181" from-port="0" to-layer="1182" to-port="1" />
-		<edge from-layer="1182" from-port="2" to-layer="1183" to-port="1" />
+		<edge from-layer="1177" from-port="2" to-layer="1179" to-port="0" />
+		<edge from-layer="1178" from-port="0" to-layer="1179" to-port="1" />
+		<edge from-layer="1179" from-port="2" to-layer="1181" to-port="0" />
+		<edge from-layer="1180" from-port="0" to-layer="1181" to-port="1" />
+		<edge from-layer="1181" from-port="2" to-layer="1182" to-port="0" />
+		<edge from-layer="1182" from-port="1" to-layer="1183" to-port="1" />
 		<edge from-layer="1183" from-port="2" to-layer="1184" to-port="1" />
 		<edge from-layer="1184" from-port="2" to-layer="1186" to-port="0" />
 		<edge from-layer="1184" from-port="2" to-layer="1190" to-port="0" />
@@ -27306,12 +27164,12 @@
 		<edge from-layer="1194" from-port="1" to-layer="1197" to-port="0" />
 		<edge from-layer="1195" from-port="0" to-layer="1196" to-port="0" />
 		<edge from-layer="1196" from-port="1" to-layer="1197" to-port="1" />
-		<edge from-layer="1197" from-port="2" to-layer="1200" to-port="0" />
-		<edge from-layer="1198" from-port="0" to-layer="1199" to-port="0" />
-		<edge from-layer="1199" from-port="1" to-layer="1200" to-port="1" />
-		<edge from-layer="1200" from-port="2" to-layer="1202" to-port="0" />
-		<edge from-layer="1201" from-port="0" to-layer="1202" to-port="1" />
-		<edge from-layer="1202" from-port="2" to-layer="1203" to-port="1" />
+		<edge from-layer="1197" from-port="2" to-layer="1199" to-port="0" />
+		<edge from-layer="1198" from-port="0" to-layer="1199" to-port="1" />
+		<edge from-layer="1199" from-port="2" to-layer="1201" to-port="0" />
+		<edge from-layer="1200" from-port="0" to-layer="1201" to-port="1" />
+		<edge from-layer="1201" from-port="2" to-layer="1202" to-port="0" />
+		<edge from-layer="1202" from-port="1" to-layer="1203" to-port="1" />
 		<edge from-layer="1203" from-port="2" to-layer="1204" to-port="1" />
 		<edge from-layer="1204" from-port="2" to-layer="1205" to-port="1" />
 		<edge from-layer="1205" from-port="2" to-layer="1206" to-port="0" />
@@ -27335,12 +27193,12 @@
 		<edge from-layer="1220" from-port="1" to-layer="1223" to-port="0" />
 		<edge from-layer="1221" from-port="0" to-layer="1222" to-port="0" />
 		<edge from-layer="1222" from-port="1" to-layer="1223" to-port="1" />
-		<edge from-layer="1223" from-port="2" to-layer="1226" to-port="0" />
-		<edge from-layer="1224" from-port="0" to-layer="1225" to-port="0" />
-		<edge from-layer="1225" from-port="1" to-layer="1226" to-port="1" />
-		<edge from-layer="1226" from-port="2" to-layer="1228" to-port="0" />
-		<edge from-layer="1227" from-port="0" to-layer="1228" to-port="1" />
-		<edge from-layer="1228" from-port="2" to-layer="1229" to-port="1" />
+		<edge from-layer="1223" from-port="2" to-layer="1225" to-port="0" />
+		<edge from-layer="1224" from-port="0" to-layer="1225" to-port="1" />
+		<edge from-layer="1225" from-port="2" to-layer="1227" to-port="0" />
+		<edge from-layer="1226" from-port="0" to-layer="1227" to-port="1" />
+		<edge from-layer="1227" from-port="2" to-layer="1228" to-port="0" />
+		<edge from-layer="1228" from-port="1" to-layer="1229" to-port="1" />
 		<edge from-layer="1229" from-port="2" to-layer="1230" to-port="1" />
 		<edge from-layer="1230" from-port="2" to-layer="1232" to-port="0" />
 		<edge from-layer="1231" from-port="0" to-layer="1232" to-port="1" />
@@ -27356,12 +27214,12 @@
 		<edge from-layer="1240" from-port="1" to-layer="1243" to-port="0" />
 		<edge from-layer="1241" from-port="0" to-layer="1242" to-port="0" />
 		<edge from-layer="1242" from-port="1" to-layer="1243" to-port="1" />
-		<edge from-layer="1243" from-port="2" to-layer="1246" to-port="0" />
-		<edge from-layer="1244" from-port="0" to-layer="1245" to-port="0" />
-		<edge from-layer="1245" from-port="1" to-layer="1246" to-port="1" />
-		<edge from-layer="1246" from-port="2" to-layer="1248" to-port="0" />
-		<edge from-layer="1247" from-port="0" to-layer="1248" to-port="1" />
-		<edge from-layer="1248" from-port="2" to-layer="1249" to-port="1" />
+		<edge from-layer="1243" from-port="2" to-layer="1245" to-port="0" />
+		<edge from-layer="1244" from-port="0" to-layer="1245" to-port="1" />
+		<edge from-layer="1245" from-port="2" to-layer="1247" to-port="0" />
+		<edge from-layer="1246" from-port="0" to-layer="1247" to-port="1" />
+		<edge from-layer="1247" from-port="2" to-layer="1248" to-port="0" />
+		<edge from-layer="1248" from-port="1" to-layer="1249" to-port="1" />
 		<edge from-layer="1249" from-port="2" to-layer="1251" to-port="0" />
 		<edge from-layer="1250" from-port="0" to-layer="1251" to-port="1" />
 		<edge from-layer="1251" from-port="2" to-layer="1253" to-port="0" />
@@ -27377,12 +27235,12 @@
 		<edge from-layer="1261" from-port="1" to-layer="1264" to-port="0" />
 		<edge from-layer="1262" from-port="0" to-layer="1263" to-port="0" />
 		<edge from-layer="1263" from-port="1" to-layer="1264" to-port="1" />
-		<edge from-layer="1264" from-port="2" to-layer="1267" to-port="0" />
-		<edge from-layer="1265" from-port="0" to-layer="1266" to-port="0" />
-		<edge from-layer="1266" from-port="1" to-layer="1267" to-port="1" />
-		<edge from-layer="1267" from-port="2" to-layer="1269" to-port="0" />
-		<edge from-layer="1268" from-port="0" to-layer="1269" to-port="1" />
-		<edge from-layer="1269" from-port="2" to-layer="1270" to-port="1" />
+		<edge from-layer="1264" from-port="2" to-layer="1266" to-port="0" />
+		<edge from-layer="1265" from-port="0" to-layer="1266" to-port="1" />
+		<edge from-layer="1266" from-port="2" to-layer="1268" to-port="0" />
+		<edge from-layer="1267" from-port="0" to-layer="1268" to-port="1" />
+		<edge from-layer="1268" from-port="2" to-layer="1269" to-port="0" />
+		<edge from-layer="1269" from-port="1" to-layer="1270" to-port="1" />
 		<edge from-layer="1270" from-port="2" to-layer="1271" to-port="1" />
 		<edge from-layer="1271" from-port="2" to-layer="1273" to-port="0" />
 		<edge from-layer="1272" from-port="0" to-layer="1273" to-port="1" />
@@ -27391,11 +27249,11 @@
 		<edge from-layer="1275" from-port="2" to-layer="1276" to-port="1" />
 		<edge from-layer="1276" from-port="2" to-layer="1278" to-port="0" />
 		<edge from-layer="1277" from-port="0" to-layer="1278" to-port="1" />
-		<edge from-layer="1278" from-port="2" to-layer="1279" to-port="0" />
 		<edge from-layer="1278" from-port="2" to-layer="1289" to-port="0" />
+		<edge from-layer="1278" from-port="2" to-layer="1279" to-port="0" />
 		<edge from-layer="1279" from-port="1" to-layer="1286" to-port="0" />
-		<edge from-layer="1280" from-port="0" to-layer="1283" to-port="0" />
 		<edge from-layer="1280" from-port="0" to-layer="1286" to-port="1" />
+		<edge from-layer="1280" from-port="0" to-layer="1283" to-port="0" />
 		<edge from-layer="1281" from-port="0" to-layer="1286" to-port="2" />
 		<edge from-layer="1282" from-port="0" to-layer="1284" to-port="0" />
 		<edge from-layer="1283" from-port="1" to-layer="1284" to-port="1" />
@@ -27409,12 +27267,12 @@
 		<edge from-layer="1291" from-port="1" to-layer="1294" to-port="0" />
 		<edge from-layer="1292" from-port="0" to-layer="1293" to-port="0" />
 		<edge from-layer="1293" from-port="1" to-layer="1294" to-port="1" />
-		<edge from-layer="1294" from-port="2" to-layer="1297" to-port="0" />
-		<edge from-layer="1295" from-port="0" to-layer="1296" to-port="0" />
-		<edge from-layer="1296" from-port="1" to-layer="1297" to-port="1" />
-		<edge from-layer="1297" from-port="2" to-layer="1299" to-port="0" />
-		<edge from-layer="1298" from-port="0" to-layer="1299" to-port="1" />
-		<edge from-layer="1299" from-port="2" to-layer="1300" to-port="1" />
+		<edge from-layer="1294" from-port="2" to-layer="1296" to-port="0" />
+		<edge from-layer="1295" from-port="0" to-layer="1296" to-port="1" />
+		<edge from-layer="1296" from-port="2" to-layer="1298" to-port="0" />
+		<edge from-layer="1297" from-port="0" to-layer="1298" to-port="1" />
+		<edge from-layer="1298" from-port="2" to-layer="1299" to-port="0" />
+		<edge from-layer="1299" from-port="1" to-layer="1300" to-port="1" />
 		<edge from-layer="1300" from-port="2" to-layer="1301" to-port="1" />
 		<edge from-layer="1301" from-port="2" to-layer="1302" to-port="1" />
 		<edge from-layer="1302" from-port="2" to-layer="1303" to-port="0" />
@@ -27436,12 +27294,12 @@
 		<edge from-layer="1317" from-port="1" to-layer="1320" to-port="0" />
 		<edge from-layer="1318" from-port="0" to-layer="1319" to-port="0" />
 		<edge from-layer="1319" from-port="1" to-layer="1320" to-port="1" />
-		<edge from-layer="1320" from-port="2" to-layer="1323" to-port="0" />
-		<edge from-layer="1321" from-port="0" to-layer="1322" to-port="0" />
-		<edge from-layer="1322" from-port="1" to-layer="1323" to-port="1" />
-		<edge from-layer="1323" from-port="2" to-layer="1325" to-port="0" />
-		<edge from-layer="1324" from-port="0" to-layer="1325" to-port="1" />
-		<edge from-layer="1325" from-port="2" to-layer="1326" to-port="1" />
+		<edge from-layer="1320" from-port="2" to-layer="1322" to-port="0" />
+		<edge from-layer="1321" from-port="0" to-layer="1322" to-port="1" />
+		<edge from-layer="1322" from-port="2" to-layer="1324" to-port="0" />
+		<edge from-layer="1323" from-port="0" to-layer="1324" to-port="1" />
+		<edge from-layer="1324" from-port="2" to-layer="1325" to-port="0" />
+		<edge from-layer="1325" from-port="1" to-layer="1326" to-port="1" />
 		<edge from-layer="1326" from-port="2" to-layer="1327" to-port="1" />
 		<edge from-layer="1327" from-port="2" to-layer="1329" to-port="0" />
 		<edge from-layer="1327" from-port="2" to-layer="1333" to-port="0" />
@@ -27457,12 +27315,12 @@
 		<edge from-layer="1337" from-port="1" to-layer="1340" to-port="0" />
 		<edge from-layer="1338" from-port="0" to-layer="1339" to-port="0" />
 		<edge from-layer="1339" from-port="1" to-layer="1340" to-port="1" />
-		<edge from-layer="1340" from-port="2" to-layer="1343" to-port="0" />
-		<edge from-layer="1341" from-port="0" to-layer="1342" to-port="0" />
-		<edge from-layer="1342" from-port="1" to-layer="1343" to-port="1" />
-		<edge from-layer="1343" from-port="2" to-layer="1345" to-port="0" />
-		<edge from-layer="1344" from-port="0" to-layer="1345" to-port="1" />
-		<edge from-layer="1345" from-port="2" to-layer="1346" to-port="1" />
+		<edge from-layer="1340" from-port="2" to-layer="1342" to-port="0" />
+		<edge from-layer="1341" from-port="0" to-layer="1342" to-port="1" />
+		<edge from-layer="1342" from-port="2" to-layer="1344" to-port="0" />
+		<edge from-layer="1343" from-port="0" to-layer="1344" to-port="1" />
+		<edge from-layer="1344" from-port="2" to-layer="1345" to-port="0" />
+		<edge from-layer="1345" from-port="1" to-layer="1346" to-port="1" />
 		<edge from-layer="1346" from-port="2" to-layer="1347" to-port="1" />
 		<edge from-layer="1347" from-port="2" to-layer="1348" to-port="1" />
 		<edge from-layer="1348" from-port="2" to-layer="1349" to-port="0" />
@@ -27479,19 +27337,19 @@
 		<edge from-layer="1358" from-port="2" to-layer="1360" to-port="0" />
 		<edge from-layer="1359" from-port="0" to-layer="1360" to-port="1" />
 		<edge from-layer="1360" from-port="2" to-layer="1361" to-port="0" />
-		<edge from-layer="1361" from-port="1" to-layer="1413" to-port="0" />
 		<edge from-layer="1361" from-port="1" to-layer="1392" to-port="0" />
+		<edge from-layer="1361" from-port="1" to-layer="1413" to-port="0" />
 		<edge from-layer="1361" from-port="1" to-layer="1372" to-port="0" />
 		<edge from-layer="1362" from-port="0" to-layer="1363" to-port="0" />
 		<edge from-layer="1363" from-port="1" to-layer="1366" to-port="0" />
 		<edge from-layer="1364" from-port="0" to-layer="1365" to-port="0" />
 		<edge from-layer="1365" from-port="1" to-layer="1366" to-port="1" />
-		<edge from-layer="1366" from-port="2" to-layer="1369" to-port="0" />
-		<edge from-layer="1367" from-port="0" to-layer="1368" to-port="0" />
-		<edge from-layer="1368" from-port="1" to-layer="1369" to-port="1" />
-		<edge from-layer="1369" from-port="2" to-layer="1371" to-port="0" />
-		<edge from-layer="1370" from-port="0" to-layer="1371" to-port="1" />
-		<edge from-layer="1371" from-port="2" to-layer="1372" to-port="1" />
+		<edge from-layer="1366" from-port="2" to-layer="1368" to-port="0" />
+		<edge from-layer="1367" from-port="0" to-layer="1368" to-port="1" />
+		<edge from-layer="1368" from-port="2" to-layer="1370" to-port="0" />
+		<edge from-layer="1369" from-port="0" to-layer="1370" to-port="1" />
+		<edge from-layer="1370" from-port="2" to-layer="1371" to-port="0" />
+		<edge from-layer="1371" from-port="1" to-layer="1372" to-port="1" />
 		<edge from-layer="1372" from-port="2" to-layer="1373" to-port="1" />
 		<edge from-layer="1373" from-port="2" to-layer="1375" to-port="0" />
 		<edge from-layer="1374" from-port="0" to-layer="1375" to-port="1" />
@@ -27500,19 +27358,19 @@
 		<edge from-layer="1377" from-port="2" to-layer="1381" to-port="0" />
 		<edge from-layer="1378" from-port="0" to-layer="1380" to-port="0" />
 		<edge from-layer="1379" from-port="0" to-layer="1380" to-port="1" />
-		<edge from-layer="1380" from-port="2" to-layer="1381" to-port="1" />
 		<edge from-layer="1380" from-port="2" to-layer="1397" to-port="1" />
+		<edge from-layer="1380" from-port="2" to-layer="1381" to-port="1" />
 		<edge from-layer="1381" from-port="2" to-layer="1398" to-port="0" />
 		<edge from-layer="1382" from-port="0" to-layer="1383" to-port="0" />
 		<edge from-layer="1383" from-port="1" to-layer="1386" to-port="0" />
 		<edge from-layer="1384" from-port="0" to-layer="1385" to-port="0" />
 		<edge from-layer="1385" from-port="1" to-layer="1386" to-port="1" />
-		<edge from-layer="1386" from-port="2" to-layer="1389" to-port="0" />
-		<edge from-layer="1387" from-port="0" to-layer="1388" to-port="0" />
-		<edge from-layer="1388" from-port="1" to-layer="1389" to-port="1" />
-		<edge from-layer="1389" from-port="2" to-layer="1391" to-port="0" />
-		<edge from-layer="1390" from-port="0" to-layer="1391" to-port="1" />
-		<edge from-layer="1391" from-port="2" to-layer="1392" to-port="1" />
+		<edge from-layer="1386" from-port="2" to-layer="1388" to-port="0" />
+		<edge from-layer="1387" from-port="0" to-layer="1388" to-port="1" />
+		<edge from-layer="1388" from-port="2" to-layer="1390" to-port="0" />
+		<edge from-layer="1389" from-port="0" to-layer="1390" to-port="1" />
+		<edge from-layer="1390" from-port="2" to-layer="1391" to-port="0" />
+		<edge from-layer="1391" from-port="1" to-layer="1392" to-port="1" />
 		<edge from-layer="1392" from-port="2" to-layer="1394" to-port="0" />
 		<edge from-layer="1393" from-port="0" to-layer="1394" to-port="1" />
 		<edge from-layer="1394" from-port="2" to-layer="1396" to-port="0" />
@@ -27528,12 +27386,12 @@
 		<edge from-layer="1404" from-port="1" to-layer="1407" to-port="0" />
 		<edge from-layer="1405" from-port="0" to-layer="1406" to-port="0" />
 		<edge from-layer="1406" from-port="1" to-layer="1407" to-port="1" />
-		<edge from-layer="1407" from-port="2" to-layer="1410" to-port="0" />
-		<edge from-layer="1408" from-port="0" to-layer="1409" to-port="0" />
-		<edge from-layer="1409" from-port="1" to-layer="1410" to-port="1" />
-		<edge from-layer="1410" from-port="2" to-layer="1412" to-port="0" />
-		<edge from-layer="1411" from-port="0" to-layer="1412" to-port="1" />
-		<edge from-layer="1412" from-port="2" to-layer="1413" to-port="1" />
+		<edge from-layer="1407" from-port="2" to-layer="1409" to-port="0" />
+		<edge from-layer="1408" from-port="0" to-layer="1409" to-port="1" />
+		<edge from-layer="1409" from-port="2" to-layer="1411" to-port="0" />
+		<edge from-layer="1410" from-port="0" to-layer="1411" to-port="1" />
+		<edge from-layer="1411" from-port="2" to-layer="1412" to-port="0" />
+		<edge from-layer="1412" from-port="1" to-layer="1413" to-port="1" />
 		<edge from-layer="1413" from-port="2" to-layer="1414" to-port="1" />
 		<edge from-layer="1414" from-port="2" to-layer="1416" to-port="0" />
 		<edge from-layer="1415" from-port="0" to-layer="1416" to-port="1" />
@@ -27545,8 +27403,8 @@
 		<edge from-layer="1421" from-port="2" to-layer="1422" to-port="0" />
 		<edge from-layer="1421" from-port="2" to-layer="1432" to-port="0" />
 		<edge from-layer="1422" from-port="1" to-layer="1429" to-port="0" />
-		<edge from-layer="1423" from-port="0" to-layer="1426" to-port="0" />
 		<edge from-layer="1423" from-port="0" to-layer="1429" to-port="1" />
+		<edge from-layer="1423" from-port="0" to-layer="1426" to-port="0" />
 		<edge from-layer="1424" from-port="0" to-layer="1429" to-port="2" />
 		<edge from-layer="1425" from-port="0" to-layer="1427" to-port="0" />
 		<edge from-layer="1426" from-port="1" to-layer="1427" to-port="1" />
@@ -27560,12 +27418,12 @@
 		<edge from-layer="1434" from-port="1" to-layer="1437" to-port="0" />
 		<edge from-layer="1435" from-port="0" to-layer="1436" to-port="0" />
 		<edge from-layer="1436" from-port="1" to-layer="1437" to-port="1" />
-		<edge from-layer="1437" from-port="2" to-layer="1440" to-port="0" />
-		<edge from-layer="1438" from-port="0" to-layer="1439" to-port="0" />
-		<edge from-layer="1439" from-port="1" to-layer="1440" to-port="1" />
-		<edge from-layer="1440" from-port="2" to-layer="1442" to-port="0" />
-		<edge from-layer="1441" from-port="0" to-layer="1442" to-port="1" />
-		<edge from-layer="1442" from-port="2" to-layer="1443" to-port="1" />
+		<edge from-layer="1437" from-port="2" to-layer="1439" to-port="0" />
+		<edge from-layer="1438" from-port="0" to-layer="1439" to-port="1" />
+		<edge from-layer="1439" from-port="2" to-layer="1441" to-port="0" />
+		<edge from-layer="1440" from-port="0" to-layer="1441" to-port="1" />
+		<edge from-layer="1441" from-port="2" to-layer="1442" to-port="0" />
+		<edge from-layer="1442" from-port="1" to-layer="1443" to-port="1" />
 		<edge from-layer="1443" from-port="2" to-layer="1444" to-port="1" />
 		<edge from-layer="1444" from-port="2" to-layer="1445" to-port="1" />
 		<edge from-layer="1445" from-port="2" to-layer="1446" to-port="0" />
@@ -27587,12 +27445,12 @@
 		<edge from-layer="1460" from-port="1" to-layer="1463" to-port="0" />
 		<edge from-layer="1461" from-port="0" to-layer="1462" to-port="0" />
 		<edge from-layer="1462" from-port="1" to-layer="1463" to-port="1" />
-		<edge from-layer="1463" from-port="2" to-layer="1466" to-port="0" />
-		<edge from-layer="1464" from-port="0" to-layer="1465" to-port="0" />
-		<edge from-layer="1465" from-port="1" to-layer="1466" to-port="1" />
-		<edge from-layer="1466" from-port="2" to-layer="1468" to-port="0" />
-		<edge from-layer="1467" from-port="0" to-layer="1468" to-port="1" />
-		<edge from-layer="1468" from-port="2" to-layer="1469" to-port="1" />
+		<edge from-layer="1463" from-port="2" to-layer="1465" to-port="0" />
+		<edge from-layer="1464" from-port="0" to-layer="1465" to-port="1" />
+		<edge from-layer="1465" from-port="2" to-layer="1467" to-port="0" />
+		<edge from-layer="1466" from-port="0" to-layer="1467" to-port="1" />
+		<edge from-layer="1467" from-port="2" to-layer="1468" to-port="0" />
+		<edge from-layer="1468" from-port="1" to-layer="1469" to-port="1" />
 		<edge from-layer="1469" from-port="2" to-layer="1470" to-port="1" />
 		<edge from-layer="1470" from-port="2" to-layer="1472" to-port="0" />
 		<edge from-layer="1470" from-port="2" to-layer="1476" to-port="0" />
@@ -27608,12 +27466,12 @@
 		<edge from-layer="1480" from-port="1" to-layer="1483" to-port="0" />
 		<edge from-layer="1481" from-port="0" to-layer="1482" to-port="0" />
 		<edge from-layer="1482" from-port="1" to-layer="1483" to-port="1" />
-		<edge from-layer="1483" from-port="2" to-layer="1486" to-port="0" />
-		<edge from-layer="1484" from-port="0" to-layer="1485" to-port="0" />
-		<edge from-layer="1485" from-port="1" to-layer="1486" to-port="1" />
-		<edge from-layer="1486" from-port="2" to-layer="1488" to-port="0" />
-		<edge from-layer="1487" from-port="0" to-layer="1488" to-port="1" />
-		<edge from-layer="1488" from-port="2" to-layer="1489" to-port="1" />
+		<edge from-layer="1483" from-port="2" to-layer="1485" to-port="0" />
+		<edge from-layer="1484" from-port="0" to-layer="1485" to-port="1" />
+		<edge from-layer="1485" from-port="2" to-layer="1487" to-port="0" />
+		<edge from-layer="1486" from-port="0" to-layer="1487" to-port="1" />
+		<edge from-layer="1487" from-port="2" to-layer="1488" to-port="0" />
+		<edge from-layer="1488" from-port="1" to-layer="1489" to-port="1" />
 		<edge from-layer="1489" from-port="2" to-layer="1490" to-port="1" />
 		<edge from-layer="1490" from-port="2" to-layer="1491" to-port="1" />
 		<edge from-layer="1491" from-port="2" to-layer="1492" to-port="0" />
@@ -27637,12 +27495,12 @@
 		<edge from-layer="1506" from-port="1" to-layer="1509" to-port="0" />
 		<edge from-layer="1507" from-port="0" to-layer="1508" to-port="0" />
 		<edge from-layer="1508" from-port="1" to-layer="1509" to-port="1" />
-		<edge from-layer="1509" from-port="2" to-layer="1512" to-port="0" />
-		<edge from-layer="1510" from-port="0" to-layer="1511" to-port="0" />
-		<edge from-layer="1511" from-port="1" to-layer="1512" to-port="1" />
-		<edge from-layer="1512" from-port="2" to-layer="1514" to-port="0" />
-		<edge from-layer="1513" from-port="0" to-layer="1514" to-port="1" />
-		<edge from-layer="1514" from-port="2" to-layer="1515" to-port="1" />
+		<edge from-layer="1509" from-port="2" to-layer="1511" to-port="0" />
+		<edge from-layer="1510" from-port="0" to-layer="1511" to-port="1" />
+		<edge from-layer="1511" from-port="2" to-layer="1513" to-port="0" />
+		<edge from-layer="1512" from-port="0" to-layer="1513" to-port="1" />
+		<edge from-layer="1513" from-port="2" to-layer="1514" to-port="0" />
+		<edge from-layer="1514" from-port="1" to-layer="1515" to-port="1" />
 		<edge from-layer="1515" from-port="2" to-layer="1516" to-port="1" />
 		<edge from-layer="1516" from-port="2" to-layer="1518" to-port="0" />
 		<edge from-layer="1517" from-port="0" to-layer="1518" to-port="1" />
@@ -27658,12 +27516,12 @@
 		<edge from-layer="1526" from-port="1" to-layer="1529" to-port="0" />
 		<edge from-layer="1527" from-port="0" to-layer="1528" to-port="0" />
 		<edge from-layer="1528" from-port="1" to-layer="1529" to-port="1" />
-		<edge from-layer="1529" from-port="2" to-layer="1532" to-port="0" />
-		<edge from-layer="1530" from-port="0" to-layer="1531" to-port="0" />
-		<edge from-layer="1531" from-port="1" to-layer="1532" to-port="1" />
-		<edge from-layer="1532" from-port="2" to-layer="1534" to-port="0" />
-		<edge from-layer="1533" from-port="0" to-layer="1534" to-port="1" />
-		<edge from-layer="1534" from-port="2" to-layer="1535" to-port="1" />
+		<edge from-layer="1529" from-port="2" to-layer="1531" to-port="0" />
+		<edge from-layer="1530" from-port="0" to-layer="1531" to-port="1" />
+		<edge from-layer="1531" from-port="2" to-layer="1533" to-port="0" />
+		<edge from-layer="1532" from-port="0" to-layer="1533" to-port="1" />
+		<edge from-layer="1533" from-port="2" to-layer="1534" to-port="0" />
+		<edge from-layer="1534" from-port="1" to-layer="1535" to-port="1" />
 		<edge from-layer="1535" from-port="2" to-layer="1537" to-port="0" />
 		<edge from-layer="1536" from-port="0" to-layer="1537" to-port="1" />
 		<edge from-layer="1537" from-port="2" to-layer="1539" to-port="0" />
@@ -27679,12 +27537,12 @@
 		<edge from-layer="1547" from-port="1" to-layer="1550" to-port="0" />
 		<edge from-layer="1548" from-port="0" to-layer="1549" to-port="0" />
 		<edge from-layer="1549" from-port="1" to-layer="1550" to-port="1" />
-		<edge from-layer="1550" from-port="2" to-layer="1553" to-port="0" />
-		<edge from-layer="1551" from-port="0" to-layer="1552" to-port="0" />
-		<edge from-layer="1552" from-port="1" to-layer="1553" to-port="1" />
-		<edge from-layer="1553" from-port="2" to-layer="1555" to-port="0" />
-		<edge from-layer="1554" from-port="0" to-layer="1555" to-port="1" />
-		<edge from-layer="1555" from-port="2" to-layer="1556" to-port="1" />
+		<edge from-layer="1550" from-port="2" to-layer="1552" to-port="0" />
+		<edge from-layer="1551" from-port="0" to-layer="1552" to-port="1" />
+		<edge from-layer="1552" from-port="2" to-layer="1554" to-port="0" />
+		<edge from-layer="1553" from-port="0" to-layer="1554" to-port="1" />
+		<edge from-layer="1554" from-port="2" to-layer="1555" to-port="0" />
+		<edge from-layer="1555" from-port="1" to-layer="1556" to-port="1" />
 		<edge from-layer="1556" from-port="2" to-layer="1557" to-port="1" />
 		<edge from-layer="1557" from-port="2" to-layer="1559" to-port="0" />
 		<edge from-layer="1558" from-port="0" to-layer="1559" to-port="1" />
@@ -27693,11 +27551,11 @@
 		<edge from-layer="1561" from-port="2" to-layer="1562" to-port="1" />
 		<edge from-layer="1562" from-port="2" to-layer="1564" to-port="0" />
 		<edge from-layer="1563" from-port="0" to-layer="1564" to-port="1" />
-		<edge from-layer="1564" from-port="2" to-layer="1575" to-port="0" />
 		<edge from-layer="1564" from-port="2" to-layer="1565" to-port="0" />
+		<edge from-layer="1564" from-port="2" to-layer="1575" to-port="0" />
 		<edge from-layer="1565" from-port="1" to-layer="1572" to-port="0" />
-		<edge from-layer="1566" from-port="0" to-layer="1572" to-port="1" />
 		<edge from-layer="1566" from-port="0" to-layer="1569" to-port="0" />
+		<edge from-layer="1566" from-port="0" to-layer="1572" to-port="1" />
 		<edge from-layer="1567" from-port="0" to-layer="1572" to-port="2" />
 		<edge from-layer="1568" from-port="0" to-layer="1570" to-port="0" />
 		<edge from-layer="1569" from-port="1" to-layer="1570" to-port="1" />
@@ -27711,12 +27569,12 @@
 		<edge from-layer="1577" from-port="1" to-layer="1580" to-port="0" />
 		<edge from-layer="1578" from-port="0" to-layer="1579" to-port="0" />
 		<edge from-layer="1579" from-port="1" to-layer="1580" to-port="1" />
-		<edge from-layer="1580" from-port="2" to-layer="1583" to-port="0" />
-		<edge from-layer="1581" from-port="0" to-layer="1582" to-port="0" />
-		<edge from-layer="1582" from-port="1" to-layer="1583" to-port="1" />
-		<edge from-layer="1583" from-port="2" to-layer="1585" to-port="0" />
-		<edge from-layer="1584" from-port="0" to-layer="1585" to-port="1" />
-		<edge from-layer="1585" from-port="2" to-layer="1586" to-port="1" />
+		<edge from-layer="1580" from-port="2" to-layer="1582" to-port="0" />
+		<edge from-layer="1581" from-port="0" to-layer="1582" to-port="1" />
+		<edge from-layer="1582" from-port="2" to-layer="1584" to-port="0" />
+		<edge from-layer="1583" from-port="0" to-layer="1584" to-port="1" />
+		<edge from-layer="1584" from-port="2" to-layer="1585" to-port="0" />
+		<edge from-layer="1585" from-port="1" to-layer="1586" to-port="1" />
 		<edge from-layer="1586" from-port="2" to-layer="1587" to-port="1" />
 		<edge from-layer="1587" from-port="2" to-layer="1588" to-port="1" />
 		<edge from-layer="1588" from-port="2" to-layer="1589" to-port="0" />
@@ -27738,12 +27596,12 @@
 		<edge from-layer="1603" from-port="1" to-layer="1606" to-port="0" />
 		<edge from-layer="1604" from-port="0" to-layer="1605" to-port="0" />
 		<edge from-layer="1605" from-port="1" to-layer="1606" to-port="1" />
-		<edge from-layer="1606" from-port="2" to-layer="1609" to-port="0" />
-		<edge from-layer="1607" from-port="0" to-layer="1608" to-port="0" />
-		<edge from-layer="1608" from-port="1" to-layer="1609" to-port="1" />
-		<edge from-layer="1609" from-port="2" to-layer="1611" to-port="0" />
-		<edge from-layer="1610" from-port="0" to-layer="1611" to-port="1" />
-		<edge from-layer="1611" from-port="2" to-layer="1612" to-port="1" />
+		<edge from-layer="1606" from-port="2" to-layer="1608" to-port="0" />
+		<edge from-layer="1607" from-port="0" to-layer="1608" to-port="1" />
+		<edge from-layer="1608" from-port="2" to-layer="1610" to-port="0" />
+		<edge from-layer="1609" from-port="0" to-layer="1610" to-port="1" />
+		<edge from-layer="1610" from-port="2" to-layer="1611" to-port="0" />
+		<edge from-layer="1611" from-port="1" to-layer="1612" to-port="1" />
 		<edge from-layer="1612" from-port="2" to-layer="1613" to-port="1" />
 		<edge from-layer="1613" from-port="2" to-layer="1615" to-port="0" />
 		<edge from-layer="1613" from-port="2" to-layer="1619" to-port="0" />
@@ -27759,12 +27617,12 @@
 		<edge from-layer="1623" from-port="1" to-layer="1626" to-port="0" />
 		<edge from-layer="1624" from-port="0" to-layer="1625" to-port="0" />
 		<edge from-layer="1625" from-port="1" to-layer="1626" to-port="1" />
-		<edge from-layer="1626" from-port="2" to-layer="1629" to-port="0" />
-		<edge from-layer="1627" from-port="0" to-layer="1628" to-port="0" />
-		<edge from-layer="1628" from-port="1" to-layer="1629" to-port="1" />
-		<edge from-layer="1629" from-port="2" to-layer="1631" to-port="0" />
-		<edge from-layer="1630" from-port="0" to-layer="1631" to-port="1" />
-		<edge from-layer="1631" from-port="2" to-layer="1632" to-port="1" />
+		<edge from-layer="1626" from-port="2" to-layer="1628" to-port="0" />
+		<edge from-layer="1627" from-port="0" to-layer="1628" to-port="1" />
+		<edge from-layer="1628" from-port="2" to-layer="1630" to-port="0" />
+		<edge from-layer="1629" from-port="0" to-layer="1630" to-port="1" />
+		<edge from-layer="1630" from-port="2" to-layer="1631" to-port="0" />
+		<edge from-layer="1631" from-port="1" to-layer="1632" to-port="1" />
 		<edge from-layer="1632" from-port="2" to-layer="1633" to-port="1" />
 		<edge from-layer="1633" from-port="2" to-layer="1634" to-port="1" />
 		<edge from-layer="1634" from-port="2" to-layer="1635" to-port="0" />
@@ -27788,12 +27646,12 @@
 		<edge from-layer="1649" from-port="1" to-layer="1652" to-port="0" />
 		<edge from-layer="1650" from-port="0" to-layer="1651" to-port="0" />
 		<edge from-layer="1651" from-port="1" to-layer="1652" to-port="1" />
-		<edge from-layer="1652" from-port="2" to-layer="1655" to-port="0" />
-		<edge from-layer="1653" from-port="0" to-layer="1654" to-port="0" />
-		<edge from-layer="1654" from-port="1" to-layer="1655" to-port="1" />
-		<edge from-layer="1655" from-port="2" to-layer="1657" to-port="0" />
-		<edge from-layer="1656" from-port="0" to-layer="1657" to-port="1" />
-		<edge from-layer="1657" from-port="2" to-layer="1658" to-port="1" />
+		<edge from-layer="1652" from-port="2" to-layer="1654" to-port="0" />
+		<edge from-layer="1653" from-port="0" to-layer="1654" to-port="1" />
+		<edge from-layer="1654" from-port="2" to-layer="1656" to-port="0" />
+		<edge from-layer="1655" from-port="0" to-layer="1656" to-port="1" />
+		<edge from-layer="1656" from-port="2" to-layer="1657" to-port="0" />
+		<edge from-layer="1657" from-port="1" to-layer="1658" to-port="1" />
 		<edge from-layer="1658" from-port="2" to-layer="1659" to-port="1" />
 		<edge from-layer="1659" from-port="2" to-layer="1661" to-port="0" />
 		<edge from-layer="1660" from-port="0" to-layer="1661" to-port="1" />
@@ -27809,12 +27667,12 @@
 		<edge from-layer="1669" from-port="1" to-layer="1672" to-port="0" />
 		<edge from-layer="1670" from-port="0" to-layer="1671" to-port="0" />
 		<edge from-layer="1671" from-port="1" to-layer="1672" to-port="1" />
-		<edge from-layer="1672" from-port="2" to-layer="1675" to-port="0" />
-		<edge from-layer="1673" from-port="0" to-layer="1674" to-port="0" />
-		<edge from-layer="1674" from-port="1" to-layer="1675" to-port="1" />
-		<edge from-layer="1675" from-port="2" to-layer="1677" to-port="0" />
-		<edge from-layer="1676" from-port="0" to-layer="1677" to-port="1" />
-		<edge from-layer="1677" from-port="2" to-layer="1678" to-port="1" />
+		<edge from-layer="1672" from-port="2" to-layer="1674" to-port="0" />
+		<edge from-layer="1673" from-port="0" to-layer="1674" to-port="1" />
+		<edge from-layer="1674" from-port="2" to-layer="1676" to-port="0" />
+		<edge from-layer="1675" from-port="0" to-layer="1676" to-port="1" />
+		<edge from-layer="1676" from-port="2" to-layer="1677" to-port="0" />
+		<edge from-layer="1677" from-port="1" to-layer="1678" to-port="1" />
 		<edge from-layer="1678" from-port="2" to-layer="1680" to-port="0" />
 		<edge from-layer="1679" from-port="0" to-layer="1680" to-port="1" />
 		<edge from-layer="1680" from-port="2" to-layer="1682" to-port="0" />
@@ -27830,12 +27688,12 @@
 		<edge from-layer="1690" from-port="1" to-layer="1693" to-port="0" />
 		<edge from-layer="1691" from-port="0" to-layer="1692" to-port="0" />
 		<edge from-layer="1692" from-port="1" to-layer="1693" to-port="1" />
-		<edge from-layer="1693" from-port="2" to-layer="1696" to-port="0" />
-		<edge from-layer="1694" from-port="0" to-layer="1695" to-port="0" />
-		<edge from-layer="1695" from-port="1" to-layer="1696" to-port="1" />
-		<edge from-layer="1696" from-port="2" to-layer="1698" to-port="0" />
-		<edge from-layer="1697" from-port="0" to-layer="1698" to-port="1" />
-		<edge from-layer="1698" from-port="2" to-layer="1699" to-port="1" />
+		<edge from-layer="1693" from-port="2" to-layer="1695" to-port="0" />
+		<edge from-layer="1694" from-port="0" to-layer="1695" to-port="1" />
+		<edge from-layer="1695" from-port="2" to-layer="1697" to-port="0" />
+		<edge from-layer="1696" from-port="0" to-layer="1697" to-port="1" />
+		<edge from-layer="1697" from-port="2" to-layer="1698" to-port="0" />
+		<edge from-layer="1698" from-port="1" to-layer="1699" to-port="1" />
 		<edge from-layer="1699" from-port="2" to-layer="1700" to-port="1" />
 		<edge from-layer="1700" from-port="2" to-layer="1702" to-port="0" />
 		<edge from-layer="1701" from-port="0" to-layer="1702" to-port="1" />
@@ -27862,12 +27720,12 @@
 		<edge from-layer="1720" from-port="1" to-layer="1723" to-port="0" />
 		<edge from-layer="1721" from-port="0" to-layer="1722" to-port="0" />
 		<edge from-layer="1722" from-port="1" to-layer="1723" to-port="1" />
-		<edge from-layer="1723" from-port="2" to-layer="1726" to-port="0" />
-		<edge from-layer="1724" from-port="0" to-layer="1725" to-port="0" />
-		<edge from-layer="1725" from-port="1" to-layer="1726" to-port="1" />
-		<edge from-layer="1726" from-port="2" to-layer="1728" to-port="0" />
-		<edge from-layer="1727" from-port="0" to-layer="1728" to-port="1" />
-		<edge from-layer="1728" from-port="2" to-layer="1729" to-port="1" />
+		<edge from-layer="1723" from-port="2" to-layer="1725" to-port="0" />
+		<edge from-layer="1724" from-port="0" to-layer="1725" to-port="1" />
+		<edge from-layer="1725" from-port="2" to-layer="1727" to-port="0" />
+		<edge from-layer="1726" from-port="0" to-layer="1727" to-port="1" />
+		<edge from-layer="1727" from-port="2" to-layer="1728" to-port="0" />
+		<edge from-layer="1728" from-port="1" to-layer="1729" to-port="1" />
 		<edge from-layer="1729" from-port="2" to-layer="1730" to-port="1" />
 		<edge from-layer="1730" from-port="2" to-layer="1731" to-port="1" />
 		<edge from-layer="1731" from-port="2" to-layer="1732" to-port="0" />
@@ -27889,12 +27747,12 @@
 		<edge from-layer="1746" from-port="1" to-layer="1749" to-port="0" />
 		<edge from-layer="1747" from-port="0" to-layer="1748" to-port="0" />
 		<edge from-layer="1748" from-port="1" to-layer="1749" to-port="1" />
-		<edge from-layer="1749" from-port="2" to-layer="1752" to-port="0" />
-		<edge from-layer="1750" from-port="0" to-layer="1751" to-port="0" />
-		<edge from-layer="1751" from-port="1" to-layer="1752" to-port="1" />
-		<edge from-layer="1752" from-port="2" to-layer="1754" to-port="0" />
-		<edge from-layer="1753" from-port="0" to-layer="1754" to-port="1" />
-		<edge from-layer="1754" from-port="2" to-layer="1755" to-port="1" />
+		<edge from-layer="1749" from-port="2" to-layer="1751" to-port="0" />
+		<edge from-layer="1750" from-port="0" to-layer="1751" to-port="1" />
+		<edge from-layer="1751" from-port="2" to-layer="1753" to-port="0" />
+		<edge from-layer="1752" from-port="0" to-layer="1753" to-port="1" />
+		<edge from-layer="1753" from-port="2" to-layer="1754" to-port="0" />
+		<edge from-layer="1754" from-port="1" to-layer="1755" to-port="1" />
 		<edge from-layer="1755" from-port="2" to-layer="1756" to-port="1" />
 		<edge from-layer="1756" from-port="2" to-layer="1758" to-port="0" />
 		<edge from-layer="1756" from-port="2" to-layer="1762" to-port="0" />
@@ -27910,10 +27768,10 @@
 		<edge from-layer="1766" from-port="1" to-layer="1769" to-port="0" />
 		<edge from-layer="1767" from-port="0" to-layer="1768" to-port="0" />
 		<edge from-layer="1768" from-port="1" to-layer="1769" to-port="1" />
-		<edge from-layer="1769" from-port="2" to-layer="1772" to-port="0" />
-		<edge from-layer="1770" from-port="0" to-layer="1771" to-port="0" />
-		<edge from-layer="1771" from-port="1" to-layer="1772" to-port="1" />
-		<edge from-layer="1772" from-port="2" to-layer="1773" to-port="1" />
+		<edge from-layer="1769" from-port="2" to-layer="1771" to-port="0" />
+		<edge from-layer="1770" from-port="0" to-layer="1771" to-port="1" />
+		<edge from-layer="1771" from-port="2" to-layer="1772" to-port="0" />
+		<edge from-layer="1772" from-port="1" to-layer="1773" to-port="1" />
 		<edge from-layer="1773" from-port="2" to-layer="1774" to-port="1" />
 		<edge from-layer="1774" from-port="2" to-layer="1775" to-port="1" />
 		<edge from-layer="1775" from-port="2" to-layer="1776" to-port="0" />
@@ -27935,16 +27793,12 @@
 			<weight_compression>
 				<all_layers value="False" />
 				<awq value="False" />
+				<gptq value="False" />
 				<group_size value="128" />
-				<ignored_scope>
-					<names value="[]" />
-					<patterns value="[]" />
-					<subgraphs value="[]" />
-					<types value="[]" />
-					<validate value="True" />
-				</ignored_scope>
+				<ignored_scope value="[]" />
 				<mode value="int4_sym" />
 				<ratio value="1" />
+				<scale_estimation value="False" />
 				<sensitivity_metric value="weight_quantization_error" />
 			</weight_compression>
 		</nncf>