diff --git "a/multilang/ggml-base-encoder-openvino-int4/ggml-base-encoder-openvino.xml" "b/multilang/ggml-base-encoder-openvino-int4/ggml-base-encoder-openvino.xml"
--- "a/multilang/ggml-base-encoder-openvino-int4/ggml-base-encoder-openvino.xml"
+++ "b/multilang/ggml-base-encoder-openvino-int4/ggml-base-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_649" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>512</dim>
@@ -31,7 +31,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>512</dim>
 					<dim>80</dim>
 					<dim>3</dim>
@@ -49,7 +49,7 @@
 			</output>
 		</layer>
 		<layer id="4" name="Convert_652" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>512</dim>
@@ -58,29 +58,29 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>512</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="5" name="Subtract_653" type="Subtract" version="opset1">
+		<layer id="5" name="onnx::Conv_752/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>512</dim>
 					<dim>80</dim>
 					<dim>3</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>512</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>512</dim>
 					<dim>80</dim>
 					<dim>3</dim>
@@ -97,39 +97,39 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="7" name="onnx::Conv_752/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="7" name="onnx::Conv_752/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>80</dim>
+					<dim>3</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>512</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>512</dim>
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>80</dim>
+					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="8" name="onnx::Conv_752/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="8" name="onnx::Conv_752/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>512</dim>
 					<dim>80</dim>
 					<dim>3</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>512</dim>
-					<dim>1</dim>
-					<dim>1</dim>
-				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32">
 					<dim>512</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>512</dim>
@@ -154,7 +154,7 @@
 				<port id="2" precision="FP32">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -193,7 +193,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</input>
 			<output>
@@ -298,7 +298,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</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>512</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -326,7 +326,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</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>512</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -343,14 +343,14 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/Erf_output_0">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -366,7 +366,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</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>512</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -384,19 +384,19 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/Mul_output_0">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -412,7 +412,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</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>512</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -435,7 +435,7 @@
 			</output>
 		</layer>
 		<layer id="30" name="Convert_660" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>512</dim>
@@ -444,7 +444,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>512</dim>
 					<dim>512</dim>
 					<dim>3</dim>
@@ -462,7 +462,7 @@
 			</output>
 		</layer>
 		<layer id="32" name="Convert_663" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>512</dim>
@@ -471,29 +471,29 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>512</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="33" name="Subtract_664" type="Subtract" version="opset1">
+		<layer id="33" name="onnx::Conv_754/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>512</dim>
 					<dim>512</dim>
 					<dim>3</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>512</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>512</dim>
 					<dim>512</dim>
 					<dim>3</dim>
@@ -510,39 +510,39 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="35" name="onnx::Conv_754/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="35" name="onnx::Conv_754/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+					<dim>3</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>512</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>512</dim>
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>512</dim>
+					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="36" name="onnx::Conv_754/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="36" name="onnx::Conv_754/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>512</dim>
 					<dim>512</dim>
 					<dim>3</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>512</dim>
-					<dim>1</dim>
-					<dim>1</dim>
-				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 					<dim>3</dim>
@@ -555,7 +555,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>512</dim>
@@ -567,7 +567,7 @@
 				<port id="2" precision="FP32">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -606,7 +606,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
@@ -711,7 +711,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</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>512</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -739,7 +739,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</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>512</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -756,14 +756,14 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/Erf_1_output_0">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -779,7 +779,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</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>512</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -797,19 +797,19 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/Mul_2_output_0">
 					<dim>1</dim>
 					<dim>512</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -825,7 +825,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</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>512</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -850,7 +850,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="59" name="onnx::Add_760" type="Const" version="opset1">
-			<data element_type="f32" shape="1000, 512" offset="916532" size="2048000" />
+			<data element_type="f32" shape="1500, 512" offset="916532" size="3072000" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_760">
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 			</output>
@@ -878,18 +878,18 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/Add_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 			</output>
@@ -899,14 +899,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 			</output>
@@ -916,20 +916,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="63" name="onnx::Add_785" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="2964532" size="2048" />
+			<data element_type="f32" shape="512" offset="3988532" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_785">
 					<dim>512</dim>
@@ -937,7 +937,7 @@
 			</output>
 		</layer>
 		<layer id="64" name="onnx::Add_762" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="2966580" size="2048" />
+			<data element_type="f32" shape="512" offset="3990580" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_762">
 					<dim>512</dim>
@@ -945,13 +945,13 @@
 			</output>
 		</layer>
 		<layer id="65" name="Constant_154" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="66" name="Constant_153" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" 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>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="70" name="blocks.0.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="2968644" size="2048" />
+			<data element_type="f32" shape="512" offset="3992644" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.0.attn_ln.weight">
 					<dim>512</dim>
@@ -1008,7 +1008,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="72" name="blocks.0.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="2970692" size="2048" />
+			<data element_type="f32" shape="512" offset="3994692" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.0.attn_ln.bias">
 					<dim>512</dim>
@@ -1036,7 +1036,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -1056,20 +1056,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="75" name="onnx::MatMul_763" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="2972740" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="3996740" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -1079,7 +1079,7 @@
 			</output>
 		</layer>
 		<layer id="76" name="Convert_692" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -1088,7 +1088,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -1096,7 +1096,7 @@
 			</output>
 		</layer>
 		<layer id="77" name="onnx::MatMul_763/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4127812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -1104,32 +1104,32 @@
 			</output>
 		</layer>
 		<layer id="78" name="Convert_694" 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_695" type="Subtract" version="opset1">
+		<layer id="79" name="onnx::MatMul_763/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -1137,7 +1137,7 @@
 			</output>
 		</layer>
 		<layer id="80" name="onnx::MatMul_763/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="3103813" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="4127813" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -1146,57 +1146,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="81" name="onnx::MatMul_763/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="82" name="onnx::MatMul_763/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="81" name="onnx::MatMul_763/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="83" name="Constant_700" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="82" name="Constant_699" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="84" name="Reshape_701" type="Reshape" version="opset1">
+		<layer id="83" name="Reshape_700" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -1206,7 +1189,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="84" name="onnx::MatMul_763/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -1217,7 +1215,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -1241,20 +1239,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="87" name="/blocks.0/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="89" name="Constant_170" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" 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>8</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>8</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="3107989" size="4" />
+			<data element_type="f32" shape="" offset="4131989" 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="3107993" size="4" />
+			<data element_type="f32" shape="" offset="4131993" 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>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="95" name="onnx::MatMul_765" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="3107997" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="4131997" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -1364,7 +1362,7 @@
 			</output>
 		</layer>
 		<layer id="96" name="Convert_681" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -1373,7 +1371,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -1381,7 +1379,7 @@
 			</output>
 		</layer>
 		<layer id="97" name="onnx::MatMul_765/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4263069" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -1389,32 +1387,32 @@
 			</output>
 		</layer>
 		<layer id="98" name="Convert_683" 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_684" type="Subtract" version="opset1">
+		<layer id="99" name="onnx::MatMul_765/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -1422,7 +1420,7 @@
 			</output>
 		</layer>
 		<layer id="100" name="onnx::MatMul_765/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="3239069" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="4263070" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -1431,57 +1429,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="101" name="onnx::MatMul_765/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="102" name="onnx::MatMul_765/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="101" name="onnx::MatMul_765/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="103" name="Constant_689" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="102" name="Constant_688" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="104" name="Reshape_690" type="Reshape" version="opset1">
+		<layer id="103" name="Reshape_689" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -1491,7 +1472,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="104" name="onnx::MatMul_765/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -1502,7 +1498,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="106" name="/blocks.0/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="108" name="Constant_175" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3243165" size="32" />
+			<data element_type="i64" shape="4" offset="4267166" 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>8</dim>
 					<dim>64</dim>
 				</port>
@@ -1572,7 +1568,7 @@
 					<dim>1</dim>
 					<dim>8</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -1583,7 +1579,7 @@
 					<dim>1</dim>
 					<dim>8</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>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</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>8</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>8</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>8</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>8</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>8</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>8</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="115" name="onnx::Add_767" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="3243197" size="2048" />
+			<data element_type="f32" shape="512" offset="4267198" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_767">
 					<dim>512</dim>
@@ -1687,7 +1683,7 @@
 			</output>
 		</layer>
 		<layer id="116" name="onnx::MatMul_768" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="3245245" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="4269246" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -1697,7 +1693,7 @@
 			</output>
 		</layer>
 		<layer id="117" name="Convert_670" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -1706,7 +1702,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -1714,7 +1710,7 @@
 			</output>
 		</layer>
 		<layer id="118" name="onnx::MatMul_768/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4400318" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -1722,32 +1718,32 @@
 			</output>
 		</layer>
 		<layer id="119" name="Convert_672" 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_673" type="Subtract" version="opset1">
+		<layer id="120" name="onnx::MatMul_768/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -1755,7 +1751,7 @@
 			</output>
 		</layer>
 		<layer id="121" name="onnx::MatMul_768/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="3376317" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="4400319" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -1764,57 +1760,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="122" name="onnx::MatMul_768/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="123" name="onnx::MatMul_768/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="122" name="onnx::MatMul_768/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="124" name="Constant_678" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="123" name="Constant_677" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="125" name="Reshape_679" type="Reshape" version="opset1">
+		<layer id="124" name="Reshape_678" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -1824,7 +1803,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="125" name="onnx::MatMul_768/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -1835,7 +1829,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -1859,20 +1853,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="128" name="/blocks.0/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="130" name="Constant_180" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" 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>8</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>8</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>8</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="133" name="Constant_187" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -1967,7 +1961,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</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>8</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="2968636" size="8" />
+			<data element_type="i64" shape="1" offset="3992636" 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="2968636" size="8" />
+			<data element_type="i64" shape="1" offset="3992636" 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="2968628" size="8" />
+			<data element_type="i64" shape="1" offset="3992628" 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>8</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="146" name="onnx::MatMul_786" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="3380413" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="4404415" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -2140,7 +2134,7 @@
 			</output>
 		</layer>
 		<layer id="147" name="Convert_703" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -2149,7 +2143,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -2157,7 +2151,7 @@
 			</output>
 		</layer>
 		<layer id="148" name="onnx::MatMul_786/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4263069" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -2165,32 +2159,32 @@
 			</output>
 		</layer>
 		<layer id="149" name="Convert_705" 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_706" type="Subtract" version="opset1">
+		<layer id="150" name="onnx::MatMul_786/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -2198,7 +2192,7 @@
 			</output>
 		</layer>
 		<layer id="151" name="onnx::MatMul_786/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="3511485" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="4535487" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -2207,57 +2201,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="152" name="onnx::MatMul_786/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="153" name="onnx::MatMul_786/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="152" name="onnx::MatMul_786/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="154" name="Constant_711" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="153" name="Constant_710" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="155" name="Reshape_712" type="Reshape" version="opset1">
+		<layer id="154" name="Reshape_711" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -2267,7 +2244,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="155" name="onnx::MatMul_786/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -2278,7 +2270,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -2302,14 +2294,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -2319,19 +2311,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -2341,20 +2333,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="160" name="onnx::Add_791" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="3515581" size="2048" />
+			<data element_type="f32" shape="512" offset="4539583" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_791">
 					<dim>512</dim>
@@ -2362,7 +2354,7 @@
 			</output>
 		</layer>
 		<layer id="161" name="onnx::Add_788" type="Const" version="opset1">
-			<data element_type="f32" shape="2048" offset="3517629" size="8192" />
+			<data element_type="f32" shape="2048" offset="4541631" size="8192" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_788">
 					<dim>2048</dim>
@@ -2370,13 +2362,13 @@
 			</output>
 		</layer>
 		<layer id="162" name="Constant_213" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="163" name="Constant_212" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" 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>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="167" name="blocks.0.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="3525821" size="2048" />
+			<data element_type="f32" shape="512" offset="4549823" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.0.mlp_ln.weight">
 					<dim>512</dim>
@@ -2433,7 +2425,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="169" name="blocks.0.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="3527869" size="2048" />
+			<data element_type="f32" shape="512" offset="4551871" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.0.mlp_ln.bias">
 					<dim>512</dim>
@@ -2461,7 +2453,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -2481,20 +2473,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="172" name="onnx::MatMul_789" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 2048" offset="3529917" size="524288" />
+			<data element_type="u4" shape="4, 128, 2048" offset="4553919" size="524288" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -2504,7 +2496,7 @@
 			</output>
 		</layer>
 		<layer id="173" name="Convert_714" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -2513,7 +2505,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -2521,7 +2513,7 @@
 			</output>
 		</layer>
 		<layer id="174" name="onnx::MatMul_789/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5078207" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -2529,32 +2521,32 @@
 			</output>
 		</layer>
 		<layer id="175" name="Convert_716" 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_717" type="Subtract" version="opset1">
+		<layer id="176" name="onnx::MatMul_789/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</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>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -2562,7 +2554,7 @@
 			</output>
 		</layer>
 		<layer id="177" name="onnx::MatMul_789/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 2048" offset="4054205" size="16384" />
+			<data element_type="f16" shape="4, 1, 2048" offset="5078208" size="16384" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -2571,57 +2563,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="178" name="onnx::MatMul_789/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="178" name="onnx::MatMul_789/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
-					<dim>1</dim>
+					<dim>128</dim>
 					<dim>2048</dim>
 				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>2048</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="179" name="onnx::MatMul_789/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>4</dim>
-					<dim>128</dim>
-					<dim>2048</dim>
-				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>2048</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="180" name="Constant_722" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="4070589" size="16" />
+		<layer id="179" name="Constant_721" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="5094592" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="181" name="Reshape_723" type="Reshape" version="opset1">
+		<layer id="180" name="Reshape_722" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -2631,7 +2606,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>2048</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="181" name="onnx::MatMul_789/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>2048</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>2048</dim>
 				</port>
@@ -2642,7 +2632,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>2048</dim>
 				</port>
 			</output>
@@ -2666,14 +2656,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -2689,7 +2679,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -2706,14 +2696,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -2729,7 +2719,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -2747,19 +2737,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -2775,7 +2765,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="192" name="onnx::MatMul_792" type="Const" version="opset1">
-			<data element_type="u4" shape="16, 128, 512" offset="4070605" size="524288" />
+			<data element_type="u4" shape="16, 128, 512" offset="5094608" size="524288" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>16</dim>
@@ -2799,7 +2789,7 @@
 			</output>
 		</layer>
 		<layer id="193" name="Convert_725" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>16</dim>
@@ -2808,7 +2798,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -2816,7 +2806,7 @@
 			</output>
 		</layer>
 		<layer id="194" name="onnx::MatMul_792/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4400318" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -2824,32 +2814,32 @@
 			</output>
 		</layer>
 		<layer id="195" name="Convert_727" 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_728" type="Subtract" version="opset1">
+		<layer id="196" name="onnx::MatMul_792/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</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>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -2857,7 +2847,7 @@
 			</output>
 		</layer>
 		<layer id="197" name="onnx::MatMul_792/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="16, 1, 512" offset="4594893" size="16384" />
+			<data element_type="f16" shape="16, 1, 512" offset="5618896" size="16384" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>16</dim>
@@ -2866,57 +2856,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="198" name="onnx::MatMul_792/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>16</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>16</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="199" name="onnx::MatMul_792/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="198" name="onnx::MatMul_792/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>16</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="200" name="Constant_733" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="4611277" size="16" />
+		<layer id="199" name="Constant_732" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="5635280" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="201" name="Reshape_734" type="Reshape" version="opset1">
+		<layer id="200" name="Reshape_733" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -2926,7 +2899,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>2048</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="201" name="onnx::MatMul_792/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>2048</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>2048</dim>
 					<dim>512</dim>
 				</port>
@@ -2937,7 +2925,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>512</dim>
 				</port>
 			</output>
@@ -2961,14 +2949,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -2978,19 +2966,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -3000,20 +2988,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="206" name="onnx::Add_817" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="4611293" size="2048" />
+			<data element_type="f32" shape="512" offset="5635296" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_817">
 					<dim>512</dim>
@@ -3021,7 +3009,7 @@
 			</output>
 		</layer>
 		<layer id="207" name="onnx::Add_794" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="4613341" size="2048" />
+			<data element_type="f32" shape="512" offset="5637344" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_794">
 					<dim>512</dim>
@@ -3029,13 +3017,13 @@
 			</output>
 		</layer>
 		<layer id="208" name="Constant_235" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="209" name="Constant_234" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" 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>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="213" name="blocks.1.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="4615389" size="2048" />
+			<data element_type="f32" shape="512" offset="5639392" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.1.attn_ln.weight">
 					<dim>512</dim>
@@ -3092,7 +3080,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="215" name="blocks.1.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="4617437" size="2048" />
+			<data element_type="f32" shape="512" offset="5641440" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.1.attn_ln.bias">
 					<dim>512</dim>
@@ -3120,7 +3108,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -3140,20 +3128,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="218" name="onnx::MatMul_795" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="4619485" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="5643488" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -3163,7 +3151,7 @@
 			</output>
 		</layer>
 		<layer id="219" name="Convert_758" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -3172,7 +3160,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -3180,7 +3168,7 @@
 			</output>
 		</layer>
 		<layer id="220" name="onnx::MatMul_795/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5774560" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -3188,32 +3176,32 @@
 			</output>
 		</layer>
 		<layer id="221" name="Convert_760" 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_761" type="Subtract" version="opset1">
+		<layer id="222" name="onnx::MatMul_795/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -3221,7 +3209,7 @@
 			</output>
 		</layer>
 		<layer id="223" name="onnx::MatMul_795/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="4750557" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="5774561" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -3230,57 +3218,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="224" name="onnx::MatMul_795/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="225" name="onnx::MatMul_795/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="224" name="onnx::MatMul_795/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="226" name="Constant_766" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="225" name="Constant_765" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="227" name="Reshape_767" type="Reshape" version="opset1">
+		<layer id="226" name="Reshape_766" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -3290,7 +3261,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="227" name="onnx::MatMul_795/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -3301,7 +3287,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -3325,20 +3311,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="230" name="/blocks.1/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="232" name="Constant_251" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" 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>8</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>8</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="3107989" size="4" />
+			<data element_type="f32" shape="" offset="4131989" 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="3107993" size="4" />
+			<data element_type="f32" shape="" offset="4131993" 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>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="238" name="onnx::MatMul_797" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="4754653" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="5778657" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -3448,7 +3434,7 @@
 			</output>
 		</layer>
 		<layer id="239" name="Convert_747" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -3457,7 +3443,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -3465,7 +3451,7 @@
 			</output>
 		</layer>
 		<layer id="240" name="onnx::MatMul_797/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4400318" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -3473,32 +3459,32 @@
 			</output>
 		</layer>
 		<layer id="241" name="Convert_749" 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_750" type="Subtract" version="opset1">
+		<layer id="242" name="onnx::MatMul_797/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -3506,7 +3492,7 @@
 			</output>
 		</layer>
 		<layer id="243" name="onnx::MatMul_797/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="4885725" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="5909729" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -3515,57 +3501,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="244" name="onnx::MatMul_797/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="245" name="onnx::MatMul_797/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="244" name="onnx::MatMul_797/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="246" name="Constant_755" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="245" name="Constant_754" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="247" name="Reshape_756" type="Reshape" version="opset1">
+		<layer id="246" name="Reshape_755" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -3575,7 +3544,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="247" name="onnx::MatMul_797/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -3586,7 +3570,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="249" name="/blocks.1/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="251" name="Constant_256" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3243165" size="32" />
+			<data element_type="i64" shape="4" offset="4267166" 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>8</dim>
 					<dim>64</dim>
 				</port>
@@ -3656,7 +3640,7 @@
 					<dim>1</dim>
 					<dim>8</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -3667,7 +3651,7 @@
 					<dim>1</dim>
 					<dim>8</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>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</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>8</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>8</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>8</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>8</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>8</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>8</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="258" name="onnx::Add_799" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="4889821" size="2048" />
+			<data element_type="f32" shape="512" offset="5913825" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_799">
 					<dim>512</dim>
@@ -3771,7 +3755,7 @@
 			</output>
 		</layer>
 		<layer id="259" name="onnx::MatMul_800" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="4891869" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="5915873" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -3781,7 +3765,7 @@
 			</output>
 		</layer>
 		<layer id="260" name="Convert_736" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -3790,7 +3774,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -3798,7 +3782,7 @@
 			</output>
 		</layer>
 		<layer id="261" name="onnx::MatMul_800/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5774560" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -3806,32 +3790,32 @@
 			</output>
 		</layer>
 		<layer id="262" name="Convert_738" 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_739" type="Subtract" version="opset1">
+		<layer id="263" name="onnx::MatMul_800/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -3839,7 +3823,7 @@
 			</output>
 		</layer>
 		<layer id="264" name="onnx::MatMul_800/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="5022941" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="6046945" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -3848,57 +3832,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="265" name="onnx::MatMul_800/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="266" name="onnx::MatMul_800/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="265" name="onnx::MatMul_800/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="267" name="Constant_744" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="266" name="Constant_743" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="268" name="Reshape_745" type="Reshape" version="opset1">
+		<layer id="267" name="Reshape_744" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -3908,7 +3875,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="268" name="onnx::MatMul_800/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -3919,7 +3901,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -3943,20 +3925,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="271" name="/blocks.1/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="273" name="Constant_261" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" 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>8</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>8</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>8</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="276" name="Constant_268" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -4051,7 +4033,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</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>8</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="2968636" size="8" />
+			<data element_type="i64" shape="1" offset="3992636" 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="2968636" size="8" />
+			<data element_type="i64" shape="1" offset="3992636" 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="2968628" size="8" />
+			<data element_type="i64" shape="1" offset="3992628" 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>8</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="289" name="onnx::MatMul_818" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="5027037" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="6051041" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -4224,7 +4206,7 @@
 			</output>
 		</layer>
 		<layer id="290" name="Convert_769" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -4233,7 +4215,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -4241,7 +4223,7 @@
 			</output>
 		</layer>
 		<layer id="291" name="onnx::MatMul_818/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5774560" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -4249,32 +4231,32 @@
 			</output>
 		</layer>
 		<layer id="292" name="Convert_771" 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_772" type="Subtract" version="opset1">
+		<layer id="293" name="onnx::MatMul_818/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -4282,7 +4264,7 @@
 			</output>
 		</layer>
 		<layer id="294" name="onnx::MatMul_818/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="5158109" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="6182113" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -4291,57 +4273,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="295" name="onnx::MatMul_818/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="296" name="onnx::MatMul_818/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="295" name="onnx::MatMul_818/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="297" name="Constant_777" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="296" name="Constant_776" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="298" name="Reshape_778" type="Reshape" version="opset1">
+		<layer id="297" name="Reshape_777" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -4351,7 +4316,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="298" name="onnx::MatMul_818/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -4362,7 +4342,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -4386,14 +4366,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -4403,19 +4383,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -4425,20 +4405,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="303" name="onnx::Add_823" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="5162205" size="2048" />
+			<data element_type="f32" shape="512" offset="6186209" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_823">
 					<dim>512</dim>
@@ -4446,7 +4426,7 @@
 			</output>
 		</layer>
 		<layer id="304" name="onnx::Add_820" type="Const" version="opset1">
-			<data element_type="f32" shape="2048" offset="5164253" size="8192" />
+			<data element_type="f32" shape="2048" offset="6188257" size="8192" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_820">
 					<dim>2048</dim>
@@ -4454,13 +4434,13 @@
 			</output>
 		</layer>
 		<layer id="305" name="Constant_294" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="306" name="Constant_293" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" 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>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="310" name="blocks.1.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="5172445" size="2048" />
+			<data element_type="f32" shape="512" offset="6196449" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.1.mlp_ln.weight">
 					<dim>512</dim>
@@ -4517,7 +4497,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="312" name="blocks.1.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="5174493" size="2048" />
+			<data element_type="f32" shape="512" offset="6198497" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.1.mlp_ln.bias">
 					<dim>512</dim>
@@ -4545,7 +4525,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -4565,20 +4545,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="315" name="onnx::MatMul_821" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 2048" offset="5176541" size="524288" />
+			<data element_type="u4" shape="4, 128, 2048" offset="6200545" size="524288" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -4588,7 +4568,7 @@
 			</output>
 		</layer>
 		<layer id="316" name="Convert_780" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -4597,7 +4577,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -4605,7 +4585,7 @@
 			</output>
 		</layer>
 		<layer id="317" name="onnx::MatMul_821/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4400318" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -4613,32 +4593,32 @@
 			</output>
 		</layer>
 		<layer id="318" name="Convert_782" 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_783" type="Subtract" version="opset1">
+		<layer id="319" name="onnx::MatMul_821/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</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>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -4646,7 +4626,7 @@
 			</output>
 		</layer>
 		<layer id="320" name="onnx::MatMul_821/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 2048" offset="5700829" size="16384" />
+			<data element_type="f16" shape="4, 1, 2048" offset="6724833" size="16384" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -4655,57 +4635,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="321" name="onnx::MatMul_821/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>2048</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>2048</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="322" name="onnx::MatMul_821/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="321" name="onnx::MatMul_821/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>2048</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="323" name="Constant_788" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="4070589" size="16" />
+		<layer id="322" name="Constant_787" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="5094592" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="324" name="Reshape_789" type="Reshape" version="opset1">
+		<layer id="323" name="Reshape_788" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -4715,7 +4678,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>2048</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="324" name="onnx::MatMul_821/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>2048</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>2048</dim>
 				</port>
@@ -4726,7 +4704,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>2048</dim>
 				</port>
 			</output>
@@ -4750,14 +4728,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -4773,7 +4751,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -4790,14 +4768,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -4813,7 +4791,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -4831,19 +4809,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -4859,7 +4837,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="335" name="onnx::MatMul_824" type="Const" version="opset1">
-			<data element_type="u4" shape="16, 128, 512" offset="5717213" size="524288" />
+			<data element_type="u4" shape="16, 128, 512" offset="6741217" size="524288" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>16</dim>
@@ -4883,7 +4861,7 @@
 			</output>
 		</layer>
 		<layer id="336" name="Convert_791" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>16</dim>
@@ -4892,7 +4870,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -4900,7 +4878,7 @@
 			</output>
 		</layer>
 		<layer id="337" name="onnx::MatMul_824/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4400318" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -4908,32 +4886,32 @@
 			</output>
 		</layer>
 		<layer id="338" name="Convert_793" 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_794" type="Subtract" version="opset1">
+		<layer id="339" name="onnx::MatMul_824/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</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>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -4941,7 +4919,7 @@
 			</output>
 		</layer>
 		<layer id="340" name="onnx::MatMul_824/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="16, 1, 512" offset="6241501" size="16384" />
+			<data element_type="f16" shape="16, 1, 512" offset="7265505" size="16384" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>16</dim>
@@ -4950,57 +4928,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="341" name="onnx::MatMul_824/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>16</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>16</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="342" name="onnx::MatMul_824/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="341" name="onnx::MatMul_824/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>16</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="343" name="Constant_799" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="4611277" size="16" />
+		<layer id="342" name="Constant_798" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="5635280" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="344" name="Reshape_800" type="Reshape" version="opset1">
+		<layer id="343" name="Reshape_799" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -5010,7 +4971,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>2048</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="344" name="onnx::MatMul_824/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>2048</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>2048</dim>
 					<dim>512</dim>
 				</port>
@@ -5021,7 +4997,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>512</dim>
 				</port>
 			</output>
@@ -5045,14 +5021,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -5062,19 +5038,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -5084,20 +5060,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="349" name="onnx::Add_849" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="6257885" size="2048" />
+			<data element_type="f32" shape="512" offset="7281889" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_849">
 					<dim>512</dim>
@@ -5105,7 +5081,7 @@
 			</output>
 		</layer>
 		<layer id="350" name="onnx::Add_826" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="6259933" size="2048" />
+			<data element_type="f32" shape="512" offset="7283937" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_826">
 					<dim>512</dim>
@@ -5113,13 +5089,13 @@
 			</output>
 		</layer>
 		<layer id="351" name="Constant_316" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="352" name="Constant_315" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" 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>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="356" name="blocks.2.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="6261981" size="2048" />
+			<data element_type="f32" shape="512" offset="7285985" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.2.attn_ln.weight">
 					<dim>512</dim>
@@ -5176,7 +5152,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="358" name="blocks.2.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="6264029" size="2048" />
+			<data element_type="f32" shape="512" offset="7288033" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.2.attn_ln.bias">
 					<dim>512</dim>
@@ -5204,7 +5180,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -5224,20 +5200,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="361" name="onnx::MatMul_827" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="6266077" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="7290081" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -5247,7 +5223,7 @@
 			</output>
 		</layer>
 		<layer id="362" name="Convert_824" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -5256,7 +5232,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -5264,7 +5240,7 @@
 			</output>
 		</layer>
 		<layer id="363" name="onnx::MatMul_827/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="7421153" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -5272,32 +5248,32 @@
 			</output>
 		</layer>
 		<layer id="364" name="Convert_826" 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_827" type="Subtract" version="opset1">
+		<layer id="365" name="onnx::MatMul_827/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -5305,7 +5281,7 @@
 			</output>
 		</layer>
 		<layer id="366" name="onnx::MatMul_827/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="6397149" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="7421154" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -5314,57 +5290,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="367" name="onnx::MatMul_827/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="368" name="onnx::MatMul_827/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="367" name="onnx::MatMul_827/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="369" name="Constant_832" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="368" name="Constant_831" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="370" name="Reshape_833" type="Reshape" version="opset1">
+		<layer id="369" name="Reshape_832" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -5374,7 +5333,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="370" name="onnx::MatMul_827/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -5385,7 +5359,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -5409,20 +5383,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="373" name="/blocks.2/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="375" name="Constant_332" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" 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>8</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>8</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="3107989" size="4" />
+			<data element_type="f32" shape="" offset="4131989" 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="3107993" size="4" />
+			<data element_type="f32" shape="" offset="4131993" 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>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="381" name="onnx::MatMul_829" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="6401245" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="7425250" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -5532,7 +5506,7 @@
 			</output>
 		</layer>
 		<layer id="382" name="Convert_813" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -5541,7 +5515,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -5549,7 +5523,7 @@
 			</output>
 		</layer>
 		<layer id="383" name="onnx::MatMul_829/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4127812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -5557,32 +5531,32 @@
 			</output>
 		</layer>
 		<layer id="384" name="Convert_815" 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_816" type="Subtract" version="opset1">
+		<layer id="385" name="onnx::MatMul_829/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -5590,7 +5564,7 @@
 			</output>
 		</layer>
 		<layer id="386" name="onnx::MatMul_829/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="6532317" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="7556322" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -5599,57 +5573,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="387" name="onnx::MatMul_829/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="388" name="onnx::MatMul_829/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="387" name="onnx::MatMul_829/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="389" name="Constant_821" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="388" name="Constant_820" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="390" name="Reshape_822" type="Reshape" version="opset1">
+		<layer id="389" name="Reshape_821" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -5659,7 +5616,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="390" name="onnx::MatMul_829/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -5670,7 +5642,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="392" name="/blocks.2/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="394" name="Constant_337" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3243165" size="32" />
+			<data element_type="i64" shape="4" offset="4267166" 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>8</dim>
 					<dim>64</dim>
 				</port>
@@ -5740,7 +5712,7 @@
 					<dim>1</dim>
 					<dim>8</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -5751,7 +5723,7 @@
 					<dim>1</dim>
 					<dim>8</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>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</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>8</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>8</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>8</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>8</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>8</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>8</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="401" name="onnx::Add_831" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="6536413" size="2048" />
+			<data element_type="f32" shape="512" offset="7560418" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_831">
 					<dim>512</dim>
@@ -5855,7 +5827,7 @@
 			</output>
 		</layer>
 		<layer id="402" name="onnx::MatMul_832" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="6538461" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="7562466" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -5865,7 +5837,7 @@
 			</output>
 		</layer>
 		<layer id="403" name="Convert_802" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -5874,7 +5846,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -5882,7 +5854,7 @@
 			</output>
 		</layer>
 		<layer id="404" name="onnx::MatMul_832/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5774560" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -5890,32 +5862,32 @@
 			</output>
 		</layer>
 		<layer id="405" name="Convert_804" 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_805" type="Subtract" version="opset1">
+		<layer id="406" name="onnx::MatMul_832/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -5923,7 +5895,7 @@
 			</output>
 		</layer>
 		<layer id="407" name="onnx::MatMul_832/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="6669533" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="7693538" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -5932,57 +5904,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="408" name="onnx::MatMul_832/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="409" name="onnx::MatMul_832/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="408" name="onnx::MatMul_832/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="410" name="Constant_810" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="409" name="Constant_809" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="411" name="Reshape_811" type="Reshape" version="opset1">
+		<layer id="410" name="Reshape_810" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -5992,7 +5947,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="411" name="onnx::MatMul_832/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -6003,7 +5973,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -6027,20 +5997,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="414" name="/blocks.2/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="416" name="Constant_342" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" 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>8</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>8</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>8</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="419" name="Constant_349" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -6135,7 +6105,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</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>8</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="2968636" size="8" />
+			<data element_type="i64" shape="1" offset="3992636" 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="2968636" size="8" />
+			<data element_type="i64" shape="1" offset="3992636" 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="2968628" size="8" />
+			<data element_type="i64" shape="1" offset="3992628" 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>8</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="432" name="onnx::MatMul_850" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="6673629" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="7697634" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -6308,7 +6278,7 @@
 			</output>
 		</layer>
 		<layer id="433" name="Convert_835" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -6317,7 +6287,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -6325,7 +6295,7 @@
 			</output>
 		</layer>
 		<layer id="434" name="onnx::MatMul_850/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5078207" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -6333,32 +6303,32 @@
 			</output>
 		</layer>
 		<layer id="435" name="Convert_837" 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_838" type="Subtract" version="opset1">
+		<layer id="436" name="onnx::MatMul_850/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -6366,7 +6336,7 @@
 			</output>
 		</layer>
 		<layer id="437" name="onnx::MatMul_850/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="6804701" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="7828706" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -6375,57 +6345,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="438" name="onnx::MatMul_850/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="439" name="onnx::MatMul_850/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="438" name="onnx::MatMul_850/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="440" name="Constant_843" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="439" name="Constant_842" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="441" name="Reshape_844" type="Reshape" version="opset1">
+		<layer id="440" name="Reshape_843" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -6435,7 +6388,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="441" name="onnx::MatMul_850/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -6446,7 +6414,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -6470,14 +6438,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -6487,19 +6455,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -6509,20 +6477,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="446" name="onnx::Add_855" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="6808797" size="2048" />
+			<data element_type="f32" shape="512" offset="7832802" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_855">
 					<dim>512</dim>
@@ -6530,7 +6498,7 @@
 			</output>
 		</layer>
 		<layer id="447" name="onnx::Add_852" type="Const" version="opset1">
-			<data element_type="f32" shape="2048" offset="6810845" size="8192" />
+			<data element_type="f32" shape="2048" offset="7834850" size="8192" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_852">
 					<dim>2048</dim>
@@ -6538,13 +6506,13 @@
 			</output>
 		</layer>
 		<layer id="448" name="Constant_375" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="449" name="Constant_374" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" 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>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="453" name="blocks.2.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="6819037" size="2048" />
+			<data element_type="f32" shape="512" offset="7843042" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.2.mlp_ln.weight">
 					<dim>512</dim>
@@ -6601,7 +6569,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="455" name="blocks.2.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="6821085" size="2048" />
+			<data element_type="f32" shape="512" offset="7845090" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.2.mlp_ln.bias">
 					<dim>512</dim>
@@ -6629,7 +6597,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -6649,20 +6617,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="458" name="onnx::MatMul_853" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 2048" offset="6823133" size="524288" />
+			<data element_type="u4" shape="4, 128, 2048" offset="7847138" size="524288" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -6672,7 +6640,7 @@
 			</output>
 		</layer>
 		<layer id="459" name="Convert_846" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -6681,7 +6649,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -6689,7 +6657,7 @@
 			</output>
 		</layer>
 		<layer id="460" name="onnx::MatMul_853/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5078207" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -6697,32 +6665,32 @@
 			</output>
 		</layer>
 		<layer id="461" name="Convert_848" 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_849" type="Subtract" version="opset1">
+		<layer id="462" name="onnx::MatMul_853/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</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>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -6730,7 +6698,7 @@
 			</output>
 		</layer>
 		<layer id="463" name="onnx::MatMul_853/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 2048" offset="7347421" size="16384" />
+			<data element_type="f16" shape="4, 1, 2048" offset="8371426" size="16384" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -6739,57 +6707,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="464" name="onnx::MatMul_853/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>2048</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>2048</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="465" name="onnx::MatMul_853/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="464" name="onnx::MatMul_853/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>2048</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="466" name="Constant_854" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="4070589" size="16" />
+		<layer id="465" name="Constant_853" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="5094592" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="467" name="Reshape_855" type="Reshape" version="opset1">
+		<layer id="466" name="Reshape_854" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -6799,7 +6750,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>2048</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="467" name="onnx::MatMul_853/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>2048</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>2048</dim>
 				</port>
@@ -6810,7 +6776,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>2048</dim>
 				</port>
 			</output>
@@ -6834,14 +6800,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -6857,7 +6823,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -6874,14 +6840,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -6897,7 +6863,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -6915,19 +6881,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -6943,7 +6909,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="478" name="onnx::MatMul_856" type="Const" version="opset1">
-			<data element_type="u4" shape="16, 128, 512" offset="7363805" size="524288" />
+			<data element_type="u4" shape="16, 128, 512" offset="8387810" size="524288" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>16</dim>
@@ -6967,7 +6933,7 @@
 			</output>
 		</layer>
 		<layer id="479" name="Convert_857" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>16</dim>
@@ -6976,7 +6942,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -6984,7 +6950,7 @@
 			</output>
 		</layer>
 		<layer id="480" name="onnx::MatMul_856/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="8912098" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -6992,32 +6958,32 @@
 			</output>
 		</layer>
 		<layer id="481" name="Convert_859" 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_860" type="Subtract" version="opset1">
+		<layer id="482" name="onnx::MatMul_856/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</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>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -7025,7 +6991,7 @@
 			</output>
 		</layer>
 		<layer id="483" name="onnx::MatMul_856/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="16, 1, 512" offset="7888093" size="16384" />
+			<data element_type="f16" shape="16, 1, 512" offset="8912099" size="16384" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>16</dim>
@@ -7034,57 +7000,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="484" name="onnx::MatMul_856/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>16</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>16</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="485" name="onnx::MatMul_856/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="484" name="onnx::MatMul_856/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>16</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="486" name="Constant_865" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="4611277" size="16" />
+		<layer id="485" name="Constant_864" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="5635280" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="487" name="Reshape_866" type="Reshape" version="opset1">
+		<layer id="486" name="Reshape_865" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -7094,7 +7043,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>2048</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="487" name="onnx::MatMul_856/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>2048</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>2048</dim>
 					<dim>512</dim>
 				</port>
@@ -7105,7 +7069,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>512</dim>
 				</port>
 			</output>
@@ -7129,14 +7093,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -7146,19 +7110,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -7168,20 +7132,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="492" name="onnx::Add_881" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="7904477" size="2048" />
+			<data element_type="f32" shape="512" offset="8928483" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_881">
 					<dim>512</dim>
@@ -7189,7 +7153,7 @@
 			</output>
 		</layer>
 		<layer id="493" name="onnx::Add_858" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="7906525" size="2048" />
+			<data element_type="f32" shape="512" offset="8930531" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_858">
 					<dim>512</dim>
@@ -7197,13 +7161,13 @@
 			</output>
 		</layer>
 		<layer id="494" name="Constant_397" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="495" name="Constant_396" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" 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>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="499" name="blocks.3.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="7908573" size="2048" />
+			<data element_type="f32" shape="512" offset="8932579" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.3.attn_ln.weight">
 					<dim>512</dim>
@@ -7260,7 +7224,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="501" name="blocks.3.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="7910621" size="2048" />
+			<data element_type="f32" shape="512" offset="8934627" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.3.attn_ln.bias">
 					<dim>512</dim>
@@ -7288,7 +7252,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -7308,20 +7272,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="504" name="onnx::MatMul_859" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="7912669" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="8936675" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -7331,7 +7295,7 @@
 			</output>
 		</layer>
 		<layer id="505" name="Convert_890" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -7340,7 +7304,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -7348,7 +7312,7 @@
 			</output>
 		</layer>
 		<layer id="506" name="onnx::MatMul_859/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4400318" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -7356,32 +7320,32 @@
 			</output>
 		</layer>
 		<layer id="507" name="Convert_892" 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_893" type="Subtract" version="opset1">
+		<layer id="508" name="onnx::MatMul_859/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -7389,7 +7353,7 @@
 			</output>
 		</layer>
 		<layer id="509" name="onnx::MatMul_859/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="8043741" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="9067747" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -7398,57 +7362,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="510" name="onnx::MatMul_859/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="511" name="onnx::MatMul_859/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="510" name="onnx::MatMul_859/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="512" name="Constant_898" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="511" name="Constant_897" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="513" name="Reshape_899" type="Reshape" version="opset1">
+		<layer id="512" name="Reshape_898" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -7458,7 +7405,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="513" name="onnx::MatMul_859/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -7469,7 +7431,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -7493,20 +7455,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="516" name="/blocks.3/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="518" name="Constant_413" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" 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>8</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>8</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="3107989" size="4" />
+			<data element_type="f32" shape="" offset="4131989" 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="3107993" size="4" />
+			<data element_type="f32" shape="" offset="4131993" 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>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="524" name="onnx::MatMul_861" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="8047837" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="9071843" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -7616,7 +7578,7 @@
 			</output>
 		</layer>
 		<layer id="525" name="Convert_879" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -7625,7 +7587,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -7633,7 +7595,7 @@
 			</output>
 		</layer>
 		<layer id="526" name="onnx::MatMul_861/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="9202915" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -7641,32 +7603,32 @@
 			</output>
 		</layer>
 		<layer id="527" name="Convert_881" 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_882" type="Subtract" version="opset1">
+		<layer id="528" name="onnx::MatMul_861/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -7674,7 +7636,7 @@
 			</output>
 		</layer>
 		<layer id="529" name="onnx::MatMul_861/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="8178909" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="9202916" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -7683,57 +7645,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="530" name="onnx::MatMul_861/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="531" name="onnx::MatMul_861/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="530" name="onnx::MatMul_861/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="532" name="Constant_887" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="531" name="Constant_886" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="533" name="Reshape_888" type="Reshape" version="opset1">
+		<layer id="532" name="Reshape_887" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -7743,7 +7688,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="533" name="onnx::MatMul_861/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -7754,7 +7714,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="535" name="/blocks.3/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="537" name="Constant_418" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3243165" size="32" />
+			<data element_type="i64" shape="4" offset="4267166" 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>8</dim>
 					<dim>64</dim>
 				</port>
@@ -7824,7 +7784,7 @@
 					<dim>1</dim>
 					<dim>8</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -7835,7 +7795,7 @@
 					<dim>1</dim>
 					<dim>8</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>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</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>8</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>8</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>8</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>8</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>8</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>8</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="544" name="onnx::Add_863" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="8183005" size="2048" />
+			<data element_type="f32" shape="512" offset="9207012" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_863">
 					<dim>512</dim>
@@ -7939,7 +7899,7 @@
 			</output>
 		</layer>
 		<layer id="545" name="onnx::MatMul_864" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="8185053" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="9209060" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -7949,7 +7909,7 @@
 			</output>
 		</layer>
 		<layer id="546" name="Convert_868" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -7958,7 +7918,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -7966,7 +7926,7 @@
 			</output>
 		</layer>
 		<layer id="547" name="onnx::MatMul_864/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4400318" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -7974,32 +7934,32 @@
 			</output>
 		</layer>
 		<layer id="548" name="Convert_870" 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_871" type="Subtract" version="opset1">
+		<layer id="549" name="onnx::MatMul_864/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -8007,7 +7967,7 @@
 			</output>
 		</layer>
 		<layer id="550" name="onnx::MatMul_864/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="8316125" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="9340132" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -8016,57 +7976,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="551" name="onnx::MatMul_864/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="552" name="onnx::MatMul_864/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="551" name="onnx::MatMul_864/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="553" name="Constant_876" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="552" name="Constant_875" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="554" name="Reshape_877" type="Reshape" version="opset1">
+		<layer id="553" name="Reshape_876" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -8076,7 +8019,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="554" name="onnx::MatMul_864/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -8087,7 +8045,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -8111,20 +8069,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="557" name="/blocks.3/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="559" name="Constant_423" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" 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>8</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>8</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>8</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="562" name="Constant_430" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -8219,7 +8177,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</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>8</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="2968636" size="8" />
+			<data element_type="i64" shape="1" offset="3992636" 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="2968636" size="8" />
+			<data element_type="i64" shape="1" offset="3992636" 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="2968628" size="8" />
+			<data element_type="i64" shape="1" offset="3992628" 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>8</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="575" name="onnx::MatMul_882" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="8320221" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="9344228" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -8392,7 +8350,7 @@
 			</output>
 		</layer>
 		<layer id="576" name="Convert_901" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -8401,7 +8359,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -8409,7 +8367,7 @@
 			</output>
 		</layer>
 		<layer id="577" name="onnx::MatMul_882/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4263069" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -8417,32 +8375,32 @@
 			</output>
 		</layer>
 		<layer id="578" name="Convert_903" 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_904" type="Subtract" version="opset1">
+		<layer id="579" name="onnx::MatMul_882/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -8450,7 +8408,7 @@
 			</output>
 		</layer>
 		<layer id="580" name="onnx::MatMul_882/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="8451293" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="9475300" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -8459,57 +8417,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="581" name="onnx::MatMul_882/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="582" name="onnx::MatMul_882/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="581" name="onnx::MatMul_882/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="583" name="Constant_909" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="582" name="Constant_908" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="584" name="Reshape_910" type="Reshape" version="opset1">
+		<layer id="583" name="Reshape_909" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -8519,7 +8460,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="584" name="onnx::MatMul_882/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -8530,7 +8486,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -8554,14 +8510,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -8571,19 +8527,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -8593,20 +8549,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="589" name="onnx::Add_887" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="8455389" size="2048" />
+			<data element_type="f32" shape="512" offset="9479396" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_887">
 					<dim>512</dim>
@@ -8614,7 +8570,7 @@
 			</output>
 		</layer>
 		<layer id="590" name="onnx::Add_884" type="Const" version="opset1">
-			<data element_type="f32" shape="2048" offset="8457437" size="8192" />
+			<data element_type="f32" shape="2048" offset="9481444" size="8192" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_884">
 					<dim>2048</dim>
@@ -8622,13 +8578,13 @@
 			</output>
 		</layer>
 		<layer id="591" name="Constant_456" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="592" name="Constant_455" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" 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>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="596" name="blocks.3.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="8465629" size="2048" />
+			<data element_type="f32" shape="512" offset="9489636" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.3.mlp_ln.weight">
 					<dim>512</dim>
@@ -8685,7 +8641,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="598" name="blocks.3.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="8467677" size="2048" />
+			<data element_type="f32" shape="512" offset="9491684" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.3.mlp_ln.bias">
 					<dim>512</dim>
@@ -8713,7 +8669,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -8733,20 +8689,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="601" name="onnx::MatMul_885" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 2048" offset="8469725" size="524288" />
+			<data element_type="u4" shape="4, 128, 2048" offset="9493732" size="524288" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -8756,7 +8712,7 @@
 			</output>
 		</layer>
 		<layer id="602" name="Convert_912" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -8765,7 +8721,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -8773,7 +8729,7 @@
 			</output>
 		</layer>
 		<layer id="603" name="onnx::MatMul_885/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5078207" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -8781,32 +8737,32 @@
 			</output>
 		</layer>
 		<layer id="604" name="Convert_914" 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_915" type="Subtract" version="opset1">
+		<layer id="605" name="onnx::MatMul_885/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</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>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -8814,7 +8770,7 @@
 			</output>
 		</layer>
 		<layer id="606" name="onnx::MatMul_885/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 2048" offset="8994013" size="16384" />
+			<data element_type="f16" shape="4, 1, 2048" offset="10018020" size="16384" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -8823,57 +8779,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="607" name="onnx::MatMul_885/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>2048</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>2048</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="608" name="onnx::MatMul_885/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="607" name="onnx::MatMul_885/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>2048</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="609" name="Constant_920" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="4070589" size="16" />
+		<layer id="608" name="Constant_919" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="5094592" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="610" name="Reshape_921" type="Reshape" version="opset1">
+		<layer id="609" name="Reshape_920" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -8883,7 +8822,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>2048</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="610" name="onnx::MatMul_885/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>2048</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>2048</dim>
 				</port>
@@ -8894,7 +8848,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>2048</dim>
 				</port>
 			</output>
@@ -8918,14 +8872,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -8941,7 +8895,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -8958,14 +8912,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -8981,7 +8935,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -8999,19 +8953,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -9027,7 +8981,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="621" name="onnx::MatMul_888" type="Const" version="opset1">
-			<data element_type="u4" shape="16, 128, 512" offset="9010397" size="524288" />
+			<data element_type="u4" shape="16, 128, 512" offset="10034404" size="524288" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>16</dim>
@@ -9051,7 +9005,7 @@
 			</output>
 		</layer>
 		<layer id="622" name="Convert_923" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>16</dim>
@@ -9060,7 +9014,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -9068,7 +9022,7 @@
 			</output>
 		</layer>
 		<layer id="623" name="onnx::MatMul_888/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5774560" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -9076,32 +9030,32 @@
 			</output>
 		</layer>
 		<layer id="624" name="Convert_925" 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_926" type="Subtract" version="opset1">
+		<layer id="625" name="onnx::MatMul_888/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</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>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -9109,7 +9063,7 @@
 			</output>
 		</layer>
 		<layer id="626" name="onnx::MatMul_888/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="16, 1, 512" offset="9534685" size="16384" />
+			<data element_type="f16" shape="16, 1, 512" offset="10558692" size="16384" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>16</dim>
@@ -9118,57 +9072,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="627" name="onnx::MatMul_888/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>16</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>16</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="628" name="onnx::MatMul_888/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="627" name="onnx::MatMul_888/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>16</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="629" name="Constant_931" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="4611277" size="16" />
+		<layer id="628" name="Constant_930" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="5635280" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="630" name="Reshape_932" type="Reshape" version="opset1">
+		<layer id="629" name="Reshape_931" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -9178,7 +9115,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>2048</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="630" name="onnx::MatMul_888/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>2048</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>2048</dim>
 					<dim>512</dim>
 				</port>
@@ -9189,7 +9141,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>512</dim>
 				</port>
 			</output>
@@ -9213,14 +9165,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -9230,19 +9182,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -9252,20 +9204,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="635" name="onnx::Add_913" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="9551069" size="2048" />
+			<data element_type="f32" shape="512" offset="10575076" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_913">
 					<dim>512</dim>
@@ -9273,7 +9225,7 @@
 			</output>
 		</layer>
 		<layer id="636" name="onnx::Add_890" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="9553117" size="2048" />
+			<data element_type="f32" shape="512" offset="10577124" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_890">
 					<dim>512</dim>
@@ -9281,13 +9233,13 @@
 			</output>
 		</layer>
 		<layer id="637" name="Constant_478" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="638" name="Constant_477" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" 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>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="642" name="blocks.4.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="9555165" size="2048" />
+			<data element_type="f32" shape="512" offset="10579172" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.4.attn_ln.weight">
 					<dim>512</dim>
@@ -9344,7 +9296,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="644" name="blocks.4.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="9557213" size="2048" />
+			<data element_type="f32" shape="512" offset="10581220" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.4.attn_ln.bias">
 					<dim>512</dim>
@@ -9372,7 +9324,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -9392,20 +9344,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="647" name="onnx::MatMul_891" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="9559261" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="10583268" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -9415,7 +9367,7 @@
 			</output>
 		</layer>
 		<layer id="648" name="Convert_956" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -9424,7 +9376,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -9432,7 +9384,7 @@
 			</output>
 		</layer>
 		<layer id="649" name="onnx::MatMul_891/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5774560" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -9440,99 +9392,82 @@
 			</output>
 		</layer>
 		<layer id="650" name="Convert_958" 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_959" type="Subtract" version="opset1">
+		<layer id="651" name="onnx::MatMul_891/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>4</dim>
-					<dim>128</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="652" name="onnx::MatMul_891/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="9690333" size="4096" />
-			<output>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="653" name="onnx::MatMul_891/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
-					<dim>1</dim>
+					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-			</input>
+			</output>
+		</layer>
+		<layer id="652" name="onnx::MatMul_891/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="4, 1, 512" offset="10714340" size="4096" />
 			<output>
-				<port id="1" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="654" name="onnx::MatMul_891/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="653" name="onnx::MatMul_891/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="655" name="Constant_964" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="654" name="Constant_963" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="656" name="Reshape_965" type="Reshape" version="opset1">
+		<layer id="655" name="Reshape_964" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -9542,7 +9477,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="656" name="onnx::MatMul_891/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -9553,7 +9503,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -9577,20 +9527,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="659" name="/blocks.4/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="661" name="Constant_494" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" 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>8</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>8</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="3107989" size="4" />
+			<data element_type="f32" shape="" offset="4131989" 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="3107993" size="4" />
+			<data element_type="f32" shape="" offset="4131993" 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>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="667" name="onnx::MatMul_893" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="9694429" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="10718436" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -9700,7 +9650,7 @@
 			</output>
 		</layer>
 		<layer id="668" name="Convert_945" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -9709,7 +9659,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -9717,7 +9667,7 @@
 			</output>
 		</layer>
 		<layer id="669" name="onnx::MatMul_893/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="9202915" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -9725,32 +9675,32 @@
 			</output>
 		</layer>
 		<layer id="670" name="Convert_947" 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_948" type="Subtract" version="opset1">
+		<layer id="671" name="onnx::MatMul_893/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -9758,7 +9708,7 @@
 			</output>
 		</layer>
 		<layer id="672" name="onnx::MatMul_893/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="9825501" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="10849508" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -9767,57 +9717,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="673" name="onnx::MatMul_893/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="674" name="onnx::MatMul_893/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="673" name="onnx::MatMul_893/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="675" name="Constant_953" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="674" name="Constant_952" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="676" name="Reshape_954" type="Reshape" version="opset1">
+		<layer id="675" name="Reshape_953" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -9827,7 +9760,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="676" name="onnx::MatMul_893/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -9838,7 +9786,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="678" name="/blocks.4/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="680" name="Constant_499" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3243165" size="32" />
+			<data element_type="i64" shape="4" offset="4267166" 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>8</dim>
 					<dim>64</dim>
 				</port>
@@ -9908,7 +9856,7 @@
 					<dim>1</dim>
 					<dim>8</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -9919,7 +9867,7 @@
 					<dim>1</dim>
 					<dim>8</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>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</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>8</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>8</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>8</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>8</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>8</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>8</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="687" name="onnx::Add_895" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="9829597" size="2048" />
+			<data element_type="f32" shape="512" offset="10853604" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_895">
 					<dim>512</dim>
@@ -10023,7 +9971,7 @@
 			</output>
 		</layer>
 		<layer id="688" name="onnx::MatMul_896" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="9831645" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="10855652" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -10033,7 +9981,7 @@
 			</output>
 		</layer>
 		<layer id="689" name="Convert_934" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -10042,7 +9990,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -10050,7 +9998,7 @@
 			</output>
 		</layer>
 		<layer id="690" name="onnx::MatMul_896/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4400318" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -10058,32 +10006,32 @@
 			</output>
 		</layer>
 		<layer id="691" name="Convert_936" 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_937" type="Subtract" version="opset1">
+		<layer id="692" name="onnx::MatMul_896/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -10091,7 +10039,7 @@
 			</output>
 		</layer>
 		<layer id="693" name="onnx::MatMul_896/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="9962717" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="10986724" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -10100,57 +10048,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="694" name="onnx::MatMul_896/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="695" name="onnx::MatMul_896/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="694" name="onnx::MatMul_896/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="696" name="Constant_942" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="695" name="Constant_941" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="697" name="Reshape_943" type="Reshape" version="opset1">
+		<layer id="696" name="Reshape_942" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -10160,7 +10091,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="697" name="onnx::MatMul_896/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -10171,7 +10117,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -10195,20 +10141,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="700" name="/blocks.4/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="702" name="Constant_504" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" 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>8</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>8</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>8</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="705" name="Constant_511" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -10303,7 +10249,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</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>8</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="2968636" size="8" />
+			<data element_type="i64" shape="1" offset="3992636" 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="2968636" size="8" />
+			<data element_type="i64" shape="1" offset="3992636" 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="2968628" size="8" />
+			<data element_type="i64" shape="1" offset="3992628" 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>8</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="718" name="onnx::MatMul_914" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="9966813" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="10990820" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -10476,7 +10422,7 @@
 			</output>
 		</layer>
 		<layer id="719" name="Convert_967" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -10485,7 +10431,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -10493,7 +10439,7 @@
 			</output>
 		</layer>
 		<layer id="720" name="onnx::MatMul_914/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4400318" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -10501,32 +10447,32 @@
 			</output>
 		</layer>
 		<layer id="721" name="Convert_969" 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_970" type="Subtract" version="opset1">
+		<layer id="722" name="onnx::MatMul_914/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -10534,7 +10480,7 @@
 			</output>
 		</layer>
 		<layer id="723" name="onnx::MatMul_914/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="10097885" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="11121892" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -10543,57 +10489,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="724" name="onnx::MatMul_914/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="725" name="onnx::MatMul_914/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="724" name="onnx::MatMul_914/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="726" name="Constant_975" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="725" name="Constant_974" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="727" name="Reshape_976" type="Reshape" version="opset1">
+		<layer id="726" name="Reshape_975" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -10603,7 +10532,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="727" name="onnx::MatMul_914/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -10614,7 +10558,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -10638,14 +10582,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -10655,19 +10599,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -10677,20 +10621,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="732" name="onnx::Add_919" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="10101981" size="2048" />
+			<data element_type="f32" shape="512" offset="11125988" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_919">
 					<dim>512</dim>
@@ -10698,7 +10642,7 @@
 			</output>
 		</layer>
 		<layer id="733" name="onnx::Add_916" type="Const" version="opset1">
-			<data element_type="f32" shape="2048" offset="10104029" size="8192" />
+			<data element_type="f32" shape="2048" offset="11128036" size="8192" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_916">
 					<dim>2048</dim>
@@ -10706,13 +10650,13 @@
 			</output>
 		</layer>
 		<layer id="734" name="Constant_537" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="735" name="Constant_536" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" 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>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="739" name="blocks.4.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="10112221" size="2048" />
+			<data element_type="f32" shape="512" offset="11136228" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.4.mlp_ln.weight">
 					<dim>512</dim>
@@ -10769,7 +10713,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="741" name="blocks.4.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="10114269" size="2048" />
+			<data element_type="f32" shape="512" offset="11138276" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.4.mlp_ln.bias">
 					<dim>512</dim>
@@ -10797,7 +10741,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -10817,20 +10761,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="744" name="onnx::MatMul_917" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 2048" offset="10116317" size="524288" />
+			<data element_type="u4" shape="4, 128, 2048" offset="11140324" size="524288" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -10840,7 +10784,7 @@
 			</output>
 		</layer>
 		<layer id="745" name="Convert_978" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -10849,7 +10793,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -10857,7 +10801,7 @@
 			</output>
 		</layer>
 		<layer id="746" name="onnx::MatMul_917/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4127812" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -10865,32 +10809,32 @@
 			</output>
 		</layer>
 		<layer id="747" name="Convert_980" 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_981" type="Subtract" version="opset1">
+		<layer id="748" name="onnx::MatMul_917/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</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>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -10898,7 +10842,7 @@
 			</output>
 		</layer>
 		<layer id="749" name="onnx::MatMul_917/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 2048" offset="10640605" size="16384" />
+			<data element_type="f16" shape="4, 1, 2048" offset="11664612" size="16384" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -10907,57 +10851,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="750" name="onnx::MatMul_917/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>2048</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>2048</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="751" name="onnx::MatMul_917/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="750" name="onnx::MatMul_917/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>2048</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="752" name="Constant_986" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="4070589" size="16" />
+		<layer id="751" name="Constant_985" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="5094592" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="753" name="Reshape_987" type="Reshape" version="opset1">
+		<layer id="752" name="Reshape_986" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -10967,7 +10894,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>2048</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="753" name="onnx::MatMul_917/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>2048</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>2048</dim>
 				</port>
@@ -10978,7 +10920,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>2048</dim>
 				</port>
 			</output>
@@ -11002,14 +10944,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -11025,7 +10967,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -11042,14 +10984,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -11065,7 +11007,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -11083,19 +11025,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -11111,7 +11053,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="764" name="onnx::MatMul_920" type="Const" version="opset1">
-			<data element_type="u4" shape="16, 128, 512" offset="10656989" size="524288" />
+			<data element_type="u4" shape="16, 128, 512" offset="11680996" size="524288" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>16</dim>
@@ -11135,7 +11077,7 @@
 			</output>
 		</layer>
 		<layer id="765" name="Convert_989" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>16</dim>
@@ -11144,7 +11086,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -11152,7 +11094,7 @@
 			</output>
 		</layer>
 		<layer id="766" name="onnx::MatMul_920/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5774560" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -11160,32 +11102,32 @@
 			</output>
 		</layer>
 		<layer id="767" name="Convert_991" 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_992" type="Subtract" version="opset1">
+		<layer id="768" name="onnx::MatMul_920/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</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>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -11193,7 +11135,7 @@
 			</output>
 		</layer>
 		<layer id="769" name="onnx::MatMul_920/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="16, 1, 512" offset="11181277" size="16384" />
+			<data element_type="f16" shape="16, 1, 512" offset="12205284" size="16384" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>16</dim>
@@ -11202,57 +11144,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="770" name="onnx::MatMul_920/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>16</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>16</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="771" name="onnx::MatMul_920/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="770" name="onnx::MatMul_920/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>16</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="772" name="Constant_997" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="4611277" size="16" />
+		<layer id="771" name="Constant_996" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="5635280" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="773" name="Reshape_998" type="Reshape" version="opset1">
+		<layer id="772" name="Reshape_997" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>16</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -11262,7 +11187,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>2048</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="773" name="onnx::MatMul_920/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>2048</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>2048</dim>
 					<dim>512</dim>
 				</port>
@@ -11273,7 +11213,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>512</dim>
 				</port>
 			</output>
@@ -11297,14 +11237,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -11314,19 +11254,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -11336,20 +11276,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="778" name="onnx::Add_945" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="11197661" size="2048" />
+			<data element_type="f32" shape="512" offset="12221668" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_945">
 					<dim>512</dim>
@@ -11357,7 +11297,7 @@
 			</output>
 		</layer>
 		<layer id="779" name="onnx::Add_922" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="11199709" size="2048" />
+			<data element_type="f32" shape="512" offset="12223716" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_922">
 					<dim>512</dim>
@@ -11365,13 +11305,13 @@
 			</output>
 		</layer>
 		<layer id="780" name="Constant_559" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="781" name="Constant_558" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" 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>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="785" name="blocks.5.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="11201757" size="2048" />
+			<data element_type="f32" shape="512" offset="12225764" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.5.attn_ln.weight">
 					<dim>512</dim>
@@ -11428,7 +11368,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="787" name="blocks.5.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="11203805" size="2048" />
+			<data element_type="f32" shape="512" offset="12227812" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.5.attn_ln.bias">
 					<dim>512</dim>
@@ -11456,7 +11396,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -11476,20 +11416,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="790" name="onnx::MatMul_923" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="11205853" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="12229860" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -11499,7 +11439,7 @@
 			</output>
 		</layer>
 		<layer id="791" name="Convert_1022" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -11508,7 +11448,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -11516,7 +11456,7 @@
 			</output>
 		</layer>
 		<layer id="792" name="onnx::MatMul_923/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4400318" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -11524,32 +11464,32 @@
 			</output>
 		</layer>
 		<layer id="793" name="Convert_1024" 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_1025" type="Subtract" version="opset1">
+		<layer id="794" name="onnx::MatMul_923/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -11557,7 +11497,7 @@
 			</output>
 		</layer>
 		<layer id="795" name="onnx::MatMul_923/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="11336925" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="12360932" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -11566,57 +11506,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="796" name="onnx::MatMul_923/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="797" name="onnx::MatMul_923/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="796" name="onnx::MatMul_923/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="798" name="Constant_1030" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="797" name="Constant_1029" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="799" name="Reshape_1031" type="Reshape" version="opset1">
+		<layer id="798" name="Reshape_1030" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -11626,7 +11549,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="799" name="onnx::MatMul_923/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -11637,7 +11575,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -11661,20 +11599,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="802" name="/blocks.5/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="804" name="Constant_575" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" 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>8</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>8</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="3107989" size="4" />
+			<data element_type="f32" shape="" offset="4131989" 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="3107993" size="4" />
+			<data element_type="f32" shape="" offset="4131993" 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>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="810" name="onnx::MatMul_925" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="11341021" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="12365028" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -11784,7 +11722,7 @@
 			</output>
 		</layer>
 		<layer id="811" name="Convert_1011" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -11793,7 +11731,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -11801,7 +11739,7 @@
 			</output>
 		</layer>
 		<layer id="812" name="onnx::MatMul_925/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5774560" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -11809,32 +11747,32 @@
 			</output>
 		</layer>
 		<layer id="813" name="Convert_1013" 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_1014" type="Subtract" version="opset1">
+		<layer id="814" name="onnx::MatMul_925/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -11842,7 +11780,7 @@
 			</output>
 		</layer>
 		<layer id="815" name="onnx::MatMul_925/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="11472093" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="12496100" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -11851,57 +11789,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="816" name="onnx::MatMul_925/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="817" name="onnx::MatMul_925/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="816" name="onnx::MatMul_925/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="818" name="Constant_1019" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="817" name="Constant_1018" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="819" name="Reshape_1020" type="Reshape" version="opset1">
+		<layer id="818" name="Reshape_1019" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -11911,7 +11832,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="819" name="onnx::MatMul_925/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -11922,7 +11858,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="821" name="/blocks.5/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="823" name="Constant_580" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3243165" size="32" />
+			<data element_type="i64" shape="4" offset="4267166" 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>8</dim>
 					<dim>64</dim>
 				</port>
@@ -11992,7 +11928,7 @@
 					<dim>1</dim>
 					<dim>8</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -12003,7 +11939,7 @@
 					<dim>1</dim>
 					<dim>8</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>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</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>8</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>8</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>8</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>8</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>8</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>8</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="830" name="onnx::Add_927" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="11476189" size="2048" />
+			<data element_type="f32" shape="512" offset="12500196" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_927">
 					<dim>512</dim>
@@ -12107,7 +12043,7 @@
 			</output>
 		</layer>
 		<layer id="831" name="onnx::MatMul_928" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="11478237" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="12502244" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -12117,7 +12053,7 @@
 			</output>
 		</layer>
 		<layer id="832" name="Convert_1000" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -12126,7 +12062,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -12134,7 +12070,7 @@
 			</output>
 		</layer>
 		<layer id="833" name="onnx::MatMul_928/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="4400318" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -12142,32 +12078,32 @@
 			</output>
 		</layer>
 		<layer id="834" name="Convert_1002" 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_1003" type="Subtract" version="opset1">
+		<layer id="835" name="onnx::MatMul_928/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -12175,7 +12111,7 @@
 			</output>
 		</layer>
 		<layer id="836" name="onnx::MatMul_928/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="11609309" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="12633316" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -12184,57 +12120,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="837" name="onnx::MatMul_928/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="838" name="onnx::MatMul_928/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="837" name="onnx::MatMul_928/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="839" name="Constant_1008" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="838" name="Constant_1007" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="840" name="Reshape_1009" type="Reshape" version="opset1">
+		<layer id="839" name="Reshape_1008" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -12244,7 +12163,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="840" name="onnx::MatMul_928/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -12255,7 +12189,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -12279,20 +12213,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="843" name="/blocks.5/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107925" size="32" />
+			<data element_type="i64" shape="4" offset="4131925" 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>512</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>8</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="845" name="Constant_585" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" 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>8</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>8</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>8</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="848" name="Constant_592" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="3107957" size="32" />
+			<data element_type="i64" shape="4" offset="4131957" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -12387,7 +12321,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>8</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>8</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>8</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="2968636" size="8" />
+			<data element_type="i64" shape="1" offset="3992636" 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="2968636" size="8" />
+			<data element_type="i64" shape="1" offset="3992636" 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="2968628" size="8" />
+			<data element_type="i64" shape="1" offset="3992628" 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>8</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="861" name="onnx::MatMul_946" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 512" offset="11613405" size="131072" />
+			<data element_type="u4" shape="4, 128, 512" offset="12637412" size="131072" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -12560,7 +12494,7 @@
 			</output>
 		</layer>
 		<layer id="862" name="Convert_1033" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -12569,7 +12503,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -12577,7 +12511,7 @@
 			</output>
 		</layer>
 		<layer id="863" name="onnx::MatMul_946/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="7421153" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -12585,32 +12519,32 @@
 			</output>
 		</layer>
 		<layer id="864" name="Convert_1035" 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_1036" type="Subtract" version="opset1">
+		<layer id="865" name="onnx::MatMul_946/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</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>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -12618,7 +12552,7 @@
 			</output>
 		</layer>
 		<layer id="866" name="onnx::MatMul_946/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 512" offset="11744477" size="4096" />
+			<data element_type="f16" shape="4, 1, 512" offset="12768484" size="4096" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -12627,57 +12561,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="867" name="onnx::MatMul_946/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="868" name="onnx::MatMul_946/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="867" name="onnx::MatMul_946/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="869" name="Constant_1041" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3107909" size="16" />
+		<layer id="868" name="Constant_1040" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="4131909" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="870" name="Reshape_1042" type="Reshape" version="opset1">
+		<layer id="869" name="Reshape_1041" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>512</dim>
@@ -12687,7 +12604,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="870" name="onnx::MatMul_946/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>512</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>512</dim>
 				</port>
@@ -12698,7 +12630,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -12722,14 +12654,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -12739,19 +12671,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -12761,20 +12693,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="875" name="onnx::Add_951" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="11748573" size="2048" />
+			<data element_type="f32" shape="512" offset="12772580" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_951">
 					<dim>512</dim>
@@ -12782,7 +12714,7 @@
 			</output>
 		</layer>
 		<layer id="876" name="onnx::Add_948" type="Const" version="opset1">
-			<data element_type="f32" shape="2048" offset="11750621" size="8192" />
+			<data element_type="f32" shape="2048" offset="12774628" size="8192" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_948">
 					<dim>2048</dim>
@@ -12790,13 +12722,13 @@
 			</output>
 		</layer>
 		<layer id="877" name="Constant_618" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="878" name="Constant_617" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" 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>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="882" name="blocks.5.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="11758813" size="2048" />
+			<data element_type="f32" shape="512" offset="12782820" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.5.mlp_ln.weight">
 					<dim>512</dim>
@@ -12853,7 +12785,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="884" name="blocks.5.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="11760861" size="2048" />
+			<data element_type="f32" shape="512" offset="12784868" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.5.mlp_ln.bias">
 					<dim>512</dim>
@@ -12881,7 +12813,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -12901,20 +12833,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="887" name="onnx::MatMul_949" type="Const" version="opset1">
-			<data element_type="u4" shape="4, 128, 2048" offset="11762909" size="524288" />
+			<data element_type="u4" shape="4, 128, 2048" offset="12786916" size="524288" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -12924,7 +12856,7 @@
 			</output>
 		</layer>
 		<layer id="888" name="Convert_1044" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>4</dim>
@@ -12933,7 +12865,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -12941,7 +12873,7 @@
 			</output>
 		</layer>
 		<layer id="889" name="onnx::MatMul_949/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="3103812" size="1" />
+			<data element_type="u4" shape="1" offset="5078207" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -12949,32 +12881,32 @@
 			</output>
 		</layer>
 		<layer id="890" name="Convert_1046" 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_1047" type="Subtract" version="opset1">
+		<layer id="891" name="onnx::MatMul_949/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</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>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -12982,7 +12914,7 @@
 			</output>
 		</layer>
 		<layer id="892" name="onnx::MatMul_949/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="4, 1, 2048" offset="12287197" size="16384" />
+			<data element_type="f16" shape="4, 1, 2048" offset="13311204" size="16384" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>4</dim>
@@ -12991,57 +12923,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="893" name="onnx::MatMul_949/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>2048</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>4</dim>
-					<dim>1</dim>
-					<dim>2048</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="894" name="onnx::MatMul_949/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="893" name="onnx::MatMul_949/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>4</dim>
 					<dim>1</dim>
 					<dim>2048</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="895" name="Constant_1052" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="4070589" size="16" />
+		<layer id="894" name="Constant_1051" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="5094592" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="896" name="Reshape_1053" type="Reshape" version="opset1">
+		<layer id="895" name="Reshape_1052" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>4</dim>
 					<dim>128</dim>
 					<dim>2048</dim>
@@ -13051,7 +12966,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>512</dim>
+					<dim>2048</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="896" name="onnx::MatMul_949/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>512</dim>
+					<dim>2048</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>512</dim>
 					<dim>2048</dim>
 				</port>
@@ -13062,7 +12992,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>2048</dim>
 				</port>
 			</output>
@@ -13086,14 +13016,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -13109,7 +13039,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -13126,14 +13056,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -13149,7 +13079,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -13167,19 +13097,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
@@ -13195,7 +13125,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</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>2048</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="907" name="onnx::MatMul_952" type="Const" version="opset1">
-			<data element_type="u8" shape="2048, 512" offset="12303581" size="1048576" />
+			<data element_type="u8" shape="2048, 512" offset="13327588" size="1048576" />
 			<output>
 				<port id="0" precision="U8">
 					<dim>2048</dim>
@@ -13218,7 +13148,7 @@
 			</output>
 		</layer>
 		<layer id="908" name="Convert_1056" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>2048</dim>
@@ -13226,14 +13156,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>2048</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="909" name="onnx::MatMul_952/zero_point" type="Const" version="opset1">
-			<data element_type="u8" shape="1, 512" offset="13352157" size="512" />
+			<data element_type="u8" shape="1, 512" offset="14376164" size="512" />
 			<output>
 				<port id="0" precision="U8">
 					<dim>1</dim>
@@ -13242,7 +13172,7 @@
 			</output>
 		</layer>
 		<layer id="910" name="Convert_1059" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>1</dim>
@@ -13250,33 +13180,33 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="911" name="Subtract_1060" type="Subtract" version="opset1">
+		<layer id="911" name="onnx::MatMul_952/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>2048</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>2048</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="912" name="onnx::MatMul_952/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="1, 512" offset="13352669" size="1024" />
+			<data element_type="f16" shape="1, 512" offset="14376676" size="1024" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>1</dim>
@@ -13284,35 +13214,35 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="913" name="onnx::MatMul_952/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="913" name="onnx::MatMul_952/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
+					<dim>2048</dim>
+					<dim>512</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
+				<port id="2" precision="FP16">
+					<dim>2048</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="914" name="onnx::MatMul_952/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="914" name="onnx::MatMul_952/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>2048</dim>
 					<dim>512</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>512</dim>
-				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32">
 					<dim>2048</dim>
 					<dim>512</dim>
 				</port>
@@ -13323,7 +13253,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>2048</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -13334,7 +13264,7 @@
 			<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>512</dim>
 				</port>
 			</output>
@@ -13347,14 +13277,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -13364,19 +13294,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
@@ -13386,26 +13316,26 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</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>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="919" name="Constant_640" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968628" size="8" />
+			<data element_type="i64" shape="" offset="3992628" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="920" name="Constant_639" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2968636" size="8" />
+			<data element_type="i64" shape="" offset="3992636" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -13434,7 +13364,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -13444,13 +13374,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="924" name="ln_post.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="13353693" size="2048" />
+			<data element_type="f32" shape="512" offset="14377700" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="ln_post.weight">
 					<dim>512</dim>
@@ -13462,7 +13392,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -13472,13 +13402,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="926" name="ln_post.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="512" offset="13355741" size="2048" />
+			<data element_type="f32" shape="512" offset="14379748" size="2048" />
 			<output>
 				<port id="0" precision="FP32" names="ln_post.bias">
 					<dim>512</dim>
@@ -13490,7 +13420,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -13500,7 +13430,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/ln_post/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 			</output>
@@ -13510,14 +13440,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="output_features">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 			</output>
@@ -13526,7 +13456,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>512</dim>
 				</port>
 			</input>
@@ -13538,10 +13468,10 @@
 		<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="20" to-port="0" />
 		<edge from-layer="9" from-port="2" to-layer="13" to-port="0" />
 		<edge from-layer="10" from-port="0" to-layer="19" to-port="0" />
@@ -13570,10 +13500,10 @@
 		<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" />
@@ -13616,19 +13546,19 @@
 		<edge from-layer="71" from-port="2" to-layer="73" to-port="0" />
 		<edge from-layer="72" from-port="0" to-layer="73" to-port="1" />
 		<edge from-layer="73" from-port="2" to-layer="74" to-port="0" />
-		<edge from-layer="74" from-port="1" to-layer="126" to-port="0" />
 		<edge from-layer="74" from-port="1" to-layer="105" to-port="0" />
+		<edge from-layer="74" from-port="1" to-layer="126" to-port="0" />
 		<edge from-layer="74" from-port="1" to-layer="85" to-port="0" />
 		<edge from-layer="75" from-port="0" to-layer="76" to-port="0" />
 		<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" />
@@ -13644,12 +13574,12 @@
 		<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" />
@@ -13665,12 +13595,12 @@
 		<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" />
@@ -13679,11 +13609,11 @@
 		<edge from-layer="131" from-port="2" to-layer="132" to-port="1" />
 		<edge from-layer="132" from-port="2" to-layer="134" to-port="0" />
 		<edge from-layer="133" from-port="0" to-layer="134" to-port="1" />
-		<edge from-layer="134" from-port="2" to-layer="145" to-port="0" />
 		<edge from-layer="134" from-port="2" to-layer="135" to-port="0" />
+		<edge from-layer="134" from-port="2" to-layer="145" to-port="0" />
 		<edge from-layer="135" from-port="1" to-layer="142" to-port="0" />
-		<edge from-layer="136" from-port="0" to-layer="142" to-port="1" />
 		<edge from-layer="136" from-port="0" to-layer="139" to-port="0" />
+		<edge from-layer="136" from-port="0" to-layer="142" to-port="1" />
 		<edge from-layer="137" from-port="0" to-layer="142" to-port="2" />
 		<edge from-layer="138" from-port="0" to-layer="140" to-port="0" />
 		<edge from-layer="139" from-port="1" to-layer="140" to-port="1" />
@@ -13697,12 +13627,12 @@
 		<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" />
@@ -13724,12 +13654,12 @@
 		<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" />
@@ -13745,12 +13675,12 @@
 		<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" />
@@ -13767,19 +13697,19 @@
 		<edge from-layer="214" from-port="2" to-layer="216" to-port="0" />
 		<edge from-layer="215" from-port="0" to-layer="216" to-port="1" />
 		<edge from-layer="216" from-port="2" to-layer="217" to-port="0" />
-		<edge from-layer="217" from-port="1" to-layer="269" to-port="0" />
 		<edge from-layer="217" from-port="1" to-layer="248" to-port="0" />
+		<edge from-layer="217" from-port="1" to-layer="269" to-port="0" />
 		<edge from-layer="217" from-port="1" to-layer="228" to-port="0" />
 		<edge from-layer="218" from-port="0" to-layer="219" to-port="0" />
 		<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" />
@@ -13795,12 +13725,12 @@
 		<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" />
@@ -13816,12 +13746,12 @@
 		<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" />
@@ -13848,12 +13778,12 @@
 		<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" />
@@ -13875,12 +13805,12 @@
 		<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" />
@@ -13896,12 +13826,12 @@
 		<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" />
@@ -13925,12 +13855,12 @@
 		<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" />
@@ -13946,12 +13876,12 @@
 		<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" />
@@ -13967,12 +13897,12 @@
 		<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" />
@@ -13999,12 +13929,12 @@
 		<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" />
@@ -14026,12 +13956,12 @@
 		<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" />
@@ -14047,12 +13977,12 @@
 		<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" />
@@ -14076,12 +14006,12 @@
 		<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" />
@@ -14097,12 +14027,12 @@
 		<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" />
@@ -14118,12 +14048,12 @@
 		<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" />
@@ -14150,12 +14080,12 @@
 		<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" />
@@ -14177,12 +14107,12 @@
 		<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="618" to-port="0" />
 		<edge from-layer="612" from-port="2" to-layer="614" to-port="0" />
@@ -14198,12 +14128,12 @@
 		<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" />
@@ -14227,12 +14157,12 @@
 		<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" />
@@ -14248,12 +14178,12 @@
 		<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" />
@@ -14269,12 +14199,12 @@
 		<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" />
@@ -14301,12 +14231,12 @@
 		<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" />
@@ -14328,12 +14258,12 @@
 		<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="761" to-port="0" />
 		<edge from-layer="755" from-port="2" to-layer="757" to-port="0" />
@@ -14349,12 +14279,12 @@
 		<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" />
@@ -14378,12 +14308,12 @@
 		<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" />
@@ -14399,12 +14329,12 @@
 		<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" />
@@ -14420,12 +14350,12 @@
 		<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" />
@@ -14452,12 +14382,12 @@
 		<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" />
@@ -14479,12 +14409,12 @@
 		<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" />
@@ -14500,10 +14430,10 @@
 		<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="915" 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="914" to-port="0" />
+		<edge from-layer="914" from-port="1" to-layer="915" to-port="1" />
 		<edge from-layer="915" from-port="2" to-layer="916" to-port="1" />
 		<edge from-layer="916" from-port="2" to-layer="917" to-port="1" />
 		<edge from-layer="917" from-port="2" to-layer="918" to-port="0" />
@@ -14525,16 +14455,12 @@
 			<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>