diff --git "a/multilang/ggml-tiny-encoder-openvino-int4/ggml-tiny-encoder-openvino.xml" "b/multilang/ggml-tiny-encoder-openvino-int4/ggml-tiny-encoder-openvino.xml"
--- "a/multilang/ggml-tiny-encoder-openvino-int4/ggml-tiny-encoder-openvino.xml"
+++ "b/multilang/ggml-tiny-encoder-openvino-int4/ggml-tiny-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_457" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>384</dim>
@@ -31,7 +31,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>384</dim>
 					<dim>80</dim>
 					<dim>3</dim>
@@ -49,7 +49,7 @@
 			</output>
 		</layer>
 		<layer id="4" name="Convert_460" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>384</dim>
@@ -58,29 +58,29 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>384</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="5" name="Subtract_461" type="Subtract" version="opset1">
+		<layer id="5" name="onnx::Conv_518/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>384</dim>
 					<dim>80</dim>
 					<dim>3</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>384</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>384</dim>
 					<dim>80</dim>
 					<dim>3</dim>
@@ -97,39 +97,39 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="7" name="onnx::Conv_518/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="7" name="onnx::Conv_518/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>80</dim>
+					<dim>3</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>384</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>384</dim>
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>80</dim>
+					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="8" name="onnx::Conv_518/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="8" name="onnx::Conv_518/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>384</dim>
 					<dim>80</dim>
 					<dim>3</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>384</dim>
-					<dim>1</dim>
-					<dim>1</dim>
-				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32">
 					<dim>384</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>384</dim>
@@ -154,7 +154,7 @@
 				<port id="2" precision="FP32">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -193,7 +193,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</input>
 			<output>
@@ -298,7 +298,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</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>384</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -326,7 +326,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</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>384</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -343,14 +343,14 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/Erf_output_0">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -366,7 +366,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</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>384</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -384,19 +384,19 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/Mul_output_0">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -412,7 +412,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</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>384</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 			</output>
 		</layer>
@@ -435,7 +435,7 @@
 			</output>
 		</layer>
 		<layer id="30" name="Convert_468" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>384</dim>
@@ -444,7 +444,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>384</dim>
 					<dim>384</dim>
 					<dim>3</dim>
@@ -462,7 +462,7 @@
 			</output>
 		</layer>
 		<layer id="32" name="Convert_471" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>384</dim>
@@ -471,29 +471,29 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>384</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="33" name="Subtract_472" type="Subtract" version="opset1">
+		<layer id="33" name="onnx::Conv_520/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>384</dim>
 					<dim>384</dim>
 					<dim>3</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>384</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>384</dim>
 					<dim>384</dim>
 					<dim>3</dim>
@@ -510,39 +510,39 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="35" name="onnx::Conv_520/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="35" name="onnx::Conv_520/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+					<dim>3</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>384</dim>
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>384</dim>
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>384</dim>
+					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="36" name="onnx::Conv_520/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="36" name="onnx::Conv_520/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>384</dim>
 					<dim>384</dim>
 					<dim>3</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>384</dim>
-					<dim>1</dim>
-					<dim>1</dim>
-				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 					<dim>3</dim>
@@ -555,7 +555,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>2000</dim>
+					<dim>3000</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>384</dim>
@@ -567,7 +567,7 @@
 				<port id="2" precision="FP32">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -606,7 +606,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
@@ -711,7 +711,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</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>384</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -739,7 +739,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</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>384</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -756,14 +756,14 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/Erf_1_output_0">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -779,7 +779,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</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>384</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -797,19 +797,19 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/Mul_2_output_0">
 					<dim>1</dim>
 					<dim>384</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -825,7 +825,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</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>384</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -850,7 +850,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="59" name="onnx::Add_526" type="Const" version="opset1">
-			<data element_type="f32" shape="1000, 384" offset="539956" size="1536000" />
+			<data element_type="f32" shape="1500, 384" offset="539956" size="2304000" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_526">
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 			</output>
@@ -878,18 +878,18 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 				<port id="1" precision="FP32">
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
 				<port id="2" precision="FP32" names="/Add_2_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 			</output>
@@ -899,14 +899,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="/Cast_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 			</output>
@@ -916,20 +916,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="63" name="onnx::Add_551" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="2075956" size="1536" />
+			<data element_type="f32" shape="384" offset="2843956" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_551">
 					<dim>384</dim>
@@ -937,7 +937,7 @@
 			</output>
 		</layer>
 		<layer id="64" name="onnx::Add_528" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="2077492" size="1536" />
+			<data element_type="f32" shape="384" offset="2845492" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_528">
 					<dim>384</dim>
@@ -945,13 +945,13 @@
 			</output>
 		</layer>
 		<layer id="65" name="Constant_124" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079028" size="8" />
+			<data element_type="i64" shape="" offset="2847028" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="66" name="Constant_123" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079036" size="8" />
+			<data element_type="i64" shape="" offset="2847036" 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>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="70" name="blocks.0.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="2079044" size="1536" />
+			<data element_type="f32" shape="384" offset="2847044" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.0.attn_ln.weight">
 					<dim>384</dim>
@@ -1008,7 +1008,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="72" name="blocks.0.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="2080580" size="1536" />
+			<data element_type="f32" shape="384" offset="2848580" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.0.attn_ln.bias">
 					<dim>384</dim>
@@ -1036,7 +1036,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -1056,20 +1056,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="75" name="onnx::MatMul_529" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="2082116" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="2850116" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -1079,7 +1079,7 @@
 			</output>
 		</layer>
 		<layer id="76" name="Convert_500" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -1088,7 +1088,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -1096,7 +1096,7 @@
 			</output>
 		</layer>
 		<layer id="77" name="onnx::MatMul_529/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="2923844" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -1104,32 +1104,32 @@
 			</output>
 		</layer>
 		<layer id="78" name="Convert_502" 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_503" type="Subtract" version="opset1">
+		<layer id="79" name="onnx::MatMul_529/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -1137,7 +1137,7 @@
 			</output>
 		</layer>
 		<layer id="80" name="onnx::MatMul_529/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="2155845" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="2923845" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -1146,57 +1146,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="81" name="onnx::MatMul_529/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="82" name="onnx::MatMul_529/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="81" name="onnx::MatMul_529/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="83" name="Constant_508" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="82" name="Constant_507" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="84" name="Reshape_509" type="Reshape" version="opset1">
+		<layer id="83" name="Reshape_508" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -1206,7 +1189,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="84" name="onnx::MatMul_529/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -1217,7 +1215,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -1241,20 +1239,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="87" name="/blocks.0/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158165" size="32" />
+			<data element_type="i64" shape="4" offset="2926165" 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>384</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>6</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="89" name="Constant_140" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158197" size="32" />
+			<data element_type="i64" shape="4" offset="2926197" 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>6</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>6</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="2158229" size="4" />
+			<data element_type="f32" shape="" offset="2926229" 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="2158233" size="4" />
+			<data element_type="f32" shape="" offset="2926233" 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>6</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>6</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="95" name="onnx::MatMul_531" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="2158237" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="2926237" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -1364,7 +1362,7 @@
 			</output>
 		</layer>
 		<layer id="96" name="Convert_489" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -1373,7 +1371,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -1381,7 +1379,7 @@
 			</output>
 		</layer>
 		<layer id="97" name="onnx::MatMul_531/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="2999965" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -1389,32 +1387,32 @@
 			</output>
 		</layer>
 		<layer id="98" name="Convert_491" 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_492" type="Subtract" version="opset1">
+		<layer id="99" name="onnx::MatMul_531/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -1422,7 +1420,7 @@
 			</output>
 		</layer>
 		<layer id="100" name="onnx::MatMul_531/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="2231965" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="2999966" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -1431,57 +1429,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="101" name="onnx::MatMul_531/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="102" name="onnx::MatMul_531/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="101" name="onnx::MatMul_531/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="103" name="Constant_497" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="102" name="Constant_496" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="104" name="Reshape_498" type="Reshape" version="opset1">
+		<layer id="103" name="Reshape_497" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -1491,7 +1472,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="104" name="onnx::MatMul_531/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -1502,7 +1498,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="106" name="/blocks.0/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158165" size="32" />
+			<data element_type="i64" shape="4" offset="2926165" 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>384</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>6</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="108" name="Constant_145" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2234269" size="32" />
+			<data element_type="i64" shape="4" offset="3002270" 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>6</dim>
 					<dim>64</dim>
 				</port>
@@ -1572,7 +1568,7 @@
 					<dim>1</dim>
 					<dim>6</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -1583,7 +1579,7 @@
 					<dim>1</dim>
 					<dim>6</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>6</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>6</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>6</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>6</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>6</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>6</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>6</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>6</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>6</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>6</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="115" name="onnx::Add_533" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="2234301" size="1536" />
+			<data element_type="f32" shape="384" offset="3002302" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_533">
 					<dim>384</dim>
@@ -1687,7 +1683,7 @@
 			</output>
 		</layer>
 		<layer id="116" name="onnx::MatMul_534" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="2235837" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="3003838" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -1697,7 +1693,7 @@
 			</output>
 		</layer>
 		<layer id="117" name="Convert_478" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -1706,7 +1702,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -1714,7 +1710,7 @@
 			</output>
 		</layer>
 		<layer id="118" name="onnx::MatMul_534/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3077566" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -1722,32 +1718,32 @@
 			</output>
 		</layer>
 		<layer id="119" name="Convert_480" 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_481" type="Subtract" version="opset1">
+		<layer id="120" name="onnx::MatMul_534/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -1755,7 +1751,7 @@
 			</output>
 		</layer>
 		<layer id="121" name="onnx::MatMul_534/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="2309565" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="3077567" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -1764,57 +1760,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="122" name="onnx::MatMul_534/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="123" name="onnx::MatMul_534/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="122" name="onnx::MatMul_534/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="124" name="Constant_486" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="123" name="Constant_485" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="125" name="Reshape_487" type="Reshape" version="opset1">
+		<layer id="124" name="Reshape_486" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -1824,7 +1803,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="125" name="onnx::MatMul_534/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -1835,7 +1829,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -1859,20 +1853,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="128" name="/blocks.0/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158165" size="32" />
+			<data element_type="i64" shape="4" offset="2926165" 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>384</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>6</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="130" name="Constant_150" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158197" size="32" />
+			<data element_type="i64" shape="4" offset="2926197" 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>6</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>6</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>6</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>6</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>6</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="133" name="Constant_157" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158197" size="32" />
+			<data element_type="i64" shape="4" offset="2926197" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -1967,7 +1961,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>6</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>6</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>6</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="2079036" size="8" />
+			<data element_type="i64" shape="1" offset="2847036" 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="2079036" size="8" />
+			<data element_type="i64" shape="1" offset="2847036" 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="2079028" size="8" />
+			<data element_type="i64" shape="1" offset="2847028" 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>6</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="146" name="onnx::MatMul_552" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="2311869" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="3079871" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -2140,7 +2134,7 @@
 			</output>
 		</layer>
 		<layer id="147" name="Convert_511" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -2149,7 +2143,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -2157,7 +2151,7 @@
 			</output>
 		</layer>
 		<layer id="148" name="onnx::MatMul_552/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="2999965" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -2165,32 +2159,32 @@
 			</output>
 		</layer>
 		<layer id="149" name="Convert_513" 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_514" type="Subtract" version="opset1">
+		<layer id="150" name="onnx::MatMul_552/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -2198,7 +2192,7 @@
 			</output>
 		</layer>
 		<layer id="151" name="onnx::MatMul_552/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="2385597" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="3153599" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -2207,57 +2201,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="152" name="onnx::MatMul_552/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="153" name="onnx::MatMul_552/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="152" name="onnx::MatMul_552/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="154" name="Constant_519" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="153" name="Constant_518" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="155" name="Reshape_520" type="Reshape" version="opset1">
+		<layer id="154" name="Reshape_519" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -2267,7 +2244,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="155" name="onnx::MatMul_552/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -2278,7 +2270,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -2302,14 +2294,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -2319,19 +2311,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -2341,20 +2333,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="160" name="onnx::Add_557" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="2387901" size="1536" />
+			<data element_type="f32" shape="384" offset="3155903" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_557">
 					<dim>384</dim>
@@ -2362,7 +2354,7 @@
 			</output>
 		</layer>
 		<layer id="161" name="onnx::Add_554" type="Const" version="opset1">
-			<data element_type="f32" shape="1536" offset="2389437" size="6144" />
+			<data element_type="f32" shape="1536" offset="3157439" size="6144" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_554">
 					<dim>1536</dim>
@@ -2370,13 +2362,13 @@
 			</output>
 		</layer>
 		<layer id="162" name="Constant_183" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079028" size="8" />
+			<data element_type="i64" shape="" offset="2847028" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="163" name="Constant_182" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079036" size="8" />
+			<data element_type="i64" shape="" offset="2847036" 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>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="167" name="blocks.0.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="2395581" size="1536" />
+			<data element_type="f32" shape="384" offset="3163583" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.0.mlp_ln.weight">
 					<dim>384</dim>
@@ -2433,7 +2425,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="169" name="blocks.0.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="2397117" size="1536" />
+			<data element_type="f32" shape="384" offset="3165119" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.0.mlp_ln.bias">
 					<dim>384</dim>
@@ -2461,7 +2453,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -2481,20 +2473,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="172" name="onnx::MatMul_555" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 1536" offset="2398653" size="294912" />
+			<data element_type="u4" shape="3, 128, 1536" offset="3166655" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -2504,7 +2496,7 @@
 			</output>
 		</layer>
 		<layer id="173" name="Convert_522" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -2513,7 +2505,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
@@ -2521,7 +2513,7 @@
 			</output>
 		</layer>
 		<layer id="174" name="onnx::MatMul_555/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3461567" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -2529,32 +2521,32 @@
 			</output>
 		</layer>
 		<layer id="175" name="Convert_524" 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_525" type="Subtract" version="opset1">
+		<layer id="176" name="onnx::MatMul_555/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</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>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
@@ -2562,7 +2554,7 @@
 			</output>
 		</layer>
 		<layer id="177" name="onnx::MatMul_555/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 1536" offset="2693565" size="9216" />
+			<data element_type="f16" shape="3, 1, 1536" offset="3461568" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -2571,57 +2563,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="178" name="onnx::MatMul_555/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="178" name="onnx::MatMul_555/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
-					<dim>1</dim>
+					<dim>128</dim>
 					<dim>1536</dim>
 				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>1536</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="179" name="onnx::MatMul_555/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>3</dim>
-					<dim>128</dim>
-					<dim>1536</dim>
-				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>1536</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="180" name="Constant_530" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2702781" size="16" />
+		<layer id="179" name="Constant_529" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="3470784" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="181" name="Reshape_531" type="Reshape" version="opset1">
+		<layer id="180" name="Reshape_530" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
@@ -2631,7 +2606,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>1536</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="181" name="onnx::MatMul_555/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>1536</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>1536</dim>
 				</port>
@@ -2642,7 +2632,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>1536</dim>
 				</port>
 			</output>
@@ -2666,14 +2656,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -2689,7 +2679,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -2706,14 +2696,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -2729,7 +2719,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -2747,19 +2737,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -2775,7 +2765,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="192" name="onnx::MatMul_558" type="Const" version="opset1">
-			<data element_type="u4" shape="12, 128, 384" offset="2702797" size="294912" />
+			<data element_type="u4" shape="12, 128, 384" offset="3470800" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>12</dim>
@@ -2799,7 +2789,7 @@
 			</output>
 		</layer>
 		<layer id="193" name="Convert_533" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>12</dim>
@@ -2808,7 +2798,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -2816,7 +2806,7 @@
 			</output>
 		</layer>
 		<layer id="194" name="onnx::MatMul_558/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3461567" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -2824,32 +2814,32 @@
 			</output>
 		</layer>
 		<layer id="195" name="Convert_535" 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_536" type="Subtract" version="opset1">
+		<layer id="196" name="onnx::MatMul_558/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</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>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -2857,7 +2847,7 @@
 			</output>
 		</layer>
 		<layer id="197" name="onnx::MatMul_558/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="12, 1, 384" offset="2997709" size="9216" />
+			<data element_type="f16" shape="12, 1, 384" offset="3765712" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>12</dim>
@@ -2866,57 +2856,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="198" name="onnx::MatMul_558/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>12</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>12</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="199" name="onnx::MatMul_558/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="198" name="onnx::MatMul_558/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>12</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="200" name="Constant_541" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3006925" size="16" />
+		<layer id="199" name="Constant_540" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="3774928" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="201" name="Reshape_542" type="Reshape" version="opset1">
+		<layer id="200" name="Reshape_541" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -2926,7 +2899,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>1536</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="201" name="onnx::MatMul_558/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>1536</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>1536</dim>
 					<dim>384</dim>
 				</port>
@@ -2937,7 +2925,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>384</dim>
 				</port>
 			</output>
@@ -2961,14 +2949,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -2978,19 +2966,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -3000,20 +2988,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="206" name="onnx::Add_583" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="3006941" size="1536" />
+			<data element_type="f32" shape="384" offset="3774944" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_583">
 					<dim>384</dim>
@@ -3021,7 +3009,7 @@
 			</output>
 		</layer>
 		<layer id="207" name="onnx::Add_560" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="3008477" size="1536" />
+			<data element_type="f32" shape="384" offset="3776480" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_560">
 					<dim>384</dim>
@@ -3029,13 +3017,13 @@
 			</output>
 		</layer>
 		<layer id="208" name="Constant_205" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079028" size="8" />
+			<data element_type="i64" shape="" offset="2847028" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="209" name="Constant_204" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079036" size="8" />
+			<data element_type="i64" shape="" offset="2847036" 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>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="213" name="blocks.1.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="3010013" size="1536" />
+			<data element_type="f32" shape="384" offset="3778016" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.1.attn_ln.weight">
 					<dim>384</dim>
@@ -3092,7 +3080,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="215" name="blocks.1.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="3011549" size="1536" />
+			<data element_type="f32" shape="384" offset="3779552" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.1.attn_ln.bias">
 					<dim>384</dim>
@@ -3120,7 +3108,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -3140,20 +3128,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="218" name="onnx::MatMul_561" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="3013085" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="3781088" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -3163,7 +3151,7 @@
 			</output>
 		</layer>
 		<layer id="219" name="Convert_566" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -3172,7 +3160,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -3180,7 +3168,7 @@
 			</output>
 		</layer>
 		<layer id="220" name="onnx::MatMul_561/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3854816" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -3188,32 +3176,32 @@
 			</output>
 		</layer>
 		<layer id="221" name="Convert_568" 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_569" type="Subtract" version="opset1">
+		<layer id="222" name="onnx::MatMul_561/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -3221,7 +3209,7 @@
 			</output>
 		</layer>
 		<layer id="223" name="onnx::MatMul_561/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="3086813" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="3854817" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -3230,57 +3218,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="224" name="onnx::MatMul_561/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="225" name="onnx::MatMul_561/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="224" name="onnx::MatMul_561/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="226" name="Constant_574" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="225" name="Constant_573" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="227" name="Reshape_575" type="Reshape" version="opset1">
+		<layer id="226" name="Reshape_574" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -3290,7 +3261,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="227" name="onnx::MatMul_561/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -3301,7 +3287,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -3325,20 +3311,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="230" name="/blocks.1/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158165" size="32" />
+			<data element_type="i64" shape="4" offset="2926165" 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>384</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>6</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="232" name="Constant_221" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158197" size="32" />
+			<data element_type="i64" shape="4" offset="2926197" 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>6</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>6</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="2158229" size="4" />
+			<data element_type="f32" shape="" offset="2926229" 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="2158233" size="4" />
+			<data element_type="f32" shape="" offset="2926233" 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>6</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>6</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="238" name="onnx::MatMul_563" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="3089117" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="3857121" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -3448,7 +3434,7 @@
 			</output>
 		</layer>
 		<layer id="239" name="Convert_555" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -3457,7 +3443,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -3465,7 +3451,7 @@
 			</output>
 		</layer>
 		<layer id="240" name="onnx::MatMul_563/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="2923844" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -3473,32 +3459,32 @@
 			</output>
 		</layer>
 		<layer id="241" name="Convert_557" 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_558" type="Subtract" version="opset1">
+		<layer id="242" name="onnx::MatMul_563/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -3506,7 +3492,7 @@
 			</output>
 		</layer>
 		<layer id="243" name="onnx::MatMul_563/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="3162845" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="3930849" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -3515,57 +3501,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="244" name="onnx::MatMul_563/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="245" name="onnx::MatMul_563/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="244" name="onnx::MatMul_563/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="246" name="Constant_563" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="245" name="Constant_562" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="247" name="Reshape_564" type="Reshape" version="opset1">
+		<layer id="246" name="Reshape_563" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -3575,7 +3544,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="247" name="onnx::MatMul_563/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -3586,7 +3570,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="249" name="/blocks.1/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158165" size="32" />
+			<data element_type="i64" shape="4" offset="2926165" 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>384</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>6</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="251" name="Constant_226" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2234269" size="32" />
+			<data element_type="i64" shape="4" offset="3002270" 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>6</dim>
 					<dim>64</dim>
 				</port>
@@ -3656,7 +3640,7 @@
 					<dim>1</dim>
 					<dim>6</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -3667,7 +3651,7 @@
 					<dim>1</dim>
 					<dim>6</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>6</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>6</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>6</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>6</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>6</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>6</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>6</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>6</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>6</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>6</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="258" name="onnx::Add_565" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="3165149" size="1536" />
+			<data element_type="f32" shape="384" offset="3933153" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_565">
 					<dim>384</dim>
@@ -3771,7 +3755,7 @@
 			</output>
 		</layer>
 		<layer id="259" name="onnx::MatMul_566" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="3166685" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="3934689" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -3781,7 +3765,7 @@
 			</output>
 		</layer>
 		<layer id="260" name="Convert_544" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -3790,7 +3774,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -3798,7 +3782,7 @@
 			</output>
 		</layer>
 		<layer id="261" name="onnx::MatMul_566/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3077566" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -3806,32 +3790,32 @@
 			</output>
 		</layer>
 		<layer id="262" name="Convert_546" 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_547" type="Subtract" version="opset1">
+		<layer id="263" name="onnx::MatMul_566/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -3839,7 +3823,7 @@
 			</output>
 		</layer>
 		<layer id="264" name="onnx::MatMul_566/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="3240413" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="4008417" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -3848,57 +3832,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="265" name="onnx::MatMul_566/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="266" name="onnx::MatMul_566/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="265" name="onnx::MatMul_566/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="267" name="Constant_552" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="266" name="Constant_551" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="268" name="Reshape_553" type="Reshape" version="opset1">
+		<layer id="267" name="Reshape_552" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -3908,7 +3875,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="268" name="onnx::MatMul_566/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -3919,7 +3901,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -3943,20 +3925,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="271" name="/blocks.1/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158165" size="32" />
+			<data element_type="i64" shape="4" offset="2926165" 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>384</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>6</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="273" name="Constant_231" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158197" size="32" />
+			<data element_type="i64" shape="4" offset="2926197" 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>6</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>6</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>6</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>6</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>6</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="276" name="Constant_238" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158197" size="32" />
+			<data element_type="i64" shape="4" offset="2926197" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -4051,7 +4033,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>6</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>6</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>6</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="2079036" size="8" />
+			<data element_type="i64" shape="1" offset="2847036" 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="2079036" size="8" />
+			<data element_type="i64" shape="1" offset="2847036" 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="2079028" size="8" />
+			<data element_type="i64" shape="1" offset="2847028" 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>6</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="289" name="onnx::MatMul_584" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="3242717" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="4010721" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -4224,7 +4206,7 @@
 			</output>
 		</layer>
 		<layer id="290" name="Convert_577" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -4233,7 +4215,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -4241,7 +4223,7 @@
 			</output>
 		</layer>
 		<layer id="291" name="onnx::MatMul_584/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3461567" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -4249,32 +4231,32 @@
 			</output>
 		</layer>
 		<layer id="292" name="Convert_579" 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_580" type="Subtract" version="opset1">
+		<layer id="293" name="onnx::MatMul_584/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -4282,7 +4264,7 @@
 			</output>
 		</layer>
 		<layer id="294" name="onnx::MatMul_584/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="3316445" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="4084449" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -4291,57 +4273,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="295" name="onnx::MatMul_584/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="296" name="onnx::MatMul_584/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="295" name="onnx::MatMul_584/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="297" name="Constant_585" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="296" name="Constant_584" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="298" name="Reshape_586" type="Reshape" version="opset1">
+		<layer id="297" name="Reshape_585" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -4351,7 +4316,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="298" name="onnx::MatMul_584/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -4362,7 +4342,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -4386,14 +4366,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -4403,19 +4383,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -4425,20 +4405,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="303" name="onnx::Add_589" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="3318749" size="1536" />
+			<data element_type="f32" shape="384" offset="4086753" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_589">
 					<dim>384</dim>
@@ -4446,7 +4426,7 @@
 			</output>
 		</layer>
 		<layer id="304" name="onnx::Add_586" type="Const" version="opset1">
-			<data element_type="f32" shape="1536" offset="3320285" size="6144" />
+			<data element_type="f32" shape="1536" offset="4088289" size="6144" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_586">
 					<dim>1536</dim>
@@ -4454,13 +4434,13 @@
 			</output>
 		</layer>
 		<layer id="305" name="Constant_264" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079028" size="8" />
+			<data element_type="i64" shape="" offset="2847028" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="306" name="Constant_263" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079036" size="8" />
+			<data element_type="i64" shape="" offset="2847036" 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>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="310" name="blocks.1.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="3326429" size="1536" />
+			<data element_type="f32" shape="384" offset="4094433" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.1.mlp_ln.weight">
 					<dim>384</dim>
@@ -4517,7 +4497,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="312" name="blocks.1.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="3327965" size="1536" />
+			<data element_type="f32" shape="384" offset="4095969" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.1.mlp_ln.bias">
 					<dim>384</dim>
@@ -4545,7 +4525,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -4565,20 +4545,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="315" name="onnx::MatMul_587" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 1536" offset="3329501" size="294912" />
+			<data element_type="u4" shape="3, 128, 1536" offset="4097505" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -4588,7 +4568,7 @@
 			</output>
 		</layer>
 		<layer id="316" name="Convert_588" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -4597,7 +4577,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
@@ -4605,7 +4585,7 @@
 			</output>
 		</layer>
 		<layer id="317" name="onnx::MatMul_587/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="4392417" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -4613,32 +4593,32 @@
 			</output>
 		</layer>
 		<layer id="318" name="Convert_590" 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_591" type="Subtract" version="opset1">
+		<layer id="319" name="onnx::MatMul_587/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</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>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
@@ -4646,7 +4626,7 @@
 			</output>
 		</layer>
 		<layer id="320" name="onnx::MatMul_587/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 1536" offset="3624413" size="9216" />
+			<data element_type="f16" shape="3, 1, 1536" offset="4392418" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -4655,57 +4635,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="321" name="onnx::MatMul_587/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>1536</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>1536</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="322" name="onnx::MatMul_587/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="321" name="onnx::MatMul_587/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>1536</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="323" name="Constant_596" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2702781" size="16" />
+		<layer id="322" name="Constant_595" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="3470784" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="324" name="Reshape_597" type="Reshape" version="opset1">
+		<layer id="323" name="Reshape_596" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
@@ -4715,7 +4678,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>1536</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="324" name="onnx::MatMul_587/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>1536</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>1536</dim>
 				</port>
@@ -4726,7 +4704,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>1536</dim>
 				</port>
 			</output>
@@ -4750,14 +4728,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -4773,7 +4751,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -4790,14 +4768,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -4813,7 +4791,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -4831,19 +4809,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -4859,7 +4837,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="335" name="onnx::MatMul_590" type="Const" version="opset1">
-			<data element_type="u4" shape="12, 128, 384" offset="3633629" size="294912" />
+			<data element_type="u4" shape="12, 128, 384" offset="4401634" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>12</dim>
@@ -4883,7 +4861,7 @@
 			</output>
 		</layer>
 		<layer id="336" name="Convert_599" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>12</dim>
@@ -4892,7 +4870,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -4900,7 +4878,7 @@
 			</output>
 		</layer>
 		<layer id="337" name="onnx::MatMul_590/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="4392417" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -4908,32 +4886,32 @@
 			</output>
 		</layer>
 		<layer id="338" name="Convert_601" 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_602" type="Subtract" version="opset1">
+		<layer id="339" name="onnx::MatMul_590/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</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>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -4941,7 +4919,7 @@
 			</output>
 		</layer>
 		<layer id="340" name="onnx::MatMul_590/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="12, 1, 384" offset="3928541" size="9216" />
+			<data element_type="f16" shape="12, 1, 384" offset="4696546" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>12</dim>
@@ -4950,57 +4928,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="341" name="onnx::MatMul_590/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>12</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>12</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="342" name="onnx::MatMul_590/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="341" name="onnx::MatMul_590/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>12</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="343" name="Constant_607" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3006925" size="16" />
+		<layer id="342" name="Constant_606" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="3774928" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="344" name="Reshape_608" type="Reshape" version="opset1">
+		<layer id="343" name="Reshape_607" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -5010,7 +4971,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>1536</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="344" name="onnx::MatMul_590/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>1536</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>1536</dim>
 					<dim>384</dim>
 				</port>
@@ -5021,7 +4997,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>384</dim>
 				</port>
 			</output>
@@ -5045,14 +5021,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -5062,19 +5038,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -5084,20 +5060,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="349" name="onnx::Add_615" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="3937757" size="1536" />
+			<data element_type="f32" shape="384" offset="4705762" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_615">
 					<dim>384</dim>
@@ -5105,7 +5081,7 @@
 			</output>
 		</layer>
 		<layer id="350" name="onnx::Add_592" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="3939293" size="1536" />
+			<data element_type="f32" shape="384" offset="4707298" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_592">
 					<dim>384</dim>
@@ -5113,13 +5089,13 @@
 			</output>
 		</layer>
 		<layer id="351" name="Constant_286" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079028" size="8" />
+			<data element_type="i64" shape="" offset="2847028" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="352" name="Constant_285" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079036" size="8" />
+			<data element_type="i64" shape="" offset="2847036" 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>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="356" name="blocks.2.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="3940829" size="1536" />
+			<data element_type="f32" shape="384" offset="4708834" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.2.attn_ln.weight">
 					<dim>384</dim>
@@ -5176,7 +5152,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="358" name="blocks.2.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="3942365" size="1536" />
+			<data element_type="f32" shape="384" offset="4710370" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.2.attn_ln.bias">
 					<dim>384</dim>
@@ -5204,7 +5180,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -5224,20 +5200,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="361" name="onnx::MatMul_593" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="3943901" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="4711906" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -5247,7 +5223,7 @@
 			</output>
 		</layer>
 		<layer id="362" name="Convert_632" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -5256,7 +5232,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -5264,7 +5240,7 @@
 			</output>
 		</layer>
 		<layer id="363" name="onnx::MatMul_593/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3077566" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -5272,32 +5248,32 @@
 			</output>
 		</layer>
 		<layer id="364" name="Convert_634" 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_635" type="Subtract" version="opset1">
+		<layer id="365" name="onnx::MatMul_593/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -5305,7 +5281,7 @@
 			</output>
 		</layer>
 		<layer id="366" name="onnx::MatMul_593/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="4017629" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="4785634" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -5314,57 +5290,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="367" name="onnx::MatMul_593/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="368" name="onnx::MatMul_593/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="367" name="onnx::MatMul_593/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="369" name="Constant_640" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="368" name="Constant_639" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="370" name="Reshape_641" type="Reshape" version="opset1">
+		<layer id="369" name="Reshape_640" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -5374,7 +5333,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="370" name="onnx::MatMul_593/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -5385,7 +5359,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -5409,20 +5383,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="373" name="/blocks.2/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158165" size="32" />
+			<data element_type="i64" shape="4" offset="2926165" 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>384</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>6</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="375" name="Constant_302" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158197" size="32" />
+			<data element_type="i64" shape="4" offset="2926197" 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>6</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>6</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="2158229" size="4" />
+			<data element_type="f32" shape="" offset="2926229" 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="2158233" size="4" />
+			<data element_type="f32" shape="" offset="2926233" 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>6</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>6</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="381" name="onnx::MatMul_595" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="4019933" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="4787938" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -5532,7 +5506,7 @@
 			</output>
 		</layer>
 		<layer id="382" name="Convert_621" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -5541,7 +5515,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -5549,7 +5523,7 @@
 			</output>
 		</layer>
 		<layer id="383" name="onnx::MatMul_595/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3077566" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -5557,32 +5531,32 @@
 			</output>
 		</layer>
 		<layer id="384" name="Convert_623" 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_624" type="Subtract" version="opset1">
+		<layer id="385" name="onnx::MatMul_595/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -5590,7 +5564,7 @@
 			</output>
 		</layer>
 		<layer id="386" name="onnx::MatMul_595/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="4093661" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="4861666" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -5599,57 +5573,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="387" name="onnx::MatMul_595/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="388" name="onnx::MatMul_595/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="387" name="onnx::MatMul_595/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="389" name="Constant_629" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="388" name="Constant_628" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="390" name="Reshape_630" type="Reshape" version="opset1">
+		<layer id="389" name="Reshape_629" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -5659,7 +5616,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="390" name="onnx::MatMul_595/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -5670,7 +5642,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="392" name="/blocks.2/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158165" size="32" />
+			<data element_type="i64" shape="4" offset="2926165" 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>384</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>6</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="394" name="Constant_307" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2234269" size="32" />
+			<data element_type="i64" shape="4" offset="3002270" 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>6</dim>
 					<dim>64</dim>
 				</port>
@@ -5740,7 +5712,7 @@
 					<dim>1</dim>
 					<dim>6</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -5751,7 +5723,7 @@
 					<dim>1</dim>
 					<dim>6</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>6</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>6</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>6</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>6</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>6</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>6</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>6</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>6</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>6</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>6</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="401" name="onnx::Add_597" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="4095965" size="1536" />
+			<data element_type="f32" shape="384" offset="4863970" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_597">
 					<dim>384</dim>
@@ -5855,7 +5827,7 @@
 			</output>
 		</layer>
 		<layer id="402" name="onnx::MatMul_598" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="4097501" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="4865506" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -5865,7 +5837,7 @@
 			</output>
 		</layer>
 		<layer id="403" name="Convert_610" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -5874,7 +5846,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -5882,7 +5854,7 @@
 			</output>
 		</layer>
 		<layer id="404" name="onnx::MatMul_598/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="4939234" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -5890,32 +5862,32 @@
 			</output>
 		</layer>
 		<layer id="405" name="Convert_612" 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_613" type="Subtract" version="opset1">
+		<layer id="406" name="onnx::MatMul_598/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -5923,7 +5895,7 @@
 			</output>
 		</layer>
 		<layer id="407" name="onnx::MatMul_598/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="4171229" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="4939235" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -5932,57 +5904,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="408" name="onnx::MatMul_598/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="409" name="onnx::MatMul_598/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="408" name="onnx::MatMul_598/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="410" name="Constant_618" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="409" name="Constant_617" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="411" name="Reshape_619" type="Reshape" version="opset1">
+		<layer id="410" name="Reshape_618" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -5992,7 +5947,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="411" name="onnx::MatMul_598/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -6003,7 +5973,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -6027,20 +5997,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="414" name="/blocks.2/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158165" size="32" />
+			<data element_type="i64" shape="4" offset="2926165" 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>384</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>6</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="416" name="Constant_312" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158197" size="32" />
+			<data element_type="i64" shape="4" offset="2926197" 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>6</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>6</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>6</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>6</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>6</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="419" name="Constant_319" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158197" size="32" />
+			<data element_type="i64" shape="4" offset="2926197" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -6135,7 +6105,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>6</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>6</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>6</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="2079036" size="8" />
+			<data element_type="i64" shape="1" offset="2847036" 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="2079036" size="8" />
+			<data element_type="i64" shape="1" offset="2847036" 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="2079028" size="8" />
+			<data element_type="i64" shape="1" offset="2847028" 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>6</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="432" name="onnx::MatMul_616" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="4173533" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="4941539" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -6308,7 +6278,7 @@
 			</output>
 		</layer>
 		<layer id="433" name="Convert_643" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -6317,7 +6287,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -6325,7 +6295,7 @@
 			</output>
 		</layer>
 		<layer id="434" name="onnx::MatMul_616/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3077566" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -6333,32 +6303,32 @@
 			</output>
 		</layer>
 		<layer id="435" name="Convert_645" 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_646" type="Subtract" version="opset1">
+		<layer id="436" name="onnx::MatMul_616/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -6366,7 +6336,7 @@
 			</output>
 		</layer>
 		<layer id="437" name="onnx::MatMul_616/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="4247261" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="5015267" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -6375,57 +6345,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="438" name="onnx::MatMul_616/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="439" name="onnx::MatMul_616/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="438" name="onnx::MatMul_616/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="440" name="Constant_651" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="439" name="Constant_650" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="441" name="Reshape_652" type="Reshape" version="opset1">
+		<layer id="440" name="Reshape_651" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -6435,7 +6388,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="441" name="onnx::MatMul_616/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -6446,7 +6414,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -6470,14 +6438,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -6487,19 +6455,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -6509,20 +6477,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="446" name="onnx::Add_621" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="4249565" size="1536" />
+			<data element_type="f32" shape="384" offset="5017571" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_621">
 					<dim>384</dim>
@@ -6530,7 +6498,7 @@
 			</output>
 		</layer>
 		<layer id="447" name="onnx::Add_618" type="Const" version="opset1">
-			<data element_type="f32" shape="1536" offset="4251101" size="6144" />
+			<data element_type="f32" shape="1536" offset="5019107" size="6144" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_618">
 					<dim>1536</dim>
@@ -6538,13 +6506,13 @@
 			</output>
 		</layer>
 		<layer id="448" name="Constant_345" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079028" size="8" />
+			<data element_type="i64" shape="" offset="2847028" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="449" name="Constant_344" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079036" size="8" />
+			<data element_type="i64" shape="" offset="2847036" 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>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="453" name="blocks.2.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="4257245" size="1536" />
+			<data element_type="f32" shape="384" offset="5025251" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.2.mlp_ln.weight">
 					<dim>384</dim>
@@ -6601,7 +6569,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="455" name="blocks.2.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="4258781" size="1536" />
+			<data element_type="f32" shape="384" offset="5026787" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.2.mlp_ln.bias">
 					<dim>384</dim>
@@ -6629,7 +6597,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -6649,20 +6617,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="458" name="onnx::MatMul_619" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 1536" offset="4260317" size="294912" />
+			<data element_type="u4" shape="3, 128, 1536" offset="5028323" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -6672,7 +6640,7 @@
 			</output>
 		</layer>
 		<layer id="459" name="Convert_654" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -6681,7 +6649,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
@@ -6689,7 +6657,7 @@
 			</output>
 		</layer>
 		<layer id="460" name="onnx::MatMul_619/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3461567" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -6697,32 +6665,32 @@
 			</output>
 		</layer>
 		<layer id="461" name="Convert_656" 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_657" type="Subtract" version="opset1">
+		<layer id="462" name="onnx::MatMul_619/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</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>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
@@ -6730,7 +6698,7 @@
 			</output>
 		</layer>
 		<layer id="463" name="onnx::MatMul_619/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 1536" offset="4555229" size="9216" />
+			<data element_type="f16" shape="3, 1, 1536" offset="5323235" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -6739,57 +6707,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="464" name="onnx::MatMul_619/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>1536</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>1536</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="465" name="onnx::MatMul_619/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="464" name="onnx::MatMul_619/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>1536</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="466" name="Constant_662" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2702781" size="16" />
+		<layer id="465" name="Constant_661" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="3470784" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="467" name="Reshape_663" type="Reshape" version="opset1">
+		<layer id="466" name="Reshape_662" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
@@ -6799,7 +6750,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>1536</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="467" name="onnx::MatMul_619/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>1536</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>1536</dim>
 				</port>
@@ -6810,7 +6776,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>1536</dim>
 				</port>
 			</output>
@@ -6834,14 +6800,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -6857,7 +6823,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -6874,14 +6840,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -6897,7 +6863,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -6915,19 +6881,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -6943,7 +6909,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="478" name="onnx::MatMul_622" type="Const" version="opset1">
-			<data element_type="u4" shape="12, 128, 384" offset="4564445" size="294912" />
+			<data element_type="u4" shape="12, 128, 384" offset="5332451" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>12</dim>
@@ -6967,7 +6933,7 @@
 			</output>
 		</layer>
 		<layer id="479" name="Convert_665" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>12</dim>
@@ -6976,7 +6942,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -6984,7 +6950,7 @@
 			</output>
 		</layer>
 		<layer id="480" name="onnx::MatMul_622/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3077566" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -6992,32 +6958,32 @@
 			</output>
 		</layer>
 		<layer id="481" name="Convert_667" 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_668" type="Subtract" version="opset1">
+		<layer id="482" name="onnx::MatMul_622/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</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>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -7025,7 +6991,7 @@
 			</output>
 		</layer>
 		<layer id="483" name="onnx::MatMul_622/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="12, 1, 384" offset="4859357" size="9216" />
+			<data element_type="f16" shape="12, 1, 384" offset="5627363" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>12</dim>
@@ -7034,57 +7000,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="484" name="onnx::MatMul_622/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>12</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>12</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="485" name="onnx::MatMul_622/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="484" name="onnx::MatMul_622/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>12</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="486" name="Constant_673" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="3006925" size="16" />
+		<layer id="485" name="Constant_672" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="3774928" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="487" name="Reshape_674" type="Reshape" version="opset1">
+		<layer id="486" name="Reshape_673" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>12</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -7094,7 +7043,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>1536</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="487" name="onnx::MatMul_622/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>1536</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>1536</dim>
 					<dim>384</dim>
 				</port>
@@ -7105,7 +7069,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>384</dim>
 				</port>
 			</output>
@@ -7129,14 +7093,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -7146,19 +7110,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -7168,20 +7132,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="492" name="onnx::Add_647" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="4868573" size="1536" />
+			<data element_type="f32" shape="384" offset="5636579" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_647">
 					<dim>384</dim>
@@ -7189,7 +7153,7 @@
 			</output>
 		</layer>
 		<layer id="493" name="onnx::Add_624" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="4870109" size="1536" />
+			<data element_type="f32" shape="384" offset="5638115" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_624">
 					<dim>384</dim>
@@ -7197,13 +7161,13 @@
 			</output>
 		</layer>
 		<layer id="494" name="Constant_367" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079028" size="8" />
+			<data element_type="i64" shape="" offset="2847028" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="495" name="Constant_366" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079036" size="8" />
+			<data element_type="i64" shape="" offset="2847036" 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>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="499" name="blocks.3.attn_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="4871645" size="1536" />
+			<data element_type="f32" shape="384" offset="5639651" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.3.attn_ln.weight">
 					<dim>384</dim>
@@ -7260,7 +7224,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="501" name="blocks.3.attn_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="4873181" size="1536" />
+			<data element_type="f32" shape="384" offset="5641187" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.3.attn_ln.bias">
 					<dim>384</dim>
@@ -7288,7 +7252,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -7308,20 +7272,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="504" name="onnx::MatMul_625" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="4874717" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="5642723" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -7331,7 +7295,7 @@
 			</output>
 		</layer>
 		<layer id="505" name="Convert_698" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -7340,7 +7304,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -7348,7 +7312,7 @@
 			</output>
 		</layer>
 		<layer id="506" name="onnx::MatMul_625/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="4392417" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -7356,32 +7320,32 @@
 			</output>
 		</layer>
 		<layer id="507" name="Convert_700" 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_701" type="Subtract" version="opset1">
+		<layer id="508" name="onnx::MatMul_625/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -7389,7 +7353,7 @@
 			</output>
 		</layer>
 		<layer id="509" name="onnx::MatMul_625/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="4948445" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="5716451" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -7398,57 +7362,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="510" name="onnx::MatMul_625/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="511" name="onnx::MatMul_625/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="510" name="onnx::MatMul_625/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="512" name="Constant_706" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="511" name="Constant_705" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="513" name="Reshape_707" type="Reshape" version="opset1">
+		<layer id="512" name="Reshape_706" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -7458,7 +7405,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="513" name="onnx::MatMul_625/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -7469,7 +7431,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -7493,20 +7455,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="516" name="/blocks.3/attn/Constant_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158165" size="32" />
+			<data element_type="i64" shape="4" offset="2926165" 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>384</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>6</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="518" name="Constant_383" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158197" size="32" />
+			<data element_type="i64" shape="4" offset="2926197" 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>6</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>6</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="2158229" size="4" />
+			<data element_type="f32" shape="" offset="2926229" 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="2158233" size="4" />
+			<data element_type="f32" shape="" offset="2926233" 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>6</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>6</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="524" name="onnx::MatMul_627" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="4950749" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="5718755" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -7616,7 +7578,7 @@
 			</output>
 		</layer>
 		<layer id="525" name="Convert_687" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -7625,7 +7587,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -7633,7 +7595,7 @@
 			</output>
 		</layer>
 		<layer id="526" name="onnx::MatMul_627/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3461567" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -7641,32 +7603,32 @@
 			</output>
 		</layer>
 		<layer id="527" name="Convert_689" 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_690" type="Subtract" version="opset1">
+		<layer id="528" name="onnx::MatMul_627/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -7674,7 +7636,7 @@
 			</output>
 		</layer>
 		<layer id="529" name="onnx::MatMul_627/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="5024477" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="5792483" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -7683,57 +7645,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="530" name="onnx::MatMul_627/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="531" name="onnx::MatMul_627/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="530" name="onnx::MatMul_627/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="532" name="Constant_695" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="531" name="Constant_694" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="533" name="Reshape_696" type="Reshape" version="opset1">
+		<layer id="532" name="Reshape_695" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -7743,7 +7688,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="533" name="onnx::MatMul_627/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -7754,7 +7714,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="535" name="/blocks.3/attn/Constant_3" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158165" size="32" />
+			<data element_type="i64" shape="4" offset="2926165" 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>384</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>6</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="537" name="Constant_388" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2234269" size="32" />
+			<data element_type="i64" shape="4" offset="3002270" 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>6</dim>
 					<dim>64</dim>
 				</port>
@@ -7824,7 +7784,7 @@
 					<dim>1</dim>
 					<dim>6</dim>
 					<dim>64</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
@@ -7835,7 +7795,7 @@
 					<dim>1</dim>
 					<dim>6</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>6</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>6</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>6</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>6</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>6</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>6</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>6</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>6</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>6</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>6</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="544" name="onnx::Add_629" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="5026781" size="1536" />
+			<data element_type="f32" shape="384" offset="5794787" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_629">
 					<dim>384</dim>
@@ -7939,7 +7899,7 @@
 			</output>
 		</layer>
 		<layer id="545" name="onnx::MatMul_630" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="5028317" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="5796323" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -7949,7 +7909,7 @@
 			</output>
 		</layer>
 		<layer id="546" name="Convert_676" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -7958,7 +7918,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -7966,7 +7926,7 @@
 			</output>
 		</layer>
 		<layer id="547" name="onnx::MatMul_630/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="2999965" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -7974,32 +7934,32 @@
 			</output>
 		</layer>
 		<layer id="548" name="Convert_678" 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_679" type="Subtract" version="opset1">
+		<layer id="549" name="onnx::MatMul_630/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -8007,7 +7967,7 @@
 			</output>
 		</layer>
 		<layer id="550" name="onnx::MatMul_630/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="5102045" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="5870051" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -8016,57 +7976,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="551" name="onnx::MatMul_630/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="552" name="onnx::MatMul_630/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="551" name="onnx::MatMul_630/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="553" name="Constant_684" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="552" name="Constant_683" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="554" name="Reshape_685" type="Reshape" version="opset1">
+		<layer id="553" name="Reshape_684" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -8076,7 +8019,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="554" name="onnx::MatMul_630/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -8087,7 +8045,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -8111,20 +8069,20 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="557" name="/blocks.3/attn/Constant_4" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158165" size="32" />
+			<data element_type="i64" shape="4" offset="2926165" 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>384</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>6</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="559" name="Constant_393" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158197" size="32" />
+			<data element_type="i64" shape="4" offset="2926197" 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>6</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>6</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>6</dim>
-					<dim>1000</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
+					<dim>1500</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>6</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>6</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="562" name="Constant_400" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="2158197" size="32" />
+			<data element_type="i64" shape="4" offset="2926197" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
@@ -8219,7 +8177,7 @@
 				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>6</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>6</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>6</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="2079036" size="8" />
+			<data element_type="i64" shape="1" offset="2847036" 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="2079036" size="8" />
+			<data element_type="i64" shape="1" offset="2847036" 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="2079028" size="8" />
+			<data element_type="i64" shape="1" offset="2847028" 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>6</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="575" name="onnx::MatMul_648" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 384" offset="5104349" size="73728" />
+			<data element_type="u4" shape="3, 128, 384" offset="5872355" size="73728" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -8392,7 +8350,7 @@
 			</output>
 		</layer>
 		<layer id="576" name="Convert_709" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -8401,7 +8359,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -8409,7 +8367,7 @@
 			</output>
 		</layer>
 		<layer id="577" name="onnx::MatMul_648/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3077566" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -8417,32 +8375,32 @@
 			</output>
 		</layer>
 		<layer id="578" name="Convert_711" 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_712" type="Subtract" version="opset1">
+		<layer id="579" name="onnx::MatMul_648/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</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>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -8450,7 +8408,7 @@
 			</output>
 		</layer>
 		<layer id="580" name="onnx::MatMul_648/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 384" offset="5178077" size="2304" />
+			<data element_type="f16" shape="3, 1, 384" offset="5946083" size="2304" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -8459,57 +8417,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="581" name="onnx::MatMul_648/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="582" name="onnx::MatMul_648/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="581" name="onnx::MatMul_648/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="583" name="Constant_717" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2158149" size="16" />
+		<layer id="582" name="Constant_716" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="2926149" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="584" name="Reshape_718" type="Reshape" version="opset1">
+		<layer id="583" name="Reshape_717" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>384</dim>
@@ -8519,7 +8460,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="584" name="onnx::MatMul_648/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>384</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>384</dim>
 				</port>
@@ -8530,7 +8486,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -8554,14 +8510,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -8571,19 +8527,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -8593,20 +8549,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="589" name="onnx::Add_653" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="5180381" size="1536" />
+			<data element_type="f32" shape="384" offset="5948387" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_653">
 					<dim>384</dim>
@@ -8614,7 +8570,7 @@
 			</output>
 		</layer>
 		<layer id="590" name="onnx::Add_650" type="Const" version="opset1">
-			<data element_type="f32" shape="1536" offset="5181917" size="6144" />
+			<data element_type="f32" shape="1536" offset="5949923" size="6144" />
 			<output>
 				<port id="0" precision="FP32" names="onnx::Add_650">
 					<dim>1536</dim>
@@ -8622,13 +8578,13 @@
 			</output>
 		</layer>
 		<layer id="591" name="Constant_426" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079028" size="8" />
+			<data element_type="i64" shape="" offset="2847028" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="592" name="Constant_425" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079036" size="8" />
+			<data element_type="i64" shape="" offset="2847036" 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>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="596" name="blocks.3.mlp_ln.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="5188061" size="1536" />
+			<data element_type="f32" shape="384" offset="5956067" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.3.mlp_ln.weight">
 					<dim>384</dim>
@@ -8685,7 +8641,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="598" name="blocks.3.mlp_ln.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="5189597" size="1536" />
+			<data element_type="f32" shape="384" offset="5957603" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="blocks.3.mlp_ln.bias">
 					<dim>384</dim>
@@ -8713,7 +8669,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -8733,20 +8689,20 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="601" name="onnx::MatMul_651" type="Const" version="opset1">
-			<data element_type="u4" shape="3, 128, 1536" offset="5191133" size="294912" />
+			<data element_type="u4" shape="3, 128, 1536" offset="5959139" size="294912" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -8756,7 +8712,7 @@
 			</output>
 		</layer>
 		<layer id="602" name="Convert_720" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U4">
 					<dim>3</dim>
@@ -8765,7 +8721,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
@@ -8773,7 +8729,7 @@
 			</output>
 		</layer>
 		<layer id="603" name="onnx::MatMul_651/zero_point" type="Const" version="opset1">
-			<data element_type="u4" shape="1" offset="2155844" size="1" />
+			<data element_type="u4" shape="1" offset="3461567" size="1" />
 			<output>
 				<port id="0" precision="U4">
 					<dim>1</dim>
@@ -8781,32 +8737,32 @@
 			</output>
 		</layer>
 		<layer id="604" name="Convert_722" 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_723" type="Subtract" version="opset1">
+		<layer id="605" name="onnx::MatMul_651/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</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>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
@@ -8814,7 +8770,7 @@
 			</output>
 		</layer>
 		<layer id="606" name="onnx::MatMul_651/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="3, 1, 1536" offset="5486045" size="9216" />
+			<data element_type="f16" shape="3, 1, 1536" offset="6254051" size="9216" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>3</dim>
@@ -8823,57 +8779,40 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="607" name="onnx::MatMul_651/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="FP16">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>1536</dim>
-				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
-					<dim>3</dim>
-					<dim>1</dim>
-					<dim>1536</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="608" name="onnx::MatMul_651/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="607" name="onnx::MatMul_651/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3</dim>
 					<dim>1</dim>
 					<dim>1536</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="609" name="Constant_728" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="2702781" size="16" />
+		<layer id="608" name="Constant_727" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="3470784" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="610" name="Reshape_729" type="Reshape" version="opset1">
+		<layer id="609" name="Reshape_728" type="Reshape" version="opset1">
 			<data special_zero="false" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3</dim>
 					<dim>128</dim>
 					<dim>1536</dim>
@@ -8883,7 +8822,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>384</dim>
+					<dim>1536</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="610" name="onnx::MatMul_651/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>384</dim>
+					<dim>1536</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>384</dim>
 					<dim>1536</dim>
 				</port>
@@ -8894,7 +8848,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>1536</dim>
 				</port>
 			</output>
@@ -8918,14 +8872,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -8941,7 +8895,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -8958,14 +8912,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -8981,7 +8935,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -8999,19 +8953,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
@@ -9027,7 +8981,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</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>1536</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="621" name="onnx::MatMul_654" type="Const" version="opset1">
-			<data element_type="u8" shape="1536, 384" offset="5495261" size="589824" />
+			<data element_type="u8" shape="1536, 384" offset="6263267" size="589824" />
 			<output>
 				<port id="0" precision="U8">
 					<dim>1536</dim>
@@ -9050,7 +9004,7 @@
 			</output>
 		</layer>
 		<layer id="622" name="Convert_732" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>1536</dim>
@@ -9058,14 +9012,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1536</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="623" name="onnx::MatMul_654/zero_point" type="Const" version="opset1">
-			<data element_type="u8" shape="1, 384" offset="6085085" size="384" />
+			<data element_type="u8" shape="1, 384" offset="6853091" size="384" />
 			<output>
 				<port id="0" precision="U8">
 					<dim>1</dim>
@@ -9074,7 +9028,7 @@
 			</output>
 		</layer>
 		<layer id="624" name="Convert_735" type="Convert" version="opset1">
-			<data destination_type="f32" />
+			<data destination_type="f16" />
 			<input>
 				<port id="0" precision="U8">
 					<dim>1</dim>
@@ -9082,33 +9036,33 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="625" name="Subtract_736" type="Subtract" version="opset1">
+		<layer id="625" name="onnx::MatMul_654/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>1536</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>1536</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="626" name="onnx::MatMul_654/scale" type="Const" version="opset1">
-			<data element_type="f16" shape="1, 384" offset="6085469" size="768" />
+			<data element_type="f16" shape="1, 384" offset="6853475" size="768" />
 			<output>
 				<port id="0" precision="FP16">
 					<dim>1</dim>
@@ -9116,35 +9070,35 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="627" name="onnx::MatMul_654/scale_convert" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="627" name="onnx::MatMul_654/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP16">
+					<dim>1536</dim>
+					<dim>384</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
+				<port id="2" precision="FP16">
+					<dim>1536</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="628" name="onnx::MatMul_654/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="628" name="onnx::MatMul_654/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>1536</dim>
 					<dim>384</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>384</dim>
-				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32">
 					<dim>1536</dim>
 					<dim>384</dim>
 				</port>
@@ -9155,7 +9109,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>1536</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -9166,7 +9120,7 @@
 			<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>384</dim>
 				</port>
 			</output>
@@ -9179,14 +9133,14 @@
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -9196,19 +9150,19 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 				<port id="1" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
@@ -9218,26 +9172,26 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</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>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="633" name="Constant_448" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079028" size="8" />
+			<data element_type="i64" shape="" offset="2847028" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
 		<layer id="634" name="Constant_447" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="2079036" size="8" />
+			<data element_type="i64" shape="" offset="2847036" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
@@ -9266,7 +9220,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 				<port id="1" precision="I64">
@@ -9276,13 +9230,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="638" name="ln_post.weight" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="6086237" size="1536" />
+			<data element_type="f32" shape="384" offset="6854243" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="ln_post.weight">
 					<dim>384</dim>
@@ -9294,7 +9248,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -9304,13 +9258,13 @@
 			<output>
 				<port id="2" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 			</output>
 		</layer>
 		<layer id="640" name="ln_post.bias" type="Const" version="opset1">
-			<data element_type="f32" shape="384" offset="6087773" size="1536" />
+			<data element_type="f32" shape="384" offset="6855779" size="1536" />
 			<output>
 				<port id="0" precision="FP32" names="ln_post.bias">
 					<dim>384</dim>
@@ -9322,7 +9276,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 				<port id="1" precision="FP32">
@@ -9332,7 +9286,7 @@
 			<output>
 				<port id="2" precision="FP32" names="/ln_post/LayerNormalization_output_0">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 			</output>
@@ -9342,14 +9296,14 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 			</input>
 			<output>
 				<port id="1" precision="FP32" names="output_features">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 			</output>
@@ -9358,7 +9312,7 @@
 			<input>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
-					<dim>1000</dim>
+					<dim>1500</dim>
 					<dim>384</dim>
 				</port>
 			</input>
@@ -9370,10 +9324,10 @@
 		<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" />
@@ -9402,10 +9356,10 @@
 		<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="41" to-port="0" />
 		<edge from-layer="37" from-port="2" to-layer="48" to-port="0" />
 		<edge from-layer="38" from-port="0" to-layer="40" to-port="0" />
@@ -9455,12 +9409,12 @@
 		<edge from-layer="76" from-port="1" to-layer="79" to-port="0" />
 		<edge from-layer="77" from-port="0" to-layer="78" to-port="0" />
 		<edge from-layer="78" from-port="1" to-layer="79" to-port="1" />
-		<edge from-layer="79" from-port="2" to-layer="82" to-port="0" />
-		<edge from-layer="80" from-port="0" to-layer="81" to-port="0" />
-		<edge from-layer="81" from-port="1" to-layer="82" to-port="1" />
-		<edge from-layer="82" from-port="2" to-layer="84" to-port="0" />
-		<edge from-layer="83" from-port="0" to-layer="84" to-port="1" />
-		<edge from-layer="84" from-port="2" to-layer="85" to-port="1" />
+		<edge from-layer="79" from-port="2" to-layer="81" to-port="0" />
+		<edge from-layer="80" from-port="0" to-layer="81" to-port="1" />
+		<edge from-layer="81" from-port="2" to-layer="83" to-port="0" />
+		<edge from-layer="82" from-port="0" to-layer="83" to-port="1" />
+		<edge from-layer="83" from-port="2" to-layer="84" to-port="0" />
+		<edge from-layer="84" from-port="1" to-layer="85" to-port="1" />
 		<edge from-layer="85" from-port="2" to-layer="86" to-port="1" />
 		<edge from-layer="86" from-port="2" to-layer="88" to-port="0" />
 		<edge from-layer="87" from-port="0" to-layer="88" to-port="1" />
@@ -9476,12 +9430,12 @@
 		<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" />
@@ -9497,12 +9451,12 @@
 		<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" />
@@ -9529,12 +9483,12 @@
 		<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" />
@@ -9556,12 +9510,12 @@
 		<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" />
@@ -9577,12 +9531,12 @@
 		<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" />
@@ -9606,12 +9560,12 @@
 		<edge from-layer="219" from-port="1" to-layer="222" to-port="0" />
 		<edge from-layer="220" from-port="0" to-layer="221" to-port="0" />
 		<edge from-layer="221" from-port="1" to-layer="222" to-port="1" />
-		<edge from-layer="222" from-port="2" to-layer="225" to-port="0" />
-		<edge from-layer="223" from-port="0" to-layer="224" to-port="0" />
-		<edge from-layer="224" from-port="1" to-layer="225" to-port="1" />
-		<edge from-layer="225" from-port="2" to-layer="227" to-port="0" />
-		<edge from-layer="226" from-port="0" to-layer="227" to-port="1" />
-		<edge from-layer="227" from-port="2" to-layer="228" to-port="1" />
+		<edge from-layer="222" from-port="2" to-layer="224" to-port="0" />
+		<edge from-layer="223" from-port="0" to-layer="224" to-port="1" />
+		<edge from-layer="224" from-port="2" to-layer="226" to-port="0" />
+		<edge from-layer="225" from-port="0" to-layer="226" to-port="1" />
+		<edge from-layer="226" from-port="2" to-layer="227" to-port="0" />
+		<edge from-layer="227" from-port="1" to-layer="228" to-port="1" />
 		<edge from-layer="228" from-port="2" to-layer="229" to-port="1" />
 		<edge from-layer="229" from-port="2" to-layer="231" to-port="0" />
 		<edge from-layer="230" from-port="0" to-layer="231" to-port="1" />
@@ -9627,12 +9581,12 @@
 		<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" />
@@ -9648,12 +9602,12 @@
 		<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" />
@@ -9680,12 +9634,12 @@
 		<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" />
@@ -9707,12 +9661,12 @@
 		<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" />
@@ -9728,12 +9682,12 @@
 		<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" />
@@ -9757,12 +9711,12 @@
 		<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" />
@@ -9778,12 +9732,12 @@
 		<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" />
@@ -9799,12 +9753,12 @@
 		<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" />
@@ -9831,12 +9785,12 @@
 		<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" />
@@ -9858,12 +9812,12 @@
 		<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" />
@@ -9879,12 +9833,12 @@
 		<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" />
@@ -9901,19 +9855,19 @@
 		<edge from-layer="500" from-port="2" to-layer="502" to-port="0" />
 		<edge from-layer="501" from-port="0" to-layer="502" to-port="1" />
 		<edge from-layer="502" from-port="2" to-layer="503" to-port="0" />
-		<edge from-layer="503" from-port="1" to-layer="534" to-port="0" />
-		<edge from-layer="503" from-port="1" to-layer="514" to-port="0" />
 		<edge from-layer="503" from-port="1" to-layer="555" to-port="0" />
+		<edge from-layer="503" from-port="1" to-layer="514" to-port="0" />
+		<edge from-layer="503" from-port="1" to-layer="534" to-port="0" />
 		<edge from-layer="504" from-port="0" to-layer="505" to-port="0" />
 		<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" />
@@ -9929,12 +9883,12 @@
 		<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" />
@@ -9950,12 +9904,12 @@
 		<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" />
@@ -9982,12 +9936,12 @@
 		<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" />
@@ -10009,12 +9963,12 @@
 		<edge from-layer="602" from-port="1" to-layer="605" to-port="0" />
 		<edge from-layer="603" from-port="0" to-layer="604" to-port="0" />
 		<edge from-layer="604" from-port="1" to-layer="605" to-port="1" />
-		<edge from-layer="605" from-port="2" to-layer="608" to-port="0" />
-		<edge from-layer="606" from-port="0" to-layer="607" to-port="0" />
-		<edge from-layer="607" from-port="1" to-layer="608" to-port="1" />
-		<edge from-layer="608" from-port="2" to-layer="610" to-port="0" />
-		<edge from-layer="609" from-port="0" to-layer="610" to-port="1" />
-		<edge from-layer="610" from-port="2" to-layer="611" to-port="1" />
+		<edge from-layer="605" from-port="2" to-layer="607" to-port="0" />
+		<edge from-layer="606" from-port="0" to-layer="607" to-port="1" />
+		<edge from-layer="607" from-port="2" to-layer="609" to-port="0" />
+		<edge from-layer="608" from-port="0" to-layer="609" to-port="1" />
+		<edge from-layer="609" from-port="2" to-layer="610" to-port="0" />
+		<edge from-layer="610" from-port="1" to-layer="611" to-port="1" />
 		<edge from-layer="611" from-port="2" to-layer="612" to-port="1" />
 		<edge from-layer="612" from-port="2" to-layer="614" to-port="0" />
 		<edge from-layer="612" from-port="2" to-layer="618" to-port="0" />
@@ -10030,10 +9984,10 @@
 		<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="629" 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="628" to-port="0" />
+		<edge from-layer="628" from-port="1" to-layer="629" to-port="1" />
 		<edge from-layer="629" from-port="2" to-layer="630" to-port="1" />
 		<edge from-layer="630" from-port="2" to-layer="631" to-port="1" />
 		<edge from-layer="631" from-port="2" to-layer="632" to-port="0" />
@@ -10055,16 +10009,12 @@
 			<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>