11004 lines
365 KiB
XML
11004 lines
365 KiB
XML
<?xml version="1.0"?>
|
|
<net name="Model231" version="11">
|
|
<layers>
|
|
<layer id="2" name="input_ids" type="Parameter" version="opset1">
|
|
<data shape="?,?" element_type="i64" />
|
|
<output>
|
|
<port id="0" precision="I64" names="input_ids">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="1" name="attention_mask" type="Parameter" version="opset1">
|
|
<data shape="?,?" element_type="i64" />
|
|
<output>
|
|
<port id="0" precision="I64" names="attention_mask">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="0" name="token_type_ids" type="Parameter" version="opset1">
|
|
<data shape="?,?" element_type="i64" />
|
|
<output>
|
|
<port id="0" precision="I64" names="token_type_ids">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="3" name="Constant_28170404" type="Const" version="opset1">
|
|
<data element_type="i8" shape="30000, 128" offset="0" size="3840000" />
|
|
<output>
|
|
<port id="0" precision="I8">
|
|
<dim>30000</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="4" name="Convert_28170405" type="Convert" version="opset1">
|
|
<data destination_type="f32" />
|
|
<input>
|
|
<port id="0" precision="I8">
|
|
<dim>30000</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32">
|
|
<dim>30000</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="5" name="Constant_28170406" type="Const" version="opset1">
|
|
<data element_type="f32" shape="30000, 1" offset="3840000" size="120000" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>30000</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="6" name="__module.embeddings.word_embeddings/aten::embedding/Gather/fq_weights_0" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>30000</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>30000</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>30000</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="7" name="__module.embeddings.word_embeddings/aten::embedding/Convert" type="Convert" version="opset1">
|
|
<data destination_type="i32" />
|
|
<input>
|
|
<port id="0" precision="I64">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="I32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="8" name="__module.embeddings.word_embeddings/aten::embedding/Constant" type="Const" version="opset1">
|
|
<data element_type="i32" shape="" offset="3960000" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="9" name="__module.embeddings.word_embeddings/aten::embedding/Gather" type="Gather" version="opset8">
|
|
<data batch_dims="0" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>30000</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="2" precision="I32" />
|
|
</input>
|
|
<output>
|
|
<port id="3" precision="FP32" names="58,inputs_embeds">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="10" name="Constant_28170408" type="Const" version="opset1">
|
|
<data element_type="i8" shape="2, 128" offset="3960004" size="256" />
|
|
<output>
|
|
<port id="0" precision="I8">
|
|
<dim>2</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="11" name="Convert_28170409" type="Convert" version="opset1">
|
|
<data destination_type="f32" />
|
|
<input>
|
|
<port id="0" precision="I8">
|
|
<dim>2</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32">
|
|
<dim>2</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="12" name="Constant_28170410" type="Const" version="opset1">
|
|
<data element_type="f32" shape="2, 1" offset="3960260" size="8" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>2</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="13" name="__module.embeddings.token_type_embeddings/aten::embedding/Gather/fq_weights_0" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>2</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>2</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>2</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="14" name="__module.embeddings.token_type_embeddings/aten::embedding/Convert" type="Convert" version="opset1">
|
|
<data destination_type="i32" />
|
|
<input>
|
|
<port id="0" precision="I64">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="I32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="15" name="__module.embeddings.token_type_embeddings/aten::embedding/Constant" type="Const" version="opset1">
|
|
<data element_type="i32" shape="" offset="3960000" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="16" name="__module.embeddings.token_type_embeddings/aten::embedding/Gather" type="Gather" version="opset8">
|
|
<data batch_dims="0" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>2</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="2" precision="I32" />
|
|
</input>
|
|
<output>
|
|
<port id="3" precision="FP32" names="60,token_type_embeddings.1">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="17" name="__module.embeddings/aten::add/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="61_1">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="18" name="Constant_28170412" type="Const" version="opset1">
|
|
<data element_type="i8" shape="512, 128" offset="3960268" size="65536" />
|
|
<output>
|
|
<port id="0" precision="I8">
|
|
<dim>512</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="19" name="Convert_28170413" type="Convert" version="opset1">
|
|
<data destination_type="f32" />
|
|
<input>
|
|
<port id="0" precision="I8">
|
|
<dim>512</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32">
|
|
<dim>512</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="20" name="Constant_28170414" type="Const" version="opset1">
|
|
<data element_type="f32" shape="512, 1" offset="4025804" size="2048" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>512</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="21" name="__module.embeddings.position_embeddings/aten::embedding/Gather/fq_weights_0" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>512</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>512</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>512</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="22" name="__module.embeddings/aten::slice/Slice" type="Const" version="opset1">
|
|
<data element_type="i64" shape="1, 512" offset="4027852" size="4096" />
|
|
<output>
|
|
<port id="0" precision="I64" names="55">
|
|
<dim>1</dim>
|
|
<dim>512</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="23" name="__module.embeddings/aten::slice/Reshape" type="Const" version="opset1">
|
|
<data element_type="i64" shape="1" offset="4031948" size="8" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="24" name="ShapeOf_162951" type="ShapeOf" version="opset3">
|
|
<data output_type="i64" />
|
|
<input>
|
|
<port id="0" precision="I64">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="I64">
|
|
<dim>2</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="25" name="Constant_163062" type="Const" version="opset1">
|
|
<data element_type="i64" shape="1" offset="4031956" size="8" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="26" name="Constant_162953" type="Const" version="opset1">
|
|
<data element_type="i64" shape="" offset="4031948" size="8" />
|
|
<output>
|
|
<port id="0" precision="I64" />
|
|
</output>
|
|
</layer>
|
|
<layer id="27" name="Gather_162954" type="Gather" version="opset8">
|
|
<data batch_dims="0" />
|
|
<input>
|
|
<port id="0" precision="I64">
|
|
<dim>2</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>1</dim>
|
|
</port>
|
|
<port id="2" precision="I64" />
|
|
</input>
|
|
<output>
|
|
<port id="3" precision="I64" names="51,53,54">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="28" name="__module.embeddings/aten::slice/Reshape_2" type="Const" version="opset1">
|
|
<data element_type="i64" shape="1" offset="4031956" size="8" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="29" name="__module.embeddings/aten::slice/Reshape_3" type="Const" version="opset1">
|
|
<data element_type="i64" shape="1" offset="4031956" size="8" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="30" name="__module.embeddings/aten::slice/Slice_1" type="Slice" version="opset8">
|
|
<input>
|
|
<port id="0" precision="I64">
|
|
<dim>1</dim>
|
|
<dim>512</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>1</dim>
|
|
</port>
|
|
<port id="2" precision="I64">
|
|
<dim>1</dim>
|
|
</port>
|
|
<port id="3" precision="I64">
|
|
<dim>1</dim>
|
|
</port>
|
|
<port id="4" precision="I64">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="I64" names="56">
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="31" name="__module.embeddings.position_embeddings/aten::embedding/Convert" type="Convert" version="opset1">
|
|
<data destination_type="i32" />
|
|
<input>
|
|
<port id="0" precision="I64">
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="32" name="__module.embeddings.position_embeddings/aten::embedding/Constant" type="Const" version="opset1">
|
|
<data element_type="i32" shape="" offset="3960000" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="33" name="__module.embeddings.position_embeddings/aten::embedding/Gather" type="Gather" version="opset8">
|
|
<data batch_dims="0" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>512</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="2" precision="I32" />
|
|
</input>
|
|
<output>
|
|
<port id="3" precision="FP32" names="63,position_embeddings.1">
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="34" name="__module.embeddings/aten::add_/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="61,embeddings.1">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="35" name="__module.embeddings.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="36" name="__module.embeddings.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="37" name="Constant_162781" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 128" offset="4031968" size="512" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="38" name="__module.embeddings.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="39" name="Constant_162782" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 128" offset="4032480" size="512" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="40" name="__module.embeddings.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="68,input.1">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="41" name="__module.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 128" offset="4032992" size="512" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="42" name="__module.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="43" name="__module.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4033504" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="44" name="__module.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4033508" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="45" name="__module.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4033504" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="46" name="__module.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4033508" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="47" name="__module.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="48" name="Constant_28170416" type="Const" version="opset1">
|
|
<data element_type="i8" shape="768, 128" offset="4033512" size="98304" />
|
|
<output>
|
|
<port id="0" precision="I8">
|
|
<dim>768</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="49" name="Convert_28170417" type="Convert" version="opset1">
|
|
<data destination_type="f32" />
|
|
<input>
|
|
<port id="0" precision="I8">
|
|
<dim>768</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="50" name="Constant_28170418" type="Const" version="opset1">
|
|
<data element_type="f32" shape="768, 1" offset="4131816" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="51" name="__module.encoder.embedding_hidden_mapping_in/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="52" name="__module.encoder.embedding_hidden_mapping_in/aten::linear/MatMul" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>128</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="53" name="Constant_162783" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="4134888" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="54" name="__module.encoder.embedding_hidden_mapping_in/aten::linear/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="95,hidden_states.1">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="55" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4137960" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="56" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4137964" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="57" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4137960" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="58" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4137964" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="59" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="60" name="Constant_28170420" type="Const" version="opset1">
|
|
<data element_type="i8" shape="768, 768" offset="4137968" size="589824" />
|
|
<output>
|
|
<port id="0" precision="I8">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="61" name="Convert_28170421" type="Convert" version="opset1">
|
|
<data destination_type="f32" />
|
|
<input>
|
|
<port id="0" precision="I8">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="62" name="Constant_28170422" type="Const" version="opset1">
|
|
<data element_type="f32" shape="768, 1" offset="4727792" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="63" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="64" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="65" name="Constant_162784" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="4730864" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="66" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="110,x.1">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="67" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4733936" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="68" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4733940" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="69" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4733936" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="70" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4733940" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="71" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="72" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="73" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="120,x.3">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="74" name="Constant_155145" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="121">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="75" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="122">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="76" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4137960" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="77" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4137964" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="78" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4137960" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="79" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4137964" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="80" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="81" name="Constant_28170424" type="Const" version="opset1">
|
|
<data element_type="i8" shape="768, 768" offset="4734008" size="589824" />
|
|
<output>
|
|
<port id="0" precision="I8">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="82" name="Convert_28170425" type="Convert" version="opset1">
|
|
<data destination_type="f32" />
|
|
<input>
|
|
<port id="0" precision="I8">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="83" name="Constant_28170426" type="Const" version="opset1">
|
|
<data element_type="f32" shape="768, 1" offset="5323832" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="84" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="85" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="86" name="Constant_162785" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="5326904" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="87" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="113,x.5">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="88" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="5329976" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="89" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="5329980" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="90" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="5329976" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="91" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="5329980" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="92" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="93" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_1" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="94" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_1" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="126,x.7">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="95" name="Constant_155164" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="127">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="96" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_1" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="128,key_layer.1">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="97" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="136,attention_scores.1">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="98" name="Constant_162786" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1, 1" offset="5329984" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="99" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::div/Divide" type="Divide" version="opset1">
|
|
<data auto_broadcast="numpy" m_pythondiv="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="137,attention_scores.3">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="100" name="Constant_162788" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1, 1" offset="5329988" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="101" name="8" type="Const" version="opset1">
|
|
<data element_type="i64" shape="" offset="4031956" size="8" />
|
|
<output>
|
|
<port id="0" precision="I64" names="8" />
|
|
</output>
|
|
</layer>
|
|
<layer id="102" name="aten::unsqueeze/Unsqueeze" type="Unsqueeze" version="opset1">
|
|
<input>
|
|
<port id="0" precision="I64">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="I64" />
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="I64" names="9">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="103" name="10" type="Const" version="opset1">
|
|
<data element_type="i64" shape="" offset="5329992" size="8" />
|
|
<output>
|
|
<port id="0" precision="I64" names="10" />
|
|
</output>
|
|
</layer>
|
|
<layer id="104" name="aten::unsqueeze/Unsqueeze_1" type="Unsqueeze" version="opset1">
|
|
<input>
|
|
<port id="0" precision="I64">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="I64" />
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="I64" names="11,extended_attention_mask">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="105" name="aten::to/Convert" type="Convert" version="opset1">
|
|
<data destination_type="f32" />
|
|
<input>
|
|
<port id="0" precision="I64">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="16">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="106" name="Constant_162787" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1, 1" offset="5329988" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="107" name="aten::rsub/Multiply" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="108" name="aten::rsub/Subtract" type="Subtract" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="19">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="109" name="Constant_162789" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1, 1" offset="5330000" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="110" name="aten::mul/Multiply" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="21,attention_mask">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="111" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::add/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="138,input.3">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="112" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::softmax/Softmax" type="SoftMax" version="opset8">
|
|
<data axis="-1" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="139,input.5">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="113" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4137960" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="114" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4137964" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="115" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4137960" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="116" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="4137964" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="117" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="118" name="Constant_28170428" type="Const" version="opset1">
|
|
<data element_type="i8" shape="768, 768" offset="5330004" size="589824" />
|
|
<output>
|
|
<port id="0" precision="I8">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="119" name="Convert_28170429" type="Convert" version="opset1">
|
|
<data destination_type="f32" />
|
|
<input>
|
|
<port id="0" precision="I8">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="120" name="Constant_28170430" type="Const" version="opset1">
|
|
<data element_type="f32" shape="768, 1" offset="5919828" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="121" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="122" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="123" name="Constant_162790" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="5922900" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="124" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="116,x.9">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="125" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_2" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="126" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_2" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="132,x.11">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="127" name="Constant_155183" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="133">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="128" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_2" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="134">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="129" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_1" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="false" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="141,context_layer.1">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="130" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_1/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="5925972" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="131" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_1/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="5925976" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="132" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_1/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="5925972" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="133" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_1/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="5925976" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="134" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_1/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="135" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::transpose/ScatterElementsUpdate_1" type="Const" version="opset1">
|
|
<data element_type="i32" shape="4" offset="5925980" size="16" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="136" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::transpose/Transpose_1" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="142">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="137" name="Constant_162959" type="Const" version="opset1">
|
|
<data element_type="i64" shape="3" offset="5925996" size="24" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>3</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="138" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::flatten/Reshape" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>3</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="143">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="139" name="Constant_28170432" type="Const" version="opset1">
|
|
<data element_type="i8" shape="768, 768" offset="5926020" size="589824" />
|
|
<output>
|
|
<port id="0" precision="I8">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="140" name="Convert_28170433" type="Convert" version="opset1">
|
|
<data destination_type="f32" />
|
|
<input>
|
|
<port id="0" precision="I8">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="141" name="Constant_28170434" type="Const" version="opset1">
|
|
<data element_type="f32" shape="768, 1" offset="6515844" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="142" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="143" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/MatMul" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="144" name="Constant_162791" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6518916" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="145" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="146,input.7">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="146" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::add/Add_1" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="148">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="147" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="148" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="149" name="Constant_162792" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6521988" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="150" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="151" name="Constant_162793" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6525060" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="152" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="152">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="153" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="6528132" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="154" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="6528136" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="155" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="6528132" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="156" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="6528136" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="157" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="158" name="Constant_28170436" type="Const" version="opset1">
|
|
<data element_type="i8" shape="3072, 768" offset="6528140" size="2359296" />
|
|
<output>
|
|
<port id="0" precision="I8">
|
|
<dim>3072</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="159" name="Convert_28170437" type="Convert" version="opset1">
|
|
<data destination_type="f32" />
|
|
<input>
|
|
<port id="0" precision="I8">
|
|
<dim>3072</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32">
|
|
<dim>3072</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="160" name="Constant_28170438" type="Const" version="opset1">
|
|
<data element_type="f32" shape="3072, 1" offset="8887436" size="12288" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>3072</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="161" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>3072</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>3072</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>3072</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="162" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>3072</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="163" name="Constant_162794" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 3072" offset="8899724" size="12288" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="164" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="155,input.9">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="165" name="Constant_162795" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912012" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="166" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="156">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="167" name="Constant_162796" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912016" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="168" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::pow/Power" type="Power" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="157">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="169" name="Constant_162797" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912020" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="170" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_1" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="158">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="171" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::add/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="159">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="172" name="Constant_162798" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912024" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="173" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_2" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="160">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="174" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::tanh/Tanh" type="Tanh" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="161">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="175" name="Constant_162799" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="5329988" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="176" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::add/Add_1" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="162">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="177" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_3" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="163">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="178" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_3/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="8912028" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="179" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_3/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="8912032" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="180" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_3/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="8912028" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="181" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_3/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="8912032" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="182" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_3/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="183" name="Constant_28170440" type="Const" version="opset1">
|
|
<data element_type="i8" shape="768, 3072" offset="8912036" size="2359296" />
|
|
<output>
|
|
<port id="0" precision="I8">
|
|
<dim>768</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="184" name="Convert_28170441" type="Convert" version="opset1">
|
|
<data destination_type="f32" />
|
|
<input>
|
|
<port id="0" precision="I8">
|
|
<dim>768</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="185" name="Constant_28170442" type="Const" version="opset1">
|
|
<data element_type="f32" shape="768, 1" offset="11271332" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="186" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="187" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/MatMul" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="188" name="Constant_162800" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11274404" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="189" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="166,ffn_output.1">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="190" name="__module.encoder.albert_layer_groups.0.albert_layers.0/aten::add/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="167">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="191" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Multiply" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="192" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/MVN" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="193" name="Constant_162801" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11277476" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="194" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="195" name="Constant_162802" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11280548" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="196" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Add" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="171,hidden_states.3">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="197" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_1/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283620" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="198" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_1/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283624" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="199" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_1/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283620" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="200" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_1/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283624" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="201" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_1/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="202" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_1" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="203" name="Constant_162803" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="4730864" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="204" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_1" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="186,x.13">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="205" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_1/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283628" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="206" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_1/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283632" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="207" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_1/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283628" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="208" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_1/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283632" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="209" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_1/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="210" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_3" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="211" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_3" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="196,x.15">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="212" name="Constant_155563" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="197">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="213" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_3" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="198">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="214" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_1/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283620" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="215" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_1/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283624" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="216" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_1/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283620" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="217" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_1/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283624" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="218" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_1/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="219" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_1" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="220" name="Constant_162804" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11283636" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="221" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_1" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="189,x.17">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="222" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_1/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286708" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="223" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_1/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286712" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="224" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_1/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286708" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="225" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_1/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286712" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="226" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_1/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="227" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_4" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="228" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_4" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="202,x.19">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="229" name="Constant_155582" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="203">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="230" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_4" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="204,key_layer.3">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="231" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_2" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="212,attention_scores.5">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="232" name="Constant_162805" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1, 1" offset="5329984" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="233" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::div/Divide_1" type="Divide" version="opset1">
|
|
<data auto_broadcast="numpy" m_pythondiv="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="213,attention_scores.7">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="234" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::add/Add_2" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="214,input.11">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="235" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::softmax/Softmax_1" type="SoftMax" version="opset8">
|
|
<data axis="-1" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="215,input.13">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="236" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_1/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283620" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="237" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_1/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283624" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="238" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_1/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283620" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="239" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_1/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11283624" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="240" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_1/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="241" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_1" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="242" name="Constant_162806" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="5922900" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="243" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/Add_1" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="192,x.21">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="244" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_5" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="245" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_5" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="208,x.23">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="246" name="Constant_155601" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="209">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="247" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_5" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="210">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="248" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_3" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="false" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="217,context_layer.3">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="249" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_3/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286716" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="250" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_3/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286720" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="251" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_3/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286716" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="252" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_3/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286720" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="253" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_3/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="254" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::transpose/ScatterElementsUpdate_3" type="Const" version="opset1">
|
|
<data element_type="i32" shape="4" offset="5925980" size="16" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="255" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::transpose/Transpose_3" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="218">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="256" name="Constant_162960" type="Const" version="opset1">
|
|
<data element_type="i64" shape="3" offset="5925996" size="24" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>3</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="257" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::flatten/Reshape_1" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>3</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="219">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="258" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/MatMul_1" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="259" name="Constant_162807" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6518916" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="260" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/Add_1" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="222,input.15">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="261" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::add/Add_3" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="224">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="262" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Multiply_2" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="263" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/MVN_1" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="264" name="Constant_162808" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6521988" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="265" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Multiply_3" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="266" name="Constant_162809" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6525060" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="267" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Add_1" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="228">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="268" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_1/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286724" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="269" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_1/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286728" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="270" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_1/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286724" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="271" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_1/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286728" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="272" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_1/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="273" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_1" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>3072</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="274" name="Constant_162810" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 3072" offset="8899724" size="12288" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="275" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/Add_1" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="231,input.17">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="276" name="Constant_162811" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912012" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="277" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_4" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="232">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="278" name="Constant_162812" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912016" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="279" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::pow/Power_1" type="Power" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="233">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="280" name="Constant_162813" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912020" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="281" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_5" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="234">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="282" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::add/Add_2" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="235">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="283" name="Constant_162814" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912024" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="284" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_6" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="236">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="285" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::tanh/Tanh_1" type="Tanh" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="237">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="286" name="Constant_162815" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="5329988" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="287" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::add/Add_3" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="238">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="288" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_7" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="239">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="289" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_7/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="8912028" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="290" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_7/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286732" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="291" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_7/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="8912028" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="292" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_7/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286732" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="293" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_7/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="294" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/MatMul_1" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="295" name="Constant_162816" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11274404" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="296" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/Add_1" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="242,ffn_output.3">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="297" name="__module.encoder.albert_layer_groups.0.albert_layers.0/aten::add/Add_1" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="243">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="298" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Multiply_2" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="299" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/MVN_1" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="300" name="Constant_162817" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11277476" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="301" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Multiply_3" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="302" name="Constant_162818" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11280548" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="303" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Add_1" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="247,hidden_states.5">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="304" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_2/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286736" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="305" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_2/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286740" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="306" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_2/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286736" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="307" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_2/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286740" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="308" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_2/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="309" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_2" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="310" name="Constant_162819" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="4730864" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="311" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_2" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="262,x.25">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="312" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_2/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286744" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="313" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_2/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286748" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="314" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_2/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286744" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="315" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_2/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286748" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="316" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_2/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="317" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_6" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="318" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_6" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="272,x.27">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="319" name="Constant_155961" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="273">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="320" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_6" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="274">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="321" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_2/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286736" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="322" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_2/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286740" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="323" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_2/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286736" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="324" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_2/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286740" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="325" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_2/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="326" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_2" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="327" name="Constant_162820" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="5326904" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="328" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_2" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="265,x.29">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="329" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_2/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286752" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="330" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_2/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286756" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="331" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_2/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286752" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="332" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_2/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286756" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="333" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_2/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="334" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_7" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="335" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_7" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="278,x.31">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="336" name="Constant_155980" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="279">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="337" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_7" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="280,key_layer.5">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="338" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_4" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="288,attention_scores.9">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="339" name="Constant_162821" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1, 1" offset="5329984" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="340" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::div/Divide_2" type="Divide" version="opset1">
|
|
<data auto_broadcast="numpy" m_pythondiv="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="289,attention_scores.11">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="341" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::add/Add_4" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="290,input.19">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="342" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::softmax/Softmax_2" type="SoftMax" version="opset8">
|
|
<data axis="-1" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="291,input.21">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="343" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_2/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286736" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="344" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_2/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286740" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="345" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_2/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286736" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="346" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_2/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286740" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="347" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_2/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="348" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_2" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="349" name="Constant_162822" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="5922900" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="350" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/Add_2" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="268,x.33">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="351" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_8" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="352" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_8" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="284,x.35">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="353" name="Constant_155999" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="285">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="354" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_8" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="286">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="355" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_5" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="false" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="293,context_layer.5">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="356" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_5/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286760" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="357" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_5/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286764" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="358" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_5/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286760" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="359" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_5/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286764" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="360" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_5/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="361" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::transpose/ScatterElementsUpdate_5" type="Const" version="opset1">
|
|
<data element_type="i32" shape="4" offset="5925980" size="16" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="362" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::transpose/Transpose_5" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="294">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="363" name="Constant_162961" type="Const" version="opset1">
|
|
<data element_type="i64" shape="3" offset="5925996" size="24" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>3</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="364" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::flatten/Reshape_2" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>3</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="295">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="365" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/MatMul_2" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="366" name="Constant_162823" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6518916" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="367" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/Add_2" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="298,input.23">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="368" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::add/Add_5" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="300">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="369" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Multiply_4" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="370" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/MVN_2" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="371" name="Constant_162824" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6521988" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="372" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Multiply_5" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="373" name="Constant_162825" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6525060" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="374" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Add_2" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="304">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="375" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_2/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286768" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="376" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_2/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286772" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="377" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_2/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286768" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="378" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_2/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286772" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="379" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_2/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="380" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_2" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>3072</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="381" name="Constant_162826" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 3072" offset="8899724" size="12288" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="382" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/Add_2" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="307,input.25">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="383" name="Constant_162827" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912012" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="384" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_8" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="308">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="385" name="Constant_162828" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912016" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="386" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::pow/Power_2" type="Power" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="309">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="387" name="Constant_162829" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912020" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="388" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_9" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="310">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="389" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::add/Add_4" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="311">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="390" name="Constant_162830" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912024" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="391" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_10" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="312">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="392" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::tanh/Tanh_2" type="Tanh" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="313">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="393" name="Constant_162831" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="5329988" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="394" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::add/Add_5" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="314">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="395" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_11" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="315">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="396" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_11/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286776" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="397" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_11/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286780" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="398" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_11/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286776" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="399" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_11/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286780" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="400" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_11/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="401" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/MatMul_2" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="402" name="Constant_162832" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11274404" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="403" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/Add_2" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="318,ffn_output.5">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="404" name="__module.encoder.albert_layer_groups.0.albert_layers.0/aten::add/Add_2" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="319">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="405" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Multiply_4" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="406" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/MVN_2" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="407" name="Constant_162833" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11277476" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="408" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Multiply_5" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="409" name="Constant_162834" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11280548" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="410" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Add_2" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="323,hidden_states.7">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="411" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_3/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286784" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="412" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_3/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286788" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="413" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_3/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286784" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="414" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_3/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286788" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="415" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_3/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="416" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_3" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="417" name="Constant_162835" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="4730864" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="418" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_3" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="338,x.37">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="419" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_3/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286792" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="420" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_3/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286796" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="421" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_3/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286792" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="422" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_3/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286796" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="423" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_3/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="424" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_9" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="425" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_9" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="348,x.39">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="426" name="Constant_156359" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="349">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="427" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_9" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="350">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="428" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_3/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286784" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="429" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_3/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286788" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="430" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_3/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286784" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="431" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_3/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286788" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="432" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_3/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="433" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_3" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="434" name="Constant_162836" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11286800" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="435" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_3" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="341,x.41">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="436" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_3/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289872" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="437" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_3/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289876" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="438" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_3/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289872" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="439" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_3/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289876" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="440" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_3/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="441" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_10" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="442" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_10" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="354,x.43">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="443" name="Constant_156378" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="355">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="444" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_10" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="356,key_layer.7">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="445" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_6" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="364,attention_scores.13">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="446" name="Constant_162837" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1, 1" offset="5329984" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="447" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::div/Divide_3" type="Divide" version="opset1">
|
|
<data auto_broadcast="numpy" m_pythondiv="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="365,attention_scores.15">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="448" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::add/Add_6" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="366,input.27">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="449" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::softmax/Softmax_3" type="SoftMax" version="opset8">
|
|
<data axis="-1" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="367,input.29">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="450" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_3/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286784" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="451" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_3/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286788" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="452" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_3/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286784" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="453" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_3/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11286788" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="454" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_3/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="455" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_3" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="456" name="Constant_162838" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="5922900" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="457" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/Add_3" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="344,x.45">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="458" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_11" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="459" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_11" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="360,x.47">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="460" name="Constant_156397" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="361">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="461" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_11" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="362">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="462" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_7" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="false" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="369,context_layer.7">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="463" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_7/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289880" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="464" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_7/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289884" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="465" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_7/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289880" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="466" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_7/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289884" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="467" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_7/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="468" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::transpose/ScatterElementsUpdate_7" type="Const" version="opset1">
|
|
<data element_type="i32" shape="4" offset="5925980" size="16" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="469" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::transpose/Transpose_7" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="370">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="470" name="Constant_162962" type="Const" version="opset1">
|
|
<data element_type="i64" shape="3" offset="5925996" size="24" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>3</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="471" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::flatten/Reshape_3" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>3</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="371">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="472" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/MatMul_3" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="473" name="Constant_162839" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6518916" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="474" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/Add_3" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="374,input.31">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="475" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::add/Add_7" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="376">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="476" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Multiply_6" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="477" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/MVN_3" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="478" name="Constant_162840" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6521988" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="479" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Multiply_7" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="480" name="Constant_162841" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6525060" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="481" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Add_3" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="380">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="482" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_3/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289888" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="483" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_3/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289892" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="484" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_3/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289888" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="485" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_3/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289892" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="486" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_3/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="487" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_3" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>3072</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="488" name="Constant_162842" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 3072" offset="8899724" size="12288" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="489" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/Add_3" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="383,input.33">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="490" name="Constant_162843" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912012" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="491" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_12" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="384">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="492" name="Constant_162844" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912016" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="493" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::pow/Power_3" type="Power" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="385">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="494" name="Constant_162845" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912020" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="495" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_13" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="386">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="496" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::add/Add_6" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="387">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="497" name="Constant_162846" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912024" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="498" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_14" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="388">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="499" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::tanh/Tanh_3" type="Tanh" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="389">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="500" name="Constant_162847" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="5329988" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="501" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::add/Add_7" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="390">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="502" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_15" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="391">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="503" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_15/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289896" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="504" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_15/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289900" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="505" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_15/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289896" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="506" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_15/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289900" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="507" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_15/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="508" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/MatMul_3" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="509" name="Constant_162848" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11274404" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="510" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/Add_3" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="394,ffn_output.7">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="511" name="__module.encoder.albert_layer_groups.0.albert_layers.0/aten::add/Add_3" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="395">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="512" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Multiply_6" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="513" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/MVN_3" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="514" name="Constant_162849" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11277476" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="515" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Multiply_7" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="516" name="Constant_162850" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11280548" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="517" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Add_3" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="399,hidden_states.9">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="518" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_4/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289904" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="519" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_4/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289908" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="520" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_4/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289904" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="521" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_4/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289908" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="522" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_4/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="523" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_4" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="524" name="Constant_162851" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="4730864" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="525" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_4" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="414,x.49">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="526" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_4/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289912" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="527" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_4/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289916" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="528" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_4/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289912" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="529" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_4/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289916" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="530" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_4/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="531" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_12" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="532" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_12" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="424,x.51">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="533" name="Constant_156757" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="425">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="534" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_12" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="426">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="535" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_4/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289904" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="536" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_4/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289908" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="537" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_4/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289904" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="538" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_4/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289908" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="539" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_4/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="540" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_4" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="541" name="Constant_162852" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="5326904" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="542" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_4" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="417,x.53">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="543" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_4/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289920" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="544" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_4/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289924" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="545" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_4/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289920" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="546" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_4/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289924" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="547" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_4/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="548" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_13" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="549" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_13" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="430,x.55">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="550" name="Constant_156776" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="431">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="551" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_13" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="432,key_layer.9">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="552" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_8" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="440,attention_scores.17">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="553" name="Constant_162853" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1, 1" offset="5329984" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="554" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::div/Divide_4" type="Divide" version="opset1">
|
|
<data auto_broadcast="numpy" m_pythondiv="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="441,attention_scores.19">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="555" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::add/Add_8" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="442,input.35">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="556" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::softmax/Softmax_4" type="SoftMax" version="opset8">
|
|
<data axis="-1" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="443,input.37">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="557" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_4/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289904" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="558" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_4/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289908" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="559" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_4/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289904" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="560" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_4/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289908" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="561" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_4/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="562" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_4" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="563" name="Constant_162854" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="5922900" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="564" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/Add_4" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="420,x.57">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="565" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_14" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="566" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_14" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="436,x.59">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="567" name="Constant_156795" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="437">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="568" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_14" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="438">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="569" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_9" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="false" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="445,context_layer.9">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="570" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_9/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289928" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="571" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_9/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289932" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="572" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_9/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289928" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="573" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_9/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289932" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="574" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_9/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="575" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::transpose/ScatterElementsUpdate_9" type="Const" version="opset1">
|
|
<data element_type="i32" shape="4" offset="5925980" size="16" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="576" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::transpose/Transpose_9" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="446">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="577" name="Constant_162963" type="Const" version="opset1">
|
|
<data element_type="i64" shape="3" offset="5925996" size="24" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>3</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="578" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::flatten/Reshape_4" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>3</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="447">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="579" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/MatMul_4" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="580" name="Constant_162855" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6518916" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="581" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/Add_4" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="450,input.39">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="582" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::add/Add_9" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="452">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="583" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Multiply_8" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="584" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/MVN_4" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="585" name="Constant_162856" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6521988" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="586" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Multiply_9" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="587" name="Constant_162857" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6525060" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="588" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Add_4" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="456">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="589" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_4/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289936" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="590" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_4/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289940" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="591" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_4/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289936" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="592" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_4/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289940" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="593" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_4/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="594" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_4" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>3072</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="595" name="Constant_162858" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 3072" offset="8899724" size="12288" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="596" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/Add_4" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="459,input.41">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="597" name="Constant_162859" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912012" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="598" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_16" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="460">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="599" name="Constant_162860" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912016" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="600" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::pow/Power_4" type="Power" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="461">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="601" name="Constant_162861" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912020" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="602" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_17" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="462">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="603" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::add/Add_8" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="463">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="604" name="Constant_162862" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912024" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="605" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_18" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="464">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="606" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::tanh/Tanh_4" type="Tanh" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="465">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="607" name="Constant_162863" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="5329988" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="608" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::add/Add_9" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="466">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="609" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_19" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="467">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="610" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_19/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289944" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="611" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_19/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289948" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="612" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_19/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289944" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="613" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_19/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289948" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="614" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_19/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="615" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/MatMul_4" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="616" name="Constant_162864" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11274404" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="617" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/Add_4" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="470,ffn_output.9">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="618" name="__module.encoder.albert_layer_groups.0.albert_layers.0/aten::add/Add_4" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="471">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="619" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Multiply_8" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="620" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/MVN_4" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="621" name="Constant_162865" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11277476" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="622" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Multiply_9" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="623" name="Constant_162866" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11280548" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="624" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Add_4" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="475,hidden_states">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="625" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_5/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289952" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="626" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_5/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289956" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="627" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_5/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289952" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="628" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_5/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289956" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="629" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_5/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="630" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/MatMul_5" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="631" name="Constant_162867" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="4730864" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="632" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_5" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="490,x.61">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="633" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_5/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289960" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="634" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_5/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289964" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="635" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_5/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289960" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="636" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_5/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289964" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="637" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.query/aten::linear/Add_5/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="638" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_15" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="639" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_15" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="500,x.63">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="640" name="Constant_157155" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="501">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="641" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_15" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="502">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="642" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_5/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289952" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="643" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_5/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289956" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="644" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_5/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289952" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="645" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_5/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289956" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="646" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_5/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="647" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/MatMul_5" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="648" name="Constant_162868" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="5326904" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="649" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_5" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="493,x.65">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="650" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_5/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289968" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="651" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_5/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289972" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="652" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_5/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289968" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="653" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_5/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289972" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="654" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.key/aten::linear/Add_5/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="655" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_16" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="656" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_16" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="506,x.67">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="657" name="Constant_157174" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="507">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="658" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_16" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="508,key_layer">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="659" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_10" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="516,attention_scores.21">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="660" name="Constant_162869" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1, 1" offset="5329984" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="661" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::div/Divide_5" type="Divide" version="opset1">
|
|
<data auto_broadcast="numpy" m_pythondiv="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="517,attention_scores">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="662" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::add/Add_10" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="518,input.43">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="663" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::softmax/Softmax_5" type="SoftMax" version="opset8">
|
|
<data axis="-1" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="519,input.45">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="664" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_5/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289952" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="665" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_5/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289956" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="666" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_5/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289952" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="667" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_5/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289956" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="668" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_5/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="669" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/MatMul_5" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="670" name="Constant_162870" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="5922900" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="671" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.value/aten::linear/Add_5" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="496,x.69">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="672" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/prim::ListConstruct/Concat_17" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733944" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="673" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::view/Reshape_17" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="512,x">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="674" name="Constant_157193" type="Const" version="opset1">
|
|
<data element_type="i64" shape="4" offset="4733976" size="32" />
|
|
<output>
|
|
<port id="0" precision="I64" names="513">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="675" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::permute/Transpose_17" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="514">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="676" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_11" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="false" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="521,context_layer">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="677" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_11/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289976" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="678" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_11/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289980" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="679" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_11/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289976" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="680" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_11/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289980" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="681" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::matmul/MatMul_11/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="682" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::transpose/ScatterElementsUpdate_11" type="Const" version="opset1">
|
|
<data element_type="i32" shape="4" offset="5925980" size="16" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>4</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="683" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::transpose/Transpose_11" type="Transpose" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>-1</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>4</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="522">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="684" name="Constant_162964" type="Const" version="opset1">
|
|
<data element_type="i64" shape="3" offset="5925996" size="24" />
|
|
<output>
|
|
<port id="0" precision="I64">
|
|
<dim>3</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="685" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::flatten/Reshape_5" type="Reshape" version="opset1">
|
|
<data special_zero="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>12</dim>
|
|
<dim>64</dim>
|
|
</port>
|
|
<port id="1" precision="I64">
|
|
<dim>3</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="523">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="686" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/MatMul_5" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="687" name="Constant_162871" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6518916" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="688" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.dense/aten::linear/Add_5" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="526,input.47">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="689" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention/aten::add/Add_11" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="528">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="690" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Multiply_10" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="691" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/MVN_5" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="692" name="Constant_162872" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6521988" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="693" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Multiply_11" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="694" name="Constant_162873" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="6525060" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="695" name="__module.encoder.albert_layer_groups.0.albert_layers.0.attention.LayerNorm/aten::layer_norm/Add_5" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="532">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="696" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_5/fq_input_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289984" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="697" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_5/fq_input_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289988" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="698" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_5/fq_input_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289984" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="699" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_5/fq_input_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289988" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="700" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_5/fq_input_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="701" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/MatMul_5" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>3072</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="702" name="Constant_162874" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 3072" offset="8899724" size="12288" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="703" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn/aten::linear/Add_5" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="535,input">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="704" name="Constant_162875" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912012" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="705" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_20" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="536">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="706" name="Constant_162876" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912016" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="707" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::pow/Power_5" type="Power" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="537">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="708" name="Constant_162877" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912020" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="709" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_21" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="538">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="710" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::add/Add_10" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="539">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="711" name="Constant_162878" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="8912024" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="712" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_22" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="540">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="713" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::tanh/Tanh_5" type="Tanh" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="1" precision="FP32" names="541">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="714" name="Constant_162879" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 1" offset="5329988" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="715" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::add/Add_11" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="542">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="716" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_23" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="543">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="717" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_23/fq_output_0/input_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="8912028" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="718" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_23/fq_output_0/input_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289992" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="719" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_23/fq_output_0/output_low" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="8912028" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="720" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_23/fq_output_0/output_high" type="Const" version="opset1">
|
|
<data element_type="f32" shape="" offset="11289992" size="4" />
|
|
<output>
|
|
<port id="0" precision="FP32" />
|
|
</output>
|
|
</layer>
|
|
<layer id="721" name="__module.encoder.albert_layer_groups.0.albert_layers.0.activation/aten::mul/Multiply_23/fq_output_0" type="FakeQuantize" version="opset1">
|
|
<data levels="256" auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32" />
|
|
<port id="2" precision="FP32" />
|
|
<port id="3" precision="FP32" />
|
|
<port id="4" precision="FP32" />
|
|
</input>
|
|
<output>
|
|
<port id="5" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="722" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/MatMul_5" type="MatMul" version="opset1">
|
|
<data transpose_a="false" transpose_b="true" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>768</dim>
|
|
<dim>3072</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="723" name="Constant_162880" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11274404" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="724" name="__module.encoder.albert_layer_groups.0.albert_layers.0.ffn_output/aten::linear/Add_5" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="546,ffn_output.11">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="725" name="__module.encoder.albert_layer_groups.0.albert_layers.0/aten::add/Add_5" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="547">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="726" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Multiply_10" type="Const" version="opset1">
|
|
<data element_type="i32" shape="1" offset="4031964" size="4" />
|
|
<output>
|
|
<port id="0" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="727" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/MVN_5" type="MVN" version="opset6">
|
|
<data eps="9.999999960041972e-13" normalize_variance="true" eps_mode="INSIDE_SQRT" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="I32">
|
|
<dim>1</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="728" name="Constant_162881" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11277476" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="729" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Multiply_11" type="Multiply" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="730" name="Constant_162882" type="Const" version="opset1">
|
|
<data element_type="f32" shape="1, 1, 768" offset="11280548" size="3072" />
|
|
<output>
|
|
<port id="0" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="731" name="__module.encoder.albert_layer_groups.0.albert_layers.0.full_layer_layer_norm/aten::layer_norm/Add_5" type="Add" version="opset1">
|
|
<data auto_broadcast="numpy" />
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
<port id="1" precision="FP32">
|
|
<dim>1</dim>
|
|
<dim>1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
<output>
|
|
<port id="2" precision="FP32" names="last_hidden_state">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</output>
|
|
</layer>
|
|
<layer id="732" name="Result_159419" type="Result" version="opset1">
|
|
<input>
|
|
<port id="0" precision="FP32">
|
|
<dim>-1</dim>
|
|
<dim>-1</dim>
|
|
<dim>768</dim>
|
|
</port>
|
|
</input>
|
|
</layer>
|
|
</layers>
|
|
<edges>
|
|
<edge from-layer="0" from-port="0" to-layer="14" to-port="0" />
|
|
<edge from-layer="1" from-port="0" to-layer="102" to-port="0" />
|
|
<edge from-layer="2" from-port="0" to-layer="7" to-port="0" />
|
|
<edge from-layer="2" from-port="0" to-layer="24" 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="6" to-port="0" />
|
|
<edge from-layer="5" from-port="0" to-layer="6" to-port="1" />
|
|
<edge from-layer="6" from-port="2" to-layer="9" to-port="0" />
|
|
<edge from-layer="7" from-port="1" to-layer="9" to-port="1" />
|
|
<edge from-layer="8" from-port="0" to-layer="9" to-port="2" />
|
|
<edge from-layer="9" from-port="3" to-layer="17" to-port="0" />
|
|
<edge from-layer="10" from-port="0" to-layer="11" to-port="0" />
|
|
<edge from-layer="11" from-port="1" to-layer="13" to-port="0" />
|
|
<edge from-layer="12" from-port="0" to-layer="13" to-port="1" />
|
|
<edge from-layer="13" from-port="2" to-layer="16" to-port="0" />
|
|
<edge from-layer="14" from-port="1" to-layer="16" to-port="1" />
|
|
<edge from-layer="15" from-port="0" to-layer="16" to-port="2" />
|
|
<edge from-layer="16" from-port="3" to-layer="17" to-port="1" />
|
|
<edge from-layer="17" from-port="2" to-layer="34" to-port="0" />
|
|
<edge from-layer="18" from-port="0" to-layer="19" to-port="0" />
|
|
<edge from-layer="19" from-port="1" to-layer="21" to-port="0" />
|
|
<edge from-layer="20" from-port="0" to-layer="21" to-port="1" />
|
|
<edge from-layer="21" from-port="2" to-layer="33" to-port="0" />
|
|
<edge from-layer="22" from-port="0" to-layer="30" to-port="0" />
|
|
<edge from-layer="23" from-port="0" to-layer="30" to-port="1" />
|
|
<edge from-layer="24" from-port="1" to-layer="27" to-port="0" />
|
|
<edge from-layer="25" from-port="0" to-layer="27" to-port="1" />
|
|
<edge from-layer="26" from-port="0" to-layer="27" to-port="2" />
|
|
<edge from-layer="27" from-port="3" to-layer="30" to-port="2" />
|
|
<edge from-layer="28" from-port="0" to-layer="30" to-port="3" />
|
|
<edge from-layer="29" from-port="0" to-layer="30" to-port="4" />
|
|
<edge from-layer="30" from-port="5" to-layer="31" to-port="0" />
|
|
<edge from-layer="31" from-port="1" to-layer="33" to-port="1" />
|
|
<edge from-layer="32" from-port="0" to-layer="33" to-port="2" />
|
|
<edge from-layer="33" from-port="3" to-layer="34" to-port="1" />
|
|
<edge from-layer="34" from-port="2" to-layer="36" to-port="0" />
|
|
<edge from-layer="35" from-port="0" to-layer="36" to-port="1" />
|
|
<edge from-layer="36" from-port="2" to-layer="38" to-port="0" />
|
|
<edge from-layer="37" from-port="0" to-layer="38" to-port="1" />
|
|
<edge from-layer="38" from-port="2" to-layer="40" to-port="0" />
|
|
<edge from-layer="39" from-port="0" to-layer="40" to-port="1" />
|
|
<edge from-layer="40" from-port="2" to-layer="42" to-port="0" />
|
|
<edge from-layer="41" from-port="0" to-layer="42" to-port="1" />
|
|
<edge from-layer="42" from-port="2" to-layer="47" to-port="0" />
|
|
<edge from-layer="43" from-port="0" to-layer="47" to-port="1" />
|
|
<edge from-layer="44" from-port="0" to-layer="47" to-port="2" />
|
|
<edge from-layer="45" from-port="0" to-layer="47" to-port="3" />
|
|
<edge from-layer="46" from-port="0" to-layer="47" to-port="4" />
|
|
<edge from-layer="47" from-port="5" to-layer="52" to-port="0" />
|
|
<edge from-layer="48" from-port="0" to-layer="49" to-port="0" />
|
|
<edge from-layer="49" from-port="1" to-layer="51" to-port="0" />
|
|
<edge from-layer="50" from-port="0" to-layer="51" to-port="1" />
|
|
<edge from-layer="51" from-port="2" to-layer="52" to-port="1" />
|
|
<edge from-layer="52" from-port="2" to-layer="54" to-port="0" />
|
|
<edge from-layer="53" from-port="0" to-layer="54" to-port="1" />
|
|
<edge from-layer="54" from-port="2" to-layer="59" to-port="0" />
|
|
<edge from-layer="54" from-port="2" to-layer="117" to-port="0" />
|
|
<edge from-layer="54" from-port="2" to-layer="80" to-port="0" />
|
|
<edge from-layer="54" from-port="2" to-layer="146" to-port="0" />
|
|
<edge from-layer="55" from-port="0" to-layer="59" to-port="1" />
|
|
<edge from-layer="56" from-port="0" to-layer="59" to-port="2" />
|
|
<edge from-layer="57" from-port="0" to-layer="59" to-port="3" />
|
|
<edge from-layer="58" from-port="0" to-layer="59" to-port="4" />
|
|
<edge from-layer="59" from-port="5" to-layer="64" to-port="0" />
|
|
<edge from-layer="60" from-port="0" to-layer="61" to-port="0" />
|
|
<edge from-layer="61" from-port="1" to-layer="63" to-port="0" />
|
|
<edge from-layer="62" from-port="0" to-layer="63" to-port="1" />
|
|
<edge from-layer="63" from-port="2" to-layer="630" to-port="1" />
|
|
<edge from-layer="63" from-port="2" to-layer="64" to-port="1" />
|
|
<edge from-layer="63" from-port="2" to-layer="202" to-port="1" />
|
|
<edge from-layer="63" from-port="2" to-layer="309" to-port="1" />
|
|
<edge from-layer="63" from-port="2" to-layer="416" to-port="1" />
|
|
<edge from-layer="63" from-port="2" to-layer="523" to-port="1" />
|
|
<edge from-layer="64" from-port="2" to-layer="66" to-port="0" />
|
|
<edge from-layer="65" from-port="0" to-layer="66" to-port="1" />
|
|
<edge from-layer="66" from-port="2" to-layer="71" to-port="0" />
|
|
<edge from-layer="67" from-port="0" to-layer="71" to-port="1" />
|
|
<edge from-layer="68" from-port="0" to-layer="71" to-port="2" />
|
|
<edge from-layer="69" from-port="0" to-layer="71" to-port="3" />
|
|
<edge from-layer="70" from-port="0" to-layer="71" to-port="4" />
|
|
<edge from-layer="71" from-port="5" to-layer="73" to-port="0" />
|
|
<edge from-layer="72" from-port="0" to-layer="73" to-port="1" />
|
|
<edge from-layer="73" from-port="2" to-layer="75" to-port="0" />
|
|
<edge from-layer="74" from-port="0" to-layer="75" to-port="1" />
|
|
<edge from-layer="75" from-port="2" to-layer="97" to-port="0" />
|
|
<edge from-layer="76" from-port="0" to-layer="80" to-port="1" />
|
|
<edge from-layer="77" from-port="0" to-layer="80" to-port="2" />
|
|
<edge from-layer="78" from-port="0" to-layer="80" to-port="3" />
|
|
<edge from-layer="79" from-port="0" to-layer="80" to-port="4" />
|
|
<edge from-layer="80" from-port="5" to-layer="85" to-port="0" />
|
|
<edge from-layer="81" from-port="0" to-layer="82" to-port="0" />
|
|
<edge from-layer="82" from-port="1" 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="84" from-port="2" to-layer="326" to-port="1" />
|
|
<edge from-layer="84" from-port="2" to-layer="540" to-port="1" />
|
|
<edge from-layer="84" from-port="2" to-layer="433" to-port="1" />
|
|
<edge from-layer="84" from-port="2" to-layer="219" to-port="1" />
|
|
<edge from-layer="84" from-port="2" to-layer="647" to-port="1" />
|
|
<edge from-layer="85" from-port="2" to-layer="87" to-port="0" />
|
|
<edge from-layer="86" from-port="0" to-layer="87" to-port="1" />
|
|
<edge from-layer="87" from-port="2" to-layer="92" to-port="0" />
|
|
<edge from-layer="88" from-port="0" to-layer="92" to-port="1" />
|
|
<edge from-layer="89" from-port="0" to-layer="92" to-port="2" />
|
|
<edge from-layer="90" from-port="0" to-layer="92" to-port="3" />
|
|
<edge from-layer="91" from-port="0" to-layer="92" to-port="4" />
|
|
<edge from-layer="92" from-port="5" to-layer="94" to-port="0" />
|
|
<edge from-layer="93" from-port="0" to-layer="94" to-port="1" />
|
|
<edge from-layer="94" from-port="2" to-layer="96" to-port="0" />
|
|
<edge from-layer="95" from-port="0" to-layer="96" to-port="1" />
|
|
<edge from-layer="96" from-port="2" to-layer="97" to-port="1" />
|
|
<edge from-layer="97" from-port="2" to-layer="99" to-port="0" />
|
|
<edge from-layer="98" from-port="0" to-layer="99" to-port="1" />
|
|
<edge from-layer="99" from-port="2" to-layer="111" to-port="0" />
|
|
<edge from-layer="100" from-port="0" to-layer="108" to-port="0" />
|
|
<edge from-layer="101" from-port="0" 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="0" />
|
|
<edge from-layer="105" from-port="1" 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="108" to-port="1" />
|
|
<edge from-layer="108" from-port="2" to-layer="110" to-port="0" />
|
|
<edge from-layer="109" from-port="0" to-layer="110" to-port="1" />
|
|
<edge from-layer="110" from-port="2" to-layer="448" to-port="1" />
|
|
<edge from-layer="110" from-port="2" to-layer="341" to-port="1" />
|
|
<edge from-layer="110" from-port="2" to-layer="234" to-port="1" />
|
|
<edge from-layer="110" from-port="2" to-layer="111" to-port="1" />
|
|
<edge from-layer="110" from-port="2" to-layer="555" to-port="1" />
|
|
<edge from-layer="110" from-port="2" to-layer="662" to-port="1" />
|
|
<edge from-layer="111" from-port="2" to-layer="112" to-port="0" />
|
|
<edge from-layer="112" from-port="1" to-layer="129" to-port="0" />
|
|
<edge from-layer="113" from-port="0" to-layer="117" to-port="1" />
|
|
<edge from-layer="114" from-port="0" to-layer="117" to-port="2" />
|
|
<edge from-layer="115" from-port="0" to-layer="117" to-port="3" />
|
|
<edge from-layer="116" from-port="0" to-layer="117" to-port="4" />
|
|
<edge from-layer="117" from-port="5" to-layer="122" 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="121" to-port="0" />
|
|
<edge from-layer="120" from-port="0" to-layer="121" to-port="1" />
|
|
<edge from-layer="121" from-port="2" to-layer="348" to-port="1" />
|
|
<edge from-layer="121" from-port="2" to-layer="455" to-port="1" />
|
|
<edge from-layer="121" from-port="2" to-layer="122" to-port="1" />
|
|
<edge from-layer="121" from-port="2" to-layer="562" to-port="1" />
|
|
<edge from-layer="121" from-port="2" to-layer="241" to-port="1" />
|
|
<edge from-layer="121" from-port="2" to-layer="669" 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="126" to-port="0" />
|
|
<edge from-layer="125" from-port="0" to-layer="126" to-port="1" />
|
|
<edge from-layer="126" from-port="2" to-layer="128" to-port="0" />
|
|
<edge from-layer="127" from-port="0" to-layer="128" to-port="1" />
|
|
<edge from-layer="128" from-port="2" to-layer="129" to-port="1" />
|
|
<edge from-layer="129" from-port="2" to-layer="134" to-port="0" />
|
|
<edge from-layer="130" from-port="0" to-layer="134" to-port="1" />
|
|
<edge from-layer="131" from-port="0" to-layer="134" to-port="2" />
|
|
<edge from-layer="132" from-port="0" to-layer="134" to-port="3" />
|
|
<edge from-layer="133" from-port="0" to-layer="134" to-port="4" />
|
|
<edge from-layer="134" from-port="5" to-layer="136" to-port="0" />
|
|
<edge from-layer="135" from-port="0" to-layer="136" to-port="1" />
|
|
<edge from-layer="136" from-port="2" to-layer="138" to-port="0" />
|
|
<edge from-layer="137" from-port="0" to-layer="138" to-port="1" />
|
|
<edge from-layer="138" from-port="2" to-layer="143" to-port="0" />
|
|
<edge from-layer="139" from-port="0" to-layer="140" to-port="0" />
|
|
<edge from-layer="140" from-port="1" to-layer="142" to-port="0" />
|
|
<edge from-layer="141" from-port="0" to-layer="142" to-port="1" />
|
|
<edge from-layer="142" from-port="2" to-layer="365" to-port="1" />
|
|
<edge from-layer="142" from-port="2" to-layer="258" to-port="1" />
|
|
<edge from-layer="142" from-port="2" to-layer="686" to-port="1" />
|
|
<edge from-layer="142" from-port="2" to-layer="143" to-port="1" />
|
|
<edge from-layer="142" from-port="2" to-layer="579" to-port="1" />
|
|
<edge from-layer="142" from-port="2" to-layer="472" to-port="1" />
|
|
<edge from-layer="143" from-port="2" to-layer="145" to-port="0" />
|
|
<edge from-layer="144" from-port="0" to-layer="145" to-port="1" />
|
|
<edge from-layer="145" from-port="2" to-layer="146" to-port="1" />
|
|
<edge from-layer="146" from-port="2" to-layer="148" to-port="0" />
|
|
<edge from-layer="147" from-port="0" to-layer="148" to-port="1" />
|
|
<edge from-layer="148" from-port="2" to-layer="150" to-port="0" />
|
|
<edge from-layer="149" from-port="0" to-layer="150" 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="157" to-port="0" />
|
|
<edge from-layer="152" from-port="2" to-layer="190" to-port="1" />
|
|
<edge from-layer="153" from-port="0" to-layer="157" to-port="1" />
|
|
<edge from-layer="154" from-port="0" to-layer="157" to-port="2" />
|
|
<edge from-layer="155" from-port="0" to-layer="157" to-port="3" />
|
|
<edge from-layer="156" from-port="0" to-layer="157" to-port="4" />
|
|
<edge from-layer="157" from-port="5" to-layer="162" to-port="0" />
|
|
<edge from-layer="158" from-port="0" to-layer="159" to-port="0" />
|
|
<edge from-layer="159" from-port="1" to-layer="161" to-port="0" />
|
|
<edge from-layer="160" from-port="0" to-layer="161" to-port="1" />
|
|
<edge from-layer="161" from-port="2" to-layer="162" to-port="1" />
|
|
<edge from-layer="161" from-port="2" to-layer="594" to-port="1" />
|
|
<edge from-layer="161" from-port="2" to-layer="701" to-port="1" />
|
|
<edge from-layer="161" from-port="2" to-layer="487" to-port="1" />
|
|
<edge from-layer="161" from-port="2" to-layer="380" to-port="1" />
|
|
<edge from-layer="161" from-port="2" to-layer="273" to-port="1" />
|
|
<edge from-layer="162" from-port="2" to-layer="164" to-port="0" />
|
|
<edge from-layer="163" from-port="0" to-layer="164" to-port="1" />
|
|
<edge from-layer="164" from-port="2" to-layer="166" to-port="0" />
|
|
<edge from-layer="164" from-port="2" to-layer="168" to-port="0" />
|
|
<edge from-layer="164" from-port="2" to-layer="171" to-port="0" />
|
|
<edge from-layer="165" from-port="0" to-layer="166" to-port="1" />
|
|
<edge from-layer="166" from-port="2" to-layer="177" to-port="0" />
|
|
<edge from-layer="167" from-port="0" to-layer="168" to-port="1" />
|
|
<edge from-layer="168" from-port="2" to-layer="170" to-port="0" />
|
|
<edge from-layer="169" from-port="0" to-layer="170" to-port="1" />
|
|
<edge from-layer="170" from-port="2" to-layer="171" to-port="1" />
|
|
<edge from-layer="171" from-port="2" to-layer="173" to-port="0" />
|
|
<edge from-layer="172" from-port="0" to-layer="173" to-port="1" />
|
|
<edge from-layer="173" from-port="2" to-layer="174" to-port="0" />
|
|
<edge from-layer="174" from-port="1" to-layer="176" to-port="0" />
|
|
<edge from-layer="175" from-port="0" to-layer="176" to-port="1" />
|
|
<edge from-layer="176" from-port="2" to-layer="177" to-port="1" />
|
|
<edge from-layer="177" from-port="2" to-layer="182" to-port="0" />
|
|
<edge from-layer="178" from-port="0" to-layer="182" to-port="1" />
|
|
<edge from-layer="179" from-port="0" to-layer="182" to-port="2" />
|
|
<edge from-layer="180" from-port="0" to-layer="182" to-port="3" />
|
|
<edge from-layer="181" from-port="0" to-layer="182" to-port="4" />
|
|
<edge from-layer="182" from-port="5" to-layer="187" to-port="0" />
|
|
<edge from-layer="183" from-port="0" to-layer="184" to-port="0" />
|
|
<edge from-layer="184" from-port="1" to-layer="186" to-port="0" />
|
|
<edge from-layer="185" from-port="0" to-layer="186" to-port="1" />
|
|
<edge from-layer="186" from-port="2" to-layer="615" to-port="1" />
|
|
<edge from-layer="186" from-port="2" to-layer="508" to-port="1" />
|
|
<edge from-layer="186" from-port="2" to-layer="401" to-port="1" />
|
|
<edge from-layer="186" from-port="2" to-layer="294" to-port="1" />
|
|
<edge from-layer="186" from-port="2" to-layer="722" to-port="1" />
|
|
<edge from-layer="186" from-port="2" to-layer="187" to-port="1" />
|
|
<edge from-layer="187" from-port="2" to-layer="189" to-port="0" />
|
|
<edge from-layer="188" from-port="0" to-layer="189" to-port="1" />
|
|
<edge from-layer="189" from-port="2" to-layer="190" to-port="0" />
|
|
<edge from-layer="190" from-port="2" to-layer="192" to-port="0" />
|
|
<edge from-layer="191" from-port="0" to-layer="192" to-port="1" />
|
|
<edge from-layer="192" from-port="2" to-layer="194" to-port="0" />
|
|
<edge from-layer="193" from-port="0" to-layer="194" to-port="1" />
|
|
<edge from-layer="194" from-port="2" to-layer="196" to-port="0" />
|
|
<edge from-layer="195" from-port="0" to-layer="196" to-port="1" />
|
|
<edge from-layer="196" from-port="2" to-layer="218" to-port="0" />
|
|
<edge from-layer="196" from-port="2" to-layer="240" to-port="0" />
|
|
<edge from-layer="196" from-port="2" to-layer="261" to-port="0" />
|
|
<edge from-layer="196" from-port="2" to-layer="201" to-port="0" />
|
|
<edge from-layer="197" from-port="0" to-layer="201" to-port="1" />
|
|
<edge from-layer="198" from-port="0" to-layer="201" to-port="2" />
|
|
<edge from-layer="199" from-port="0" to-layer="201" to-port="3" />
|
|
<edge from-layer="200" from-port="0" to-layer="201" to-port="4" />
|
|
<edge from-layer="201" from-port="5" to-layer="202" to-port="0" />
|
|
<edge from-layer="202" from-port="2" to-layer="204" to-port="0" />
|
|
<edge from-layer="203" from-port="0" to-layer="204" to-port="1" />
|
|
<edge from-layer="204" from-port="2" to-layer="209" to-port="0" />
|
|
<edge from-layer="205" from-port="0" to-layer="209" to-port="1" />
|
|
<edge from-layer="206" from-port="0" to-layer="209" to-port="2" />
|
|
<edge from-layer="207" from-port="0" to-layer="209" to-port="3" />
|
|
<edge from-layer="208" from-port="0" to-layer="209" to-port="4" />
|
|
<edge from-layer="209" from-port="5" to-layer="211" to-port="0" />
|
|
<edge from-layer="210" from-port="0" to-layer="211" to-port="1" />
|
|
<edge from-layer="211" from-port="2" to-layer="213" to-port="0" />
|
|
<edge from-layer="212" from-port="0" to-layer="213" to-port="1" />
|
|
<edge from-layer="213" from-port="2" to-layer="231" to-port="0" />
|
|
<edge from-layer="214" from-port="0" to-layer="218" to-port="1" />
|
|
<edge from-layer="215" from-port="0" to-layer="218" to-port="2" />
|
|
<edge from-layer="216" from-port="0" to-layer="218" to-port="3" />
|
|
<edge from-layer="217" from-port="0" to-layer="218" to-port="4" />
|
|
<edge from-layer="218" from-port="5" to-layer="219" to-port="0" />
|
|
<edge from-layer="219" from-port="2" to-layer="221" to-port="0" />
|
|
<edge from-layer="220" from-port="0" to-layer="221" to-port="1" />
|
|
<edge from-layer="221" from-port="2" to-layer="226" to-port="0" />
|
|
<edge from-layer="222" from-port="0" to-layer="226" to-port="1" />
|
|
<edge from-layer="223" from-port="0" to-layer="226" to-port="2" />
|
|
<edge from-layer="224" from-port="0" to-layer="226" to-port="3" />
|
|
<edge from-layer="225" from-port="0" to-layer="226" to-port="4" />
|
|
<edge from-layer="226" from-port="5" to-layer="228" to-port="0" />
|
|
<edge from-layer="227" from-port="0" to-layer="228" to-port="1" />
|
|
<edge from-layer="228" from-port="2" to-layer="230" to-port="0" />
|
|
<edge from-layer="229" from-port="0" to-layer="230" to-port="1" />
|
|
<edge from-layer="230" from-port="2" to-layer="231" to-port="1" />
|
|
<edge from-layer="231" from-port="2" to-layer="233" to-port="0" />
|
|
<edge from-layer="232" from-port="0" to-layer="233" to-port="1" />
|
|
<edge from-layer="233" from-port="2" to-layer="234" to-port="0" />
|
|
<edge from-layer="234" from-port="2" to-layer="235" to-port="0" />
|
|
<edge from-layer="235" from-port="1" to-layer="248" to-port="0" />
|
|
<edge from-layer="236" from-port="0" to-layer="240" to-port="1" />
|
|
<edge from-layer="237" from-port="0" to-layer="240" to-port="2" />
|
|
<edge from-layer="238" from-port="0" to-layer="240" to-port="3" />
|
|
<edge from-layer="239" from-port="0" to-layer="240" to-port="4" />
|
|
<edge from-layer="240" from-port="5" to-layer="241" to-port="0" />
|
|
<edge from-layer="241" from-port="2" to-layer="243" to-port="0" />
|
|
<edge from-layer="242" from-port="0" to-layer="243" to-port="1" />
|
|
<edge from-layer="243" from-port="2" to-layer="245" to-port="0" />
|
|
<edge from-layer="244" from-port="0" 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="248" from-port="2" to-layer="253" to-port="0" />
|
|
<edge from-layer="249" from-port="0" to-layer="253" to-port="1" />
|
|
<edge from-layer="250" from-port="0" to-layer="253" to-port="2" />
|
|
<edge from-layer="251" from-port="0" to-layer="253" to-port="3" />
|
|
<edge from-layer="252" from-port="0" to-layer="253" to-port="4" />
|
|
<edge from-layer="253" from-port="5" to-layer="255" to-port="0" />
|
|
<edge from-layer="254" from-port="0" to-layer="255" to-port="1" />
|
|
<edge from-layer="255" from-port="2" to-layer="257" to-port="0" />
|
|
<edge from-layer="256" from-port="0" to-layer="257" to-port="1" />
|
|
<edge from-layer="257" from-port="2" to-layer="258" to-port="0" />
|
|
<edge from-layer="258" from-port="2" to-layer="260" to-port="0" />
|
|
<edge from-layer="259" from-port="0" to-layer="260" to-port="1" />
|
|
<edge from-layer="260" from-port="2" to-layer="261" to-port="1" />
|
|
<edge from-layer="261" from-port="2" to-layer="263" to-port="0" />
|
|
<edge from-layer="262" from-port="0" to-layer="263" 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="272" to-port="0" />
|
|
<edge from-layer="267" from-port="2" to-layer="297" to-port="1" />
|
|
<edge from-layer="268" from-port="0" to-layer="272" to-port="1" />
|
|
<edge from-layer="269" from-port="0" to-layer="272" to-port="2" />
|
|
<edge from-layer="270" from-port="0" to-layer="272" to-port="3" />
|
|
<edge from-layer="271" from-port="0" to-layer="272" to-port="4" />
|
|
<edge from-layer="272" from-port="5" to-layer="273" to-port="0" />
|
|
<edge from-layer="273" from-port="2" to-layer="275" to-port="0" />
|
|
<edge from-layer="274" from-port="0" to-layer="275" to-port="1" />
|
|
<edge from-layer="275" from-port="2" to-layer="277" to-port="0" />
|
|
<edge from-layer="275" from-port="2" to-layer="279" to-port="0" />
|
|
<edge from-layer="275" from-port="2" to-layer="282" to-port="0" />
|
|
<edge from-layer="276" from-port="0" to-layer="277" to-port="1" />
|
|
<edge from-layer="277" from-port="2" to-layer="288" to-port="0" />
|
|
<edge from-layer="278" from-port="0" to-layer="279" to-port="1" />
|
|
<edge from-layer="279" from-port="2" to-layer="281" to-port="0" />
|
|
<edge from-layer="280" from-port="0" to-layer="281" to-port="1" />
|
|
<edge from-layer="281" from-port="2" to-layer="282" to-port="1" />
|
|
<edge from-layer="282" from-port="2" to-layer="284" to-port="0" />
|
|
<edge from-layer="283" from-port="0" to-layer="284" to-port="1" />
|
|
<edge from-layer="284" from-port="2" to-layer="285" to-port="0" />
|
|
<edge from-layer="285" from-port="1" to-layer="287" to-port="0" />
|
|
<edge from-layer="286" from-port="0" to-layer="287" to-port="1" />
|
|
<edge from-layer="287" from-port="2" to-layer="288" to-port="1" />
|
|
<edge from-layer="288" from-port="2" to-layer="293" to-port="0" />
|
|
<edge from-layer="289" from-port="0" to-layer="293" to-port="1" />
|
|
<edge from-layer="290" from-port="0" to-layer="293" to-port="2" />
|
|
<edge from-layer="291" from-port="0" to-layer="293" to-port="3" />
|
|
<edge from-layer="292" from-port="0" to-layer="293" to-port="4" />
|
|
<edge from-layer="293" from-port="5" to-layer="294" to-port="0" />
|
|
<edge from-layer="294" from-port="2" to-layer="296" to-port="0" />
|
|
<edge from-layer="295" from-port="0" to-layer="296" to-port="1" />
|
|
<edge from-layer="296" from-port="2" to-layer="297" to-port="0" />
|
|
<edge from-layer="297" from-port="2" to-layer="299" to-port="0" />
|
|
<edge from-layer="298" from-port="0" to-layer="299" to-port="1" />
|
|
<edge from-layer="299" from-port="2" to-layer="301" to-port="0" />
|
|
<edge from-layer="300" from-port="0" to-layer="301" to-port="1" />
|
|
<edge from-layer="301" from-port="2" to-layer="303" to-port="0" />
|
|
<edge from-layer="302" from-port="0" to-layer="303" to-port="1" />
|
|
<edge from-layer="303" from-port="2" to-layer="368" to-port="0" />
|
|
<edge from-layer="303" from-port="2" to-layer="325" to-port="0" />
|
|
<edge from-layer="303" from-port="2" to-layer="308" to-port="0" />
|
|
<edge from-layer="303" from-port="2" to-layer="347" to-port="0" />
|
|
<edge from-layer="304" from-port="0" to-layer="308" to-port="1" />
|
|
<edge from-layer="305" from-port="0" to-layer="308" to-port="2" />
|
|
<edge from-layer="306" from-port="0" to-layer="308" to-port="3" />
|
|
<edge from-layer="307" from-port="0" to-layer="308" to-port="4" />
|
|
<edge from-layer="308" from-port="5" to-layer="309" to-port="0" />
|
|
<edge from-layer="309" from-port="2" to-layer="311" to-port="0" />
|
|
<edge from-layer="310" from-port="0" to-layer="311" to-port="1" />
|
|
<edge from-layer="311" from-port="2" to-layer="316" to-port="0" />
|
|
<edge from-layer="312" from-port="0" to-layer="316" to-port="1" />
|
|
<edge from-layer="313" from-port="0" to-layer="316" to-port="2" />
|
|
<edge from-layer="314" from-port="0" to-layer="316" to-port="3" />
|
|
<edge from-layer="315" from-port="0" to-layer="316" to-port="4" />
|
|
<edge from-layer="316" from-port="5" to-layer="318" to-port="0" />
|
|
<edge from-layer="317" from-port="0" to-layer="318" to-port="1" />
|
|
<edge from-layer="318" from-port="2" to-layer="320" to-port="0" />
|
|
<edge from-layer="319" from-port="0" to-layer="320" to-port="1" />
|
|
<edge from-layer="320" from-port="2" to-layer="338" to-port="0" />
|
|
<edge from-layer="321" from-port="0" to-layer="325" to-port="1" />
|
|
<edge from-layer="322" from-port="0" to-layer="325" to-port="2" />
|
|
<edge from-layer="323" from-port="0" to-layer="325" to-port="3" />
|
|
<edge from-layer="324" from-port="0" to-layer="325" to-port="4" />
|
|
<edge from-layer="325" from-port="5" to-layer="326" to-port="0" />
|
|
<edge from-layer="326" from-port="2" to-layer="328" to-port="0" />
|
|
<edge from-layer="327" from-port="0" to-layer="328" to-port="1" />
|
|
<edge from-layer="328" from-port="2" to-layer="333" to-port="0" />
|
|
<edge from-layer="329" from-port="0" to-layer="333" to-port="1" />
|
|
<edge from-layer="330" from-port="0" to-layer="333" to-port="2" />
|
|
<edge from-layer="331" from-port="0" to-layer="333" to-port="3" />
|
|
<edge from-layer="332" from-port="0" to-layer="333" to-port="4" />
|
|
<edge from-layer="333" from-port="5" to-layer="335" to-port="0" />
|
|
<edge from-layer="334" from-port="0" to-layer="335" to-port="1" />
|
|
<edge from-layer="335" from-port="2" to-layer="337" to-port="0" />
|
|
<edge from-layer="336" from-port="0" to-layer="337" to-port="1" />
|
|
<edge from-layer="337" from-port="2" to-layer="338" to-port="1" />
|
|
<edge from-layer="338" from-port="2" to-layer="340" to-port="0" />
|
|
<edge from-layer="339" from-port="0" to-layer="340" to-port="1" />
|
|
<edge from-layer="340" from-port="2" to-layer="341" to-port="0" />
|
|
<edge from-layer="341" from-port="2" to-layer="342" to-port="0" />
|
|
<edge from-layer="342" from-port="1" to-layer="355" to-port="0" />
|
|
<edge from-layer="343" from-port="0" to-layer="347" to-port="1" />
|
|
<edge from-layer="344" from-port="0" to-layer="347" to-port="2" />
|
|
<edge from-layer="345" from-port="0" to-layer="347" to-port="3" />
|
|
<edge from-layer="346" from-port="0" to-layer="347" to-port="4" />
|
|
<edge from-layer="347" from-port="5" to-layer="348" to-port="0" />
|
|
<edge from-layer="348" from-port="2" to-layer="350" to-port="0" />
|
|
<edge from-layer="349" from-port="0" to-layer="350" to-port="1" />
|
|
<edge from-layer="350" from-port="2" to-layer="352" to-port="0" />
|
|
<edge from-layer="351" from-port="0" to-layer="352" to-port="1" />
|
|
<edge from-layer="352" from-port="2" to-layer="354" to-port="0" />
|
|
<edge from-layer="353" from-port="0" to-layer="354" to-port="1" />
|
|
<edge from-layer="354" from-port="2" to-layer="355" to-port="1" />
|
|
<edge from-layer="355" from-port="2" to-layer="360" to-port="0" />
|
|
<edge from-layer="356" from-port="0" to-layer="360" to-port="1" />
|
|
<edge from-layer="357" from-port="0" to-layer="360" to-port="2" />
|
|
<edge from-layer="358" from-port="0" to-layer="360" to-port="3" />
|
|
<edge from-layer="359" from-port="0" to-layer="360" to-port="4" />
|
|
<edge from-layer="360" from-port="5" to-layer="362" to-port="0" />
|
|
<edge from-layer="361" from-port="0" to-layer="362" to-port="1" />
|
|
<edge from-layer="362" from-port="2" to-layer="364" to-port="0" />
|
|
<edge from-layer="363" from-port="0" to-layer="364" to-port="1" />
|
|
<edge from-layer="364" from-port="2" to-layer="365" to-port="0" />
|
|
<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="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="372" to-port="0" />
|
|
<edge from-layer="371" from-port="0" 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" />
|
|
<edge from-layer="374" from-port="2" to-layer="404" to-port="1" />
|
|
<edge from-layer="374" from-port="2" to-layer="379" to-port="0" />
|
|
<edge from-layer="375" from-port="0" to-layer="379" to-port="1" />
|
|
<edge from-layer="376" from-port="0" to-layer="379" to-port="2" />
|
|
<edge from-layer="377" from-port="0" to-layer="379" to-port="3" />
|
|
<edge from-layer="378" from-port="0" to-layer="379" to-port="4" />
|
|
<edge from-layer="379" from-port="5" to-layer="380" to-port="0" />
|
|
<edge from-layer="380" from-port="2" to-layer="382" to-port="0" />
|
|
<edge from-layer="381" from-port="0" to-layer="382" to-port="1" />
|
|
<edge from-layer="382" from-port="2" to-layer="389" to-port="0" />
|
|
<edge from-layer="382" from-port="2" to-layer="386" to-port="0" />
|
|
<edge from-layer="382" from-port="2" to-layer="384" to-port="0" />
|
|
<edge from-layer="383" from-port="0" to-layer="384" to-port="1" />
|
|
<edge from-layer="384" from-port="2" to-layer="395" to-port="0" />
|
|
<edge from-layer="385" from-port="0" to-layer="386" to-port="1" />
|
|
<edge from-layer="386" from-port="2" to-layer="388" to-port="0" />
|
|
<edge from-layer="387" from-port="0" to-layer="388" to-port="1" />
|
|
<edge from-layer="388" from-port="2" to-layer="389" to-port="1" />
|
|
<edge from-layer="389" from-port="2" to-layer="391" to-port="0" />
|
|
<edge from-layer="390" from-port="0" to-layer="391" to-port="1" />
|
|
<edge from-layer="391" from-port="2" to-layer="392" to-port="0" />
|
|
<edge from-layer="392" from-port="1" to-layer="394" to-port="0" />
|
|
<edge from-layer="393" from-port="0" to-layer="394" to-port="1" />
|
|
<edge from-layer="394" from-port="2" to-layer="395" to-port="1" />
|
|
<edge from-layer="395" from-port="2" to-layer="400" to-port="0" />
|
|
<edge from-layer="396" from-port="0" to-layer="400" to-port="1" />
|
|
<edge from-layer="397" from-port="0" to-layer="400" to-port="2" />
|
|
<edge from-layer="398" from-port="0" to-layer="400" to-port="3" />
|
|
<edge from-layer="399" from-port="0" to-layer="400" to-port="4" />
|
|
<edge from-layer="400" from-port="5" to-layer="401" to-port="0" />
|
|
<edge from-layer="401" from-port="2" to-layer="403" to-port="0" />
|
|
<edge from-layer="402" from-port="0" to-layer="403" to-port="1" />
|
|
<edge from-layer="403" from-port="2" to-layer="404" to-port="0" />
|
|
<edge from-layer="404" from-port="2" to-layer="406" to-port="0" />
|
|
<edge from-layer="405" from-port="0" to-layer="406" 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="415" to-port="0" />
|
|
<edge from-layer="410" from-port="2" to-layer="432" to-port="0" />
|
|
<edge from-layer="410" from-port="2" to-layer="454" to-port="0" />
|
|
<edge from-layer="410" from-port="2" to-layer="475" to-port="0" />
|
|
<edge from-layer="411" from-port="0" to-layer="415" to-port="1" />
|
|
<edge from-layer="412" from-port="0" to-layer="415" to-port="2" />
|
|
<edge from-layer="413" from-port="0" to-layer="415" to-port="3" />
|
|
<edge from-layer="414" from-port="0" to-layer="415" to-port="4" />
|
|
<edge from-layer="415" from-port="5" to-layer="416" to-port="0" />
|
|
<edge from-layer="416" from-port="2" to-layer="418" to-port="0" />
|
|
<edge from-layer="417" from-port="0" to-layer="418" to-port="1" />
|
|
<edge from-layer="418" from-port="2" to-layer="423" to-port="0" />
|
|
<edge from-layer="419" from-port="0" to-layer="423" to-port="1" />
|
|
<edge from-layer="420" from-port="0" to-layer="423" to-port="2" />
|
|
<edge from-layer="421" from-port="0" to-layer="423" to-port="3" />
|
|
<edge from-layer="422" from-port="0" to-layer="423" to-port="4" />
|
|
<edge from-layer="423" from-port="5" to-layer="425" to-port="0" />
|
|
<edge from-layer="424" from-port="0" to-layer="425" to-port="1" />
|
|
<edge from-layer="425" from-port="2" to-layer="427" to-port="0" />
|
|
<edge from-layer="426" from-port="0" to-layer="427" to-port="1" />
|
|
<edge from-layer="427" from-port="2" to-layer="445" to-port="0" />
|
|
<edge from-layer="428" from-port="0" to-layer="432" to-port="1" />
|
|
<edge from-layer="429" from-port="0" to-layer="432" to-port="2" />
|
|
<edge from-layer="430" from-port="0" to-layer="432" to-port="3" />
|
|
<edge from-layer="431" from-port="0" to-layer="432" to-port="4" />
|
|
<edge from-layer="432" from-port="5" to-layer="433" to-port="0" />
|
|
<edge from-layer="433" from-port="2" to-layer="435" to-port="0" />
|
|
<edge from-layer="434" from-port="0" to-layer="435" to-port="1" />
|
|
<edge from-layer="435" from-port="2" to-layer="440" to-port="0" />
|
|
<edge from-layer="436" from-port="0" to-layer="440" to-port="1" />
|
|
<edge from-layer="437" from-port="0" to-layer="440" to-port="2" />
|
|
<edge from-layer="438" from-port="0" to-layer="440" to-port="3" />
|
|
<edge from-layer="439" from-port="0" to-layer="440" to-port="4" />
|
|
<edge from-layer="440" from-port="5" to-layer="442" to-port="0" />
|
|
<edge from-layer="441" from-port="0" to-layer="442" to-port="1" />
|
|
<edge from-layer="442" from-port="2" to-layer="444" to-port="0" />
|
|
<edge from-layer="443" from-port="0" to-layer="444" to-port="1" />
|
|
<edge from-layer="444" from-port="2" to-layer="445" to-port="1" />
|
|
<edge from-layer="445" from-port="2" to-layer="447" to-port="0" />
|
|
<edge from-layer="446" from-port="0" to-layer="447" to-port="1" />
|
|
<edge from-layer="447" from-port="2" to-layer="448" to-port="0" />
|
|
<edge from-layer="448" from-port="2" to-layer="449" to-port="0" />
|
|
<edge from-layer="449" from-port="1" to-layer="462" to-port="0" />
|
|
<edge from-layer="450" from-port="0" to-layer="454" to-port="1" />
|
|
<edge from-layer="451" from-port="0" to-layer="454" to-port="2" />
|
|
<edge from-layer="452" from-port="0" to-layer="454" to-port="3" />
|
|
<edge from-layer="453" from-port="0" to-layer="454" to-port="4" />
|
|
<edge from-layer="454" from-port="5" to-layer="455" to-port="0" />
|
|
<edge from-layer="455" from-port="2" to-layer="457" to-port="0" />
|
|
<edge from-layer="456" from-port="0" to-layer="457" to-port="1" />
|
|
<edge from-layer="457" from-port="2" to-layer="459" to-port="0" />
|
|
<edge from-layer="458" from-port="0" to-layer="459" to-port="1" />
|
|
<edge from-layer="459" from-port="2" to-layer="461" to-port="0" />
|
|
<edge from-layer="460" from-port="0" to-layer="461" to-port="1" />
|
|
<edge from-layer="461" from-port="2" to-layer="462" to-port="1" />
|
|
<edge from-layer="462" from-port="2" to-layer="467" to-port="0" />
|
|
<edge from-layer="463" from-port="0" to-layer="467" to-port="1" />
|
|
<edge from-layer="464" from-port="0" to-layer="467" to-port="2" />
|
|
<edge from-layer="465" from-port="0" to-layer="467" to-port="3" />
|
|
<edge from-layer="466" from-port="0" to-layer="467" to-port="4" />
|
|
<edge from-layer="467" from-port="5" to-layer="469" to-port="0" />
|
|
<edge from-layer="468" from-port="0" to-layer="469" to-port="1" />
|
|
<edge from-layer="469" from-port="2" to-layer="471" to-port="0" />
|
|
<edge from-layer="470" from-port="0" to-layer="471" to-port="1" />
|
|
<edge from-layer="471" from-port="2" to-layer="472" to-port="0" />
|
|
<edge from-layer="472" from-port="2" to-layer="474" to-port="0" />
|
|
<edge from-layer="473" from-port="0" to-layer="474" to-port="1" />
|
|
<edge from-layer="474" from-port="2" to-layer="475" to-port="1" />
|
|
<edge from-layer="475" from-port="2" to-layer="477" to-port="0" />
|
|
<edge from-layer="476" from-port="0" to-layer="477" to-port="1" />
|
|
<edge from-layer="477" from-port="2" to-layer="479" to-port="0" />
|
|
<edge from-layer="478" from-port="0" to-layer="479" to-port="1" />
|
|
<edge from-layer="479" from-port="2" to-layer="481" to-port="0" />
|
|
<edge from-layer="480" from-port="0" to-layer="481" to-port="1" />
|
|
<edge from-layer="481" from-port="2" to-layer="511" to-port="1" />
|
|
<edge from-layer="481" from-port="2" to-layer="486" to-port="0" />
|
|
<edge from-layer="482" from-port="0" to-layer="486" to-port="1" />
|
|
<edge from-layer="483" from-port="0" to-layer="486" to-port="2" />
|
|
<edge from-layer="484" from-port="0" to-layer="486" to-port="3" />
|
|
<edge from-layer="485" from-port="0" to-layer="486" to-port="4" />
|
|
<edge from-layer="486" from-port="5" to-layer="487" to-port="0" />
|
|
<edge from-layer="487" from-port="2" to-layer="489" to-port="0" />
|
|
<edge from-layer="488" from-port="0" to-layer="489" to-port="1" />
|
|
<edge from-layer="489" from-port="2" to-layer="491" to-port="0" />
|
|
<edge from-layer="489" from-port="2" to-layer="493" to-port="0" />
|
|
<edge from-layer="489" from-port="2" to-layer="496" to-port="0" />
|
|
<edge from-layer="490" from-port="0" to-layer="491" to-port="1" />
|
|
<edge from-layer="491" from-port="2" to-layer="502" to-port="0" />
|
|
<edge from-layer="492" from-port="0" to-layer="493" to-port="1" />
|
|
<edge from-layer="493" from-port="2" to-layer="495" to-port="0" />
|
|
<edge from-layer="494" from-port="0" to-layer="495" to-port="1" />
|
|
<edge from-layer="495" from-port="2" to-layer="496" to-port="1" />
|
|
<edge from-layer="496" from-port="2" to-layer="498" to-port="0" />
|
|
<edge from-layer="497" from-port="0" to-layer="498" to-port="1" />
|
|
<edge from-layer="498" from-port="2" to-layer="499" to-port="0" />
|
|
<edge from-layer="499" from-port="1" to-layer="501" to-port="0" />
|
|
<edge from-layer="500" from-port="0" to-layer="501" to-port="1" />
|
|
<edge from-layer="501" from-port="2" to-layer="502" to-port="1" />
|
|
<edge from-layer="502" from-port="2" to-layer="507" to-port="0" />
|
|
<edge from-layer="503" from-port="0" to-layer="507" to-port="1" />
|
|
<edge from-layer="504" from-port="0" to-layer="507" to-port="2" />
|
|
<edge from-layer="505" from-port="0" to-layer="507" to-port="3" />
|
|
<edge from-layer="506" from-port="0" to-layer="507" to-port="4" />
|
|
<edge from-layer="507" from-port="5" to-layer="508" to-port="0" />
|
|
<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="511" to-port="0" />
|
|
<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="515" to-port="0" />
|
|
<edge from-layer="514" from-port="0" 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" />
|
|
<edge from-layer="517" from-port="2" to-layer="522" to-port="0" />
|
|
<edge from-layer="517" from-port="2" to-layer="539" to-port="0" />
|
|
<edge from-layer="517" from-port="2" to-layer="582" to-port="0" />
|
|
<edge from-layer="517" from-port="2" to-layer="561" to-port="0" />
|
|
<edge from-layer="518" from-port="0" to-layer="522" to-port="1" />
|
|
<edge from-layer="519" from-port="0" to-layer="522" to-port="2" />
|
|
<edge from-layer="520" from-port="0" to-layer="522" to-port="3" />
|
|
<edge from-layer="521" from-port="0" to-layer="522" to-port="4" />
|
|
<edge from-layer="522" from-port="5" to-layer="523" to-port="0" />
|
|
<edge from-layer="523" from-port="2" to-layer="525" to-port="0" />
|
|
<edge from-layer="524" from-port="0" to-layer="525" to-port="1" />
|
|
<edge from-layer="525" from-port="2" to-layer="530" to-port="0" />
|
|
<edge from-layer="526" from-port="0" to-layer="530" to-port="1" />
|
|
<edge from-layer="527" from-port="0" to-layer="530" to-port="2" />
|
|
<edge from-layer="528" from-port="0" to-layer="530" to-port="3" />
|
|
<edge from-layer="529" from-port="0" to-layer="530" to-port="4" />
|
|
<edge from-layer="530" from-port="5" 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="534" to-port="0" />
|
|
<edge from-layer="533" from-port="0" to-layer="534" to-port="1" />
|
|
<edge from-layer="534" from-port="2" to-layer="552" to-port="0" />
|
|
<edge from-layer="535" from-port="0" to-layer="539" to-port="1" />
|
|
<edge from-layer="536" from-port="0" to-layer="539" to-port="2" />
|
|
<edge from-layer="537" from-port="0" to-layer="539" to-port="3" />
|
|
<edge from-layer="538" from-port="0" to-layer="539" to-port="4" />
|
|
<edge from-layer="539" from-port="5" to-layer="540" to-port="0" />
|
|
<edge from-layer="540" from-port="2" to-layer="542" to-port="0" />
|
|
<edge from-layer="541" from-port="0" to-layer="542" to-port="1" />
|
|
<edge from-layer="542" from-port="2" to-layer="547" to-port="0" />
|
|
<edge from-layer="543" from-port="0" to-layer="547" to-port="1" />
|
|
<edge from-layer="544" from-port="0" to-layer="547" to-port="2" />
|
|
<edge from-layer="545" from-port="0" to-layer="547" to-port="3" />
|
|
<edge from-layer="546" from-port="0" to-layer="547" to-port="4" />
|
|
<edge from-layer="547" from-port="5" to-layer="549" to-port="0" />
|
|
<edge from-layer="548" from-port="0" to-layer="549" 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="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="0" />
|
|
<edge from-layer="555" from-port="2" to-layer="556" to-port="0" />
|
|
<edge from-layer="556" from-port="1" to-layer="569" to-port="0" />
|
|
<edge from-layer="557" from-port="0" to-layer="561" to-port="1" />
|
|
<edge from-layer="558" from-port="0" to-layer="561" to-port="2" />
|
|
<edge from-layer="559" from-port="0" to-layer="561" to-port="3" />
|
|
<edge from-layer="560" from-port="0" to-layer="561" to-port="4" />
|
|
<edge from-layer="561" from-port="5" to-layer="562" to-port="0" />
|
|
<edge from-layer="562" from-port="2" to-layer="564" to-port="0" />
|
|
<edge from-layer="563" from-port="0" to-layer="564" to-port="1" />
|
|
<edge from-layer="564" from-port="2" to-layer="566" to-port="0" />
|
|
<edge from-layer="565" from-port="0" to-layer="566" to-port="1" />
|
|
<edge from-layer="566" from-port="2" to-layer="568" to-port="0" />
|
|
<edge from-layer="567" from-port="0" to-layer="568" to-port="1" />
|
|
<edge from-layer="568" from-port="2" to-layer="569" to-port="1" />
|
|
<edge from-layer="569" from-port="2" to-layer="574" to-port="0" />
|
|
<edge from-layer="570" from-port="0" to-layer="574" to-port="1" />
|
|
<edge from-layer="571" from-port="0" to-layer="574" to-port="2" />
|
|
<edge from-layer="572" from-port="0" to-layer="574" to-port="3" />
|
|
<edge from-layer="573" from-port="0" to-layer="574" to-port="4" />
|
|
<edge from-layer="574" from-port="5" to-layer="576" to-port="0" />
|
|
<edge from-layer="575" from-port="0" to-layer="576" to-port="1" />
|
|
<edge from-layer="576" from-port="2" to-layer="578" to-port="0" />
|
|
<edge from-layer="577" from-port="0" to-layer="578" to-port="1" />
|
|
<edge from-layer="578" from-port="2" to-layer="579" to-port="0" />
|
|
<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="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="586" to-port="0" />
|
|
<edge from-layer="585" from-port="0" to-layer="586" to-port="1" />
|
|
<edge from-layer="586" from-port="2" to-layer="588" to-port="0" />
|
|
<edge from-layer="587" from-port="0" to-layer="588" to-port="1" />
|
|
<edge from-layer="588" from-port="2" to-layer="593" to-port="0" />
|
|
<edge from-layer="588" from-port="2" to-layer="618" to-port="1" />
|
|
<edge from-layer="589" from-port="0" to-layer="593" to-port="1" />
|
|
<edge from-layer="590" from-port="0" to-layer="593" to-port="2" />
|
|
<edge from-layer="591" from-port="0" to-layer="593" to-port="3" />
|
|
<edge from-layer="592" from-port="0" to-layer="593" to-port="4" />
|
|
<edge from-layer="593" from-port="5" to-layer="594" to-port="0" />
|
|
<edge from-layer="594" from-port="2" to-layer="596" to-port="0" />
|
|
<edge from-layer="595" from-port="0" to-layer="596" to-port="1" />
|
|
<edge from-layer="596" from-port="2" to-layer="603" to-port="0" />
|
|
<edge from-layer="596" from-port="2" to-layer="600" to-port="0" />
|
|
<edge from-layer="596" from-port="2" to-layer="598" to-port="0" />
|
|
<edge from-layer="597" from-port="0" to-layer="598" to-port="1" />
|
|
<edge from-layer="598" from-port="2" to-layer="609" to-port="0" />
|
|
<edge from-layer="599" from-port="0" to-layer="600" to-port="1" />
|
|
<edge from-layer="600" from-port="2" to-layer="602" to-port="0" />
|
|
<edge from-layer="601" from-port="0" to-layer="602" to-port="1" />
|
|
<edge from-layer="602" from-port="2" to-layer="603" to-port="1" />
|
|
<edge from-layer="603" from-port="2" to-layer="605" to-port="0" />
|
|
<edge from-layer="604" from-port="0" to-layer="605" to-port="1" />
|
|
<edge from-layer="605" from-port="2" to-layer="606" to-port="0" />
|
|
<edge from-layer="606" from-port="1" to-layer="608" to-port="0" />
|
|
<edge from-layer="607" from-port="0" to-layer="608" to-port="1" />
|
|
<edge from-layer="608" from-port="2" to-layer="609" to-port="1" />
|
|
<edge from-layer="609" from-port="2" to-layer="614" to-port="0" />
|
|
<edge from-layer="610" from-port="0" to-layer="614" to-port="1" />
|
|
<edge from-layer="611" from-port="0" to-layer="614" to-port="2" />
|
|
<edge from-layer="612" from-port="0" to-layer="614" to-port="3" />
|
|
<edge from-layer="613" from-port="0" to-layer="614" to-port="4" />
|
|
<edge from-layer="614" from-port="5" to-layer="615" to-port="0" />
|
|
<edge from-layer="615" from-port="2" to-layer="617" to-port="0" />
|
|
<edge from-layer="616" from-port="0" to-layer="617" to-port="1" />
|
|
<edge from-layer="617" from-port="2" to-layer="618" to-port="0" />
|
|
<edge from-layer="618" from-port="2" to-layer="620" to-port="0" />
|
|
<edge from-layer="619" from-port="0" to-layer="620" to-port="1" />
|
|
<edge from-layer="620" from-port="2" to-layer="622" to-port="0" />
|
|
<edge from-layer="621" from-port="0" to-layer="622" to-port="1" />
|
|
<edge from-layer="622" from-port="2" to-layer="624" to-port="0" />
|
|
<edge from-layer="623" from-port="0" to-layer="624" to-port="1" />
|
|
<edge from-layer="624" from-port="2" to-layer="629" to-port="0" />
|
|
<edge from-layer="624" from-port="2" to-layer="646" to-port="0" />
|
|
<edge from-layer="624" from-port="2" to-layer="668" to-port="0" />
|
|
<edge from-layer="624" from-port="2" to-layer="689" to-port="0" />
|
|
<edge from-layer="625" from-port="0" to-layer="629" to-port="1" />
|
|
<edge from-layer="626" from-port="0" to-layer="629" to-port="2" />
|
|
<edge from-layer="627" from-port="0" to-layer="629" to-port="3" />
|
|
<edge from-layer="628" from-port="0" to-layer="629" to-port="4" />
|
|
<edge from-layer="629" from-port="5" to-layer="630" to-port="0" />
|
|
<edge from-layer="630" from-port="2" to-layer="632" to-port="0" />
|
|
<edge from-layer="631" from-port="0" to-layer="632" to-port="1" />
|
|
<edge from-layer="632" from-port="2" to-layer="637" to-port="0" />
|
|
<edge from-layer="633" from-port="0" to-layer="637" to-port="1" />
|
|
<edge from-layer="634" from-port="0" to-layer="637" to-port="2" />
|
|
<edge from-layer="635" from-port="0" to-layer="637" to-port="3" />
|
|
<edge from-layer="636" from-port="0" to-layer="637" to-port="4" />
|
|
<edge from-layer="637" from-port="5" to-layer="639" to-port="0" />
|
|
<edge from-layer="638" from-port="0" to-layer="639" to-port="1" />
|
|
<edge from-layer="639" from-port="2" to-layer="641" to-port="0" />
|
|
<edge from-layer="640" from-port="0" to-layer="641" to-port="1" />
|
|
<edge from-layer="641" from-port="2" to-layer="659" to-port="0" />
|
|
<edge from-layer="642" from-port="0" to-layer="646" to-port="1" />
|
|
<edge from-layer="643" from-port="0" to-layer="646" to-port="2" />
|
|
<edge from-layer="644" from-port="0" to-layer="646" to-port="3" />
|
|
<edge from-layer="645" from-port="0" to-layer="646" to-port="4" />
|
|
<edge from-layer="646" from-port="5" to-layer="647" to-port="0" />
|
|
<edge from-layer="647" from-port="2" to-layer="649" to-port="0" />
|
|
<edge from-layer="648" from-port="0" to-layer="649" to-port="1" />
|
|
<edge from-layer="649" from-port="2" to-layer="654" to-port="0" />
|
|
<edge from-layer="650" from-port="0" to-layer="654" to-port="1" />
|
|
<edge from-layer="651" from-port="0" to-layer="654" to-port="2" />
|
|
<edge from-layer="652" from-port="0" to-layer="654" to-port="3" />
|
|
<edge from-layer="653" from-port="0" to-layer="654" to-port="4" />
|
|
<edge from-layer="654" from-port="5" to-layer="656" to-port="0" />
|
|
<edge from-layer="655" from-port="0" to-layer="656" to-port="1" />
|
|
<edge from-layer="656" from-port="2" to-layer="658" to-port="0" />
|
|
<edge from-layer="657" from-port="0" to-layer="658" to-port="1" />
|
|
<edge from-layer="658" from-port="2" to-layer="659" to-port="1" />
|
|
<edge from-layer="659" from-port="2" to-layer="661" to-port="0" />
|
|
<edge from-layer="660" from-port="0" to-layer="661" to-port="1" />
|
|
<edge from-layer="661" from-port="2" to-layer="662" to-port="0" />
|
|
<edge from-layer="662" from-port="2" to-layer="663" to-port="0" />
|
|
<edge from-layer="663" from-port="1" to-layer="676" to-port="0" />
|
|
<edge from-layer="664" from-port="0" to-layer="668" to-port="1" />
|
|
<edge from-layer="665" from-port="0" to-layer="668" to-port="2" />
|
|
<edge from-layer="666" from-port="0" to-layer="668" to-port="3" />
|
|
<edge from-layer="667" from-port="0" to-layer="668" to-port="4" />
|
|
<edge from-layer="668" from-port="5" to-layer="669" to-port="0" />
|
|
<edge from-layer="669" from-port="2" to-layer="671" to-port="0" />
|
|
<edge from-layer="670" from-port="0" to-layer="671" to-port="1" />
|
|
<edge from-layer="671" from-port="2" to-layer="673" to-port="0" />
|
|
<edge from-layer="672" from-port="0" to-layer="673" to-port="1" />
|
|
<edge from-layer="673" from-port="2" to-layer="675" to-port="0" />
|
|
<edge from-layer="674" from-port="0" to-layer="675" to-port="1" />
|
|
<edge from-layer="675" from-port="2" to-layer="676" to-port="1" />
|
|
<edge from-layer="676" from-port="2" to-layer="681" to-port="0" />
|
|
<edge from-layer="677" from-port="0" to-layer="681" to-port="1" />
|
|
<edge from-layer="678" from-port="0" to-layer="681" to-port="2" />
|
|
<edge from-layer="679" from-port="0" to-layer="681" to-port="3" />
|
|
<edge from-layer="680" from-port="0" to-layer="681" to-port="4" />
|
|
<edge from-layer="681" from-port="5" to-layer="683" to-port="0" />
|
|
<edge from-layer="682" from-port="0" to-layer="683" to-port="1" />
|
|
<edge from-layer="683" from-port="2" to-layer="685" to-port="0" />
|
|
<edge from-layer="684" from-port="0" to-layer="685" to-port="1" />
|
|
<edge from-layer="685" from-port="2" to-layer="686" to-port="0" />
|
|
<edge from-layer="686" from-port="2" to-layer="688" to-port="0" />
|
|
<edge from-layer="687" from-port="0" to-layer="688" to-port="1" />
|
|
<edge from-layer="688" from-port="2" to-layer="689" to-port="1" />
|
|
<edge from-layer="689" from-port="2" to-layer="691" to-port="0" />
|
|
<edge from-layer="690" from-port="0" to-layer="691" to-port="1" />
|
|
<edge from-layer="691" from-port="2" to-layer="693" to-port="0" />
|
|
<edge from-layer="692" from-port="0" to-layer="693" to-port="1" />
|
|
<edge from-layer="693" from-port="2" to-layer="695" to-port="0" />
|
|
<edge from-layer="694" from-port="0" to-layer="695" to-port="1" />
|
|
<edge from-layer="695" from-port="2" to-layer="700" to-port="0" />
|
|
<edge from-layer="695" from-port="2" to-layer="725" to-port="1" />
|
|
<edge from-layer="696" from-port="0" to-layer="700" to-port="1" />
|
|
<edge from-layer="697" from-port="0" to-layer="700" to-port="2" />
|
|
<edge from-layer="698" from-port="0" to-layer="700" to-port="3" />
|
|
<edge from-layer="699" from-port="0" to-layer="700" to-port="4" />
|
|
<edge from-layer="700" from-port="5" to-layer="701" to-port="0" />
|
|
<edge from-layer="701" from-port="2" to-layer="703" to-port="0" />
|
|
<edge from-layer="702" from-port="0" to-layer="703" to-port="1" />
|
|
<edge from-layer="703" from-port="2" to-layer="705" to-port="0" />
|
|
<edge from-layer="703" from-port="2" to-layer="707" to-port="0" />
|
|
<edge from-layer="703" from-port="2" to-layer="710" to-port="0" />
|
|
<edge from-layer="704" from-port="0" to-layer="705" to-port="1" />
|
|
<edge from-layer="705" from-port="2" to-layer="716" to-port="0" />
|
|
<edge from-layer="706" from-port="0" to-layer="707" to-port="1" />
|
|
<edge from-layer="707" from-port="2" to-layer="709" to-port="0" />
|
|
<edge from-layer="708" from-port="0" to-layer="709" to-port="1" />
|
|
<edge from-layer="709" from-port="2" to-layer="710" to-port="1" />
|
|
<edge from-layer="710" from-port="2" to-layer="712" to-port="0" />
|
|
<edge from-layer="711" from-port="0" to-layer="712" to-port="1" />
|
|
<edge from-layer="712" from-port="2" to-layer="713" to-port="0" />
|
|
<edge from-layer="713" from-port="1" to-layer="715" to-port="0" />
|
|
<edge from-layer="714" from-port="0" to-layer="715" to-port="1" />
|
|
<edge from-layer="715" from-port="2" to-layer="716" to-port="1" />
|
|
<edge from-layer="716" from-port="2" to-layer="721" to-port="0" />
|
|
<edge from-layer="717" from-port="0" to-layer="721" to-port="1" />
|
|
<edge from-layer="718" from-port="0" to-layer="721" to-port="2" />
|
|
<edge from-layer="719" from-port="0" to-layer="721" to-port="3" />
|
|
<edge from-layer="720" from-port="0" to-layer="721" to-port="4" />
|
|
<edge from-layer="721" from-port="5" to-layer="722" to-port="0" />
|
|
<edge from-layer="722" from-port="2" to-layer="724" to-port="0" />
|
|
<edge from-layer="723" from-port="0" to-layer="724" to-port="1" />
|
|
<edge from-layer="724" from-port="2" to-layer="725" to-port="0" />
|
|
<edge from-layer="725" from-port="2" to-layer="727" to-port="0" />
|
|
<edge from-layer="726" from-port="0" to-layer="727" to-port="1" />
|
|
<edge from-layer="727" from-port="2" to-layer="729" to-port="0" />
|
|
<edge from-layer="728" from-port="0" to-layer="729" to-port="1" />
|
|
<edge from-layer="729" from-port="2" to-layer="731" to-port="0" />
|
|
<edge from-layer="730" from-port="0" to-layer="731" to-port="1" />
|
|
<edge from-layer="731" from-port="2" to-layer="732" to-port="0" />
|
|
</edges>
|
|
<rt_info>
|
|
<Runtime_version value="2024.4.1-16618-643f23d1318-releases/2024/4" />
|
|
<conversion_parameters>
|
|
<framework value="pytorch" />
|
|
<is_python_object value="True" />
|
|
</conversion_parameters>
|
|
<nncf>
|
|
<friendly_names_were_updated value="True" />
|
|
<quantization>
|
|
<advanced_parameters value="{'overflow_fix': 'disable', 'quantize_outputs': False, 'inplace_statistics': True, 'disable_channel_alignment': True, 'disable_bias_correction': False, 'batchwise_statistics': None, 'activations_quantization_params': None, 'weights_quantization_params': None, 'activations_range_estimator_params': {'min': {'statistics_type': None, 'aggregator_type': None, 'clipping_value': None, 'quantile_outlier_prob': 0.0001}, 'max': {'statistics_type': None, 'aggregator_type': None, 'clipping_value': None, 'quantile_outlier_prob': 0.0001}}, 'weights_range_estimator_params': {'min': {'statistics_type': None, 'aggregator_type': None, 'clipping_value': None, 'quantile_outlier_prob': 0.0001}, 'max': {'statistics_type': None, 'aggregator_type': None, 'clipping_value': None, 'quantile_outlier_prob': 0.0001}}, 'bias_correction_params': {'apply_for_all_nodes': False, 'threshold': None}, 'smooth_quant_alphas': {'convolution': -1, 'matmul': 0.95}, 'smooth_quant_alpha': None, 'backend_params': {}}" />
|
|
<fast_bias_correction value="True" />
|
|
<ignored_scope>
|
|
<types value="['GroupNormalization']" />
|
|
</ignored_scope>
|
|
<model_type value="transformer" />
|
|
<preset value="mixed" />
|
|
<subset_size value="300" />
|
|
<target_device value="ANY" />
|
|
</quantization>
|
|
</nncf>
|
|
<optimum>
|
|
<optimum_intel_version value="1.20.0.dev0+b31524c" />
|
|
<optimum_version value="1.23.0" />
|
|
<pytorch_version value="2.5.0.dev20240807+cu121" />
|
|
<transformers_version value="4.43.4" />
|
|
</optimum>
|
|
</rt_info>
|
|
</net>
|