I’m trying to port a pytorch model to run on the NPU. I’ve exported both an onnx file and a .pt file using pytorch version 1.2.0 and onnx version 1.4.1. When I try to run the convert_onnx script, I get the following error:
I Start importing onnx...
I Current ONNX Model use ir_version 4 opset_version 9
Traceback (most recent call last):
File "convertonnx.py", line 40, in <module>
File "convertonnx.py", line 33, in main
File "acuitylib/vsi_nn.py", line 138, in load_onnx
File "acuitylib/app/importer/import_onnx.py", line 72, in run
File "acuitylib/converter/convert_onnx.py", line 472, in __init__
File "acuitylib/converter/convert_onnx.py", line 162, in infer_shape
File "acuitylib/converter/convert_onnx.py", line 125, in __status_update
File "acuitylib/converter/convert_onnx.py", line 159, in __model_rebuild
File "acuitylib/converter/tensor_model.py", line 330, in build_onnx_model
KeyError: ''
[2681] Failed to execute script convertonnx
When I try to use the pytorch importer, I get a similar error:
I Start importing pytorch...
mclaren_512.pt ********************
WARNING: Token 'COMMENT' defined, but not used
WARNING: There is 1 unused token
Traceback (most recent call last):
File "convertpytorch.py", line 51, in <module>
File "convertpytorch.py", line 41, in main
File "acuitylib/vsi_nn.py", line 146, in load_pytorch
File "acuitylib/app/importer/import_pytorch.py", line 92, in run
File "acuitylib/converter/convert_pytorch.py", line 526, in __init__
File "acuitylib/converter/convert_pytorch.py", line 610, in model_simplify
File "acuitylib/converter/convert_pytorch.py", line 107, in shape_pick
File "acuitylib/converter/convert_pytorch.py", line 143, in __ir_shape_inference
KeyError: 'aten::slice'
[2709] Failed to execute script convertpytorch
I checked the docs, all the layers should be supported. Can someone help me figure out what is going wrong and how to fix it?