PyTorch model conversion

johndoe · December 10, 2021, 1:16pm

I’m unable to convert a model in pytorch

This is the script to download the pretrained weights

import torch
import torchvision.models as models
resnet18 = models.resnet18(pretrained=True)

torch.save(resnet18, 'resnet18.pt')

This is the script for converting

./convert 
--model-name resnet18 
--platform pytorch 
--model resnet18.pt 
--input-size-list '3,224,224' 
--mean-values '103.94,116.78,123.68,58.82' 
--quantized-dtype asymmetric_affine 
--kboard VIM3 --print-level 1

Here’s the error:

Start import model ...
I Namespace(config=None, import='pytorch', input_size_list='3,224,224', inputs=None, model='resnet18.pt', output_data='resnet18.data', output_model='resnet18.json', outputs=None, size_with_batch=None, which='import')

I Start importing pytorch...

[50514] Failed to execute script pegasus

Traceback (most recent call last):

  File "pegasus.py", line 131, in <module>

  File "pegasus.py", line 112, in main

  File "acuitylib/app/importer/commands.py", line 286, in execute

  File "acuitylib/vsi_nn.py", line 173, in load_pytorch_by_onnx_backend

  File "torch/jit/__init__.py", line 162, in load

RuntimeError: [enforce fail at inline_container.cc:137] . PytorchStreamReader failed reading zip archive: failed finding central directory

frame #0: c10::ThrowEnforceNotMet(char const*, int, char const*, std::string const&, void const*) + 0x47 (0x7fe3f02eee17 in /home/saswat/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libc10.so)

frame #1: caffe2::serialize::PyTorchStreamReader::valid(char const*) + 0x6b (0x7fe3f327775b in /home/saswat/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libtorch.so)

frame #2: caffe2::serialize::PyTorchStreamReader::init() + 0x9a (0x7fe3f327b20a in /home/saswat/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libtorch.so)

frame #3: caffe2::serialize::PyTorchStreamReader::PyTorchStreamReader(std::string const&) + 0x60 (0x7fe3f327e270 in /home/saswat/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libtorch.so)

frame #4: torch::jit::import_ir_module(std::shared_ptr<torch::jit::script::CompilationUnit>, std::string const&, c10::optional<c10::Device>, std::unordered_map<std::string, std::string, std::hash<std::string>, std::equal_to<std::string>, std::allocator<std::pair<std::string const, std::string> > >&) + 0x38 (0x7fe3f435d088 in /home/saswat/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libtorch.so)

frame #5: <unknown function> + 0x4d6abc (0x7fe43a65eabc in /home/saswat/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libtorch_python.so)

frame #6: <unknown function> + 0x1d3f04 (0x7fe43a35bf04 in /home/saswat/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libtorch_python.so)

<omitting python frames>

frame #18: ../bin/pegasus() [0x402ca1]

frame #19: ../bin/pegasus() [0x403087]

frame #20: <unknown function> + 0x2dfd0 (0x7fe4721e4fd0 in /lib/x86_64-linux-gnu/libc.so.6)

frame #21: __libc_start_main + 0x7d (0x7fe4721e507d in /lib/x86_64-linux-gnu/libc.so.6)

frame #22: ../bin/pegasus() [0x401a9e]

My pytorch version is 1.2.0. I’ve tried with 1.4 and 1.10 but the same error occurs in each case

johndoe · December 11, 2021, 11:28am

Facing a new error now

Start import model ...
I Namespace(config=None, import='pytorch', input_size_list='3,224,224', inputs=None, model='resnet18.pt', output_data='resnet18.data', output_model='resnet18.json', outputs=None, size_with_batch=None, which='import')

I Start importing pytorch...

[6041] Failed to execute script pegasus

Traceback (most recent call last):

  File "pegasus.py", line 131, in <module>

  File "pegasus.py", line 112, in main

  File "acuitylib/app/importer/commands.py", line 286, in execute

  File "acuitylib/vsi_nn.py", line 173, in load_pytorch_by_onnx_backend

  File "torch/jit/__init__.py", line 162, in load

RuntimeError: version_number <= kMaxSupportedFileFormatVersion INTERNAL ASSERT FAILED at /pytorch/caffe2/serialize/inline_container.cc:131, please report a bug to PyTorch. Attempted to read a PyTorch file with version 3, but the maximum supported version for reading is 1. Your PyTorch installation may be too old. (init at /pytorch/caffe2/serialize/inline_container.cc:131)

frame #0: c10::Error::Error(c10::SourceLocation, std::string const&) + 0x33 (0x7f8946b74273 in /home/saswatsubhajyoti_mallick_zeotap_/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libc10.so)

frame #1: caffe2::serialize::PyTorchStreamReader::init() + 0x1e9a (0x7f8949b0300a in /home/saswatsubhajyoti_mallick_zeotap_/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libtorch.so)

frame #2: caffe2::serialize::PyTorchStreamReader::PyTorchStreamReader(std::string const&) + 0x60 (0x7f8949b04270 in /home/saswatsubhajyoti_mallick_zeotap_/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libtorch.so)

frame #3: torch::jit::import_ir_module(std::shared_ptr<torch::jit::script::CompilationUnit>, std::string const&, c10::optional<c10::Device>, std::unordered_map<std::string, std::string, std::hash<std::string>, std::equal_to<std::string>, std::allocator<std::pair<std::string const, std::string> > >&) + 0x38 (0x7f894abe3088 in /home/saswatsubhajyoti_mallick_zeotap_/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libtorch.so)

frame #4: <unknown function> + 0x4d6abc (0x7f8990ee4abc in /home/saswatsubhajyoti_mallick_zeotap_/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libtorch_python.so)

frame #5: <unknown function> + 0x1d3f04 (0x7f8990be1f04 in /home/saswatsubhajyoti_mallick_zeotap_/khadas/aml_npu_sdk/acuity-toolkit/bin/acuitylib/libtorch_python.so)

<omitting python frames>

frame #17: ../bin/pegasus() [0x402ca1]

frame #18: ../bin/pegasus() [0x403087]

frame #19: __libc_start_main + 0xe7 (0x7f89c04e5bf7 in /lib/x86_64-linux-gnu/libc.so.6)

frame #20: ../bin/pegasus() [0x401a9e]

Frank · December 14, 2021, 3:13am

@johndoe Can you share you model to me ? I will try it.

johndoe · December 14, 2021, 3:16am

@Frank Sure. Here it is: resnet18.pt - Google Drive

Frank · December 14, 2021, 3:27am

@johndoe I will test this week

johndoe · December 16, 2021, 7:12am

As a workaround, I’m saving the pretrained models as onnx and then converting them via the ‘convert’ script. The inference results are consistent

Omkar_Shende · January 22, 2022, 3:01pm

there is similar problem with keras pertrained models.
from tensorflow.keras.applications.resnet50 import ResNet50
model = ResNet50(weights=‘imagenet’)
model.save(“resnet50_keras.hdf5”)

0_import

$convert_keras
–keras-model ${NAME}.hdf5
–outputs predictions
–net-output ${NAME}.json
–data-output ${NAME}.data

2022-01-22 20:29:27.585991: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library ‘libcudart.so.10.1’; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /home/omkar/Desktop/new/aml_npu_sdk/acuity-toolkit/bin/acuitylib
2022-01-22 20:29:27.586031: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
I Start importing keras…
D Convert Keras with Keras Branch
Traceback (most recent call last):
File “convertkeras.py”, line 49, in
File “convertkeras.py”, line 39, in main
File “acuitylib/vsi_nn.py”, line 239, in load_keras
File “acuitylib/app/importer/import_keras.py”, line 202, in run
File “acuitylib/app/importer/import_keras.py”, line 154, in convert_engine_keras
File “acuitylib/converter/convert_keras.py”, line 24, in init
File “tensorflow/python/keras/saving/save.py”, line 182, in load_model
File “tensorflow/python/keras/saving/hdf5_format.py”, line 178, in load_model_from_hdf5
File “tensorflow/python/keras/saving/model_config.py”, line 55, in model_from_config
File “tensorflow/python/keras/layers/serialization.py”, line 175, in deserialize
File “tensorflow/python/keras/utils/generic_utils.py”, line 358, in deserialize_keras_object
File “tensorflow/python/keras/engine/functional.py”, line 617, in from_config
File “tensorflow/python/keras/engine/functional.py”, line 1204, in reconstruct_from_config
File “tensorflow/python/keras/engine/functional.py”, line 1186, in process_layer
File “tensorflow/python/keras/layers/serialization.py”, line 175, in deserialize
File “tensorflow/python/keras/utils/generic_utils.py”, line 360, in deserialize_keras_object
File “tensorflow/python/keras/engine/base_layer.py”, line 697, in from_config
File “tensorflow/python/keras/layers/pooling.py”, line 848, in init
File “tensorflow/python/training/tracking/base.py”, line 457, in _method_wrapper
File “/home/omkar/Desktop/new/aml_npu_sdk/acuity-toolkit/bin/acuitylib/tensorflow/python/keras/engine/base_layer_v1.py”, line 165, in init
generic_utils.validate_kwargs(kwargs, allowed_kwargs)
File “tensorflow/python/keras/utils/generic_utils.py”, line 778, in validate_kwargs
TypeError: (‘Keyword argument not understood:’, ‘keepdims’)
[20341] Failed to execute script convertkeras
output :