search

fbelderink / flutter_pytorch_mobile

A flutter plugin for pytorch model inference. Supports image models as well as custom models.
https://pub.dev/packages/pytorch_mobile
Other
98 stars 52 forks source link

Error When Loading the Model #14

Open creatornadiran opened 3 years ago

creatornadiran commented 3 years ago

OS: Windows 11 Package: Anaconda pytorch_mobile: ^0.2.1 PyTorch : 1.8.1

I'm getting this error when trying to load the model:

E/PyTorchMobile( 7949): assets/MobileNetV3_2.pt is not a proper model E/PyTorchMobile( 7949): com.facebook.jni.CppException: [enforce fail at inline_container.cc:222] . file >not found: archive/constants.pkl

All my save types:

  quantized_model = torch.quantization.convert(mobilenet)
  scripted_model = torch.jit.script(quantized_model)
  torch.save(scripted_model.state_dict(), "model_output/MobileNetV3_1.pt")
  torch.jit.save(scripted_model, "model_output/MobileNetV3_2.pt")
  scripted_model.save("model_output/MobileNetV3_3.pt")
  opt_model = optimize_for_mobile(scripted_model)
  opt_model._save_for_lite_interpreter("model_output/MobileNetV3_4.ptl")

I tried all the files and got the same error. Is there a version issue? How can I save my model properly?

palakons commented 1 year ago

did you have the issue resolved?

you could try with the official .pt

If it works, it might be because the .pt file you supplied earlier was in a different (newer) torch version

JenkinsGage commented 1 year ago

I solved it by

    # model is loaded into cpu

    model.eval()
    quantized_model = torch.quantization.convert(model)
    scripted_model = torch.jit.script(quantized_model)
    opt_model = optimize_for_mobile(scripted_model)
    opt_model.save('model.pt')

both load model and predict runs well for me and here is my platform info OS: Windows 10 Device: Android 10 PyTorch: 1.13.0 pytorch_mobile: ^0.2.2