If I modify the network structure of the target detection model, can I load my own model?

[Xhran]

I have modified a custom target detection network and trained it to get pytorch weights, can I load this model for inference using DJL? How should I do it? Thank you！

[zachgk]

Yeah. You can follow the instructions to export your model as torchscript and then load it following this example notebook. As part of this, you may need to implement a Translator class with the pre-processing and post-processing that your model uses if it doesn't use exactly the same pre/post-processing as a Translator that we already have implemented

[Xhran]

Yeah. You can follow the instructions to export your model as torchscript and then load it following this example notebook. As part of this, you may need to implement a Translator class with the pre-processing and post-processing that your model uses if it doesn't use exactly the same pre/post-processing as a Translator that we already have implemented

When I load the exported torchscript model file, I get the following error. What is the problem? Thank you!

ai.djl.translate.TranslateException: ai.djl.engine.EngineException: The following operation failed in the TorchScript interpreter. Traceback of TorchScript, serialized code (most recent call last): File "code/torch/models/yolo.py", line 71, in forward _35 = (_20).forward(_34, ) _36 = (_22).forward((_21).forward(_35, ), _29, ) _37 = (_24).forward(_33, _35, (_23).forward(_36, ), )

    _38, _39, _40, _41, = _37
    return (_41, [_38, _39, _40])
  File "code/__torch__/models/yolo.py", line 102, in forward
    y = torch.sigmoid(_22)
    _23 = torch.mul(torch.slice(y, 4, 0, 2), CONSTANTS.c0)
    _24 = torch.add(torch.sub(_23, CONSTANTS.c1), CONSTANTS.c2)
          ~~~~~~~~~ <--- HERE
    xy = torch.mul(_24, torch.select(CONSTANTS.c3, 0, 0))
    _25 = torch.mul(torch.slice(y, 4, 2, 4), CONSTANTS.c4)

Traceback of TorchScript, original code (most recent call last):
/content/drive/My Drive/pytorch/yolov5-6.0/models/yolo.py(68): forward
/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py(1090): _slow_forward
/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py(1102): _call_impl
/content/drive/My Drive/pytorch/yolov5-6.0/models/yolo.py(149): _forward_once
/content/drive/My Drive/pytorch/yolov5-6.0/models/yolo.py(126): forward
/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py(1090): _slow_forward
/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py(1102): _call_impl
/usr/local/lib/python3.7/dist-packages/torch/jit/_trace.py(965): trace_module
/usr/local/lib/python3.7/dist-packages/torch/jit/_trace.py(750): trace
export.py(56): export_torchscript
export.py(304): run
/usr/local/lib/python3.7/dist-packages/torch/autograd/grad_mode.py(28): decorate_context
export.py(359): main
export.py(364): <module>
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

[lanking520]

@Xhran your problem can be solved by adding this to the Criteria:

.optOption("mapLocation", "true")

Sample: https://docs.djl.ai/master/docs/demos/jupyter/load_pytorch_model.html#step-3-load-your-model