Pytorch, Cuda and cuDNN version ?

[ghost]

Nice work here ...

Can I ask what's the version Pytorch, Cuda and cuDNN version you're using ?

I have faced following issues and was wondering if is it due to compatibility issues ?

I am testing in Pytorch 1.10. running on Cuda 10.2 cuDNN 7. No issue faced running with Python.

erminate called after throwing an instance of 'std::runtime_error'
  what():  The following operation failed in the TorchScript interpreter.
Traceback of TorchScript, serialized code (most recent call last):
  File "code/__torch__/torchvision/models/resnet/___torch_mangle_25.py", line 6, in forward
  def forward(self: __torch__.torchvision.models.resnet.___torch_mangle_25.ResNet,
    x: Tensor) -> Tensor:
    _0 = torch.cudnn_convolution_relu(x, CONSTANTS.c0, CONSTANTS.c1, [2, 2], [3, 3], [1, 1], 1)
         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
    x0 = torch.max_pool2d(_0, [3, 3], [2, 2], [1, 1])
    _1 = torch.cudnn_convolution_relu(x0, CONSTANTS.c2, CONSTANTS.c3, [1, 1], [1, 1], [1, 1], 1)

Traceback of TorchScript, original code (most recent call last):

    graph(%input, %weight, %bias, %stride:int[], %padding:int[], %dilation:int[], %groups:int):
        %res = aten::cudnn_convolution_relu(%input, %weight, %bias, %stride, %padding, %dilation, %groups)
               ~~~~ <--- HERE
        return (%res)
RuntimeError: cuDNN filters (a.k.a. weights) must be contiguous in desired memory_format

Thanks.

[viig99]

I am using a built version Pytorch master- 1.10, cuda 11.5 and cuddn 8.3.1

[viig99]

The model_resources folder have the traced modules based on my machine configuration, you can trace one based on yours using python3 optimize_model_for_inference.py will add this to the README.md.