[BUG] <title>File "torchsparse/nn/functional/conv/func/implicit_gemm.pyx", line 160, in torchsparse.nn.functional.conv.func.implicit_gemm.ImplicitGEMMConvolutionFuntion.backward RuntimeError: shape '[27, 32, 4]' is invalid for input of size 5184

mit-han-lab / torchsparse

[MICRO'23, MLSys'22] TorchSparse: Efficient Training and Inference Framework for Sparse Convolution on GPUs.

https://torchsparse.mit.edu

MIT License

1.16k stars 132 forks source link

[BUG] <title>File "torchsparse/nn/functional/conv/func/implicit_gemm.pyx", line 160, in torchsparse.nn.functional.conv.func.implicit_gemm.ImplicitGEMMConvolutionFuntion.backward RuntimeError: shape '[27, 32, 4]' is invalid for input of size 5184 #243

Closed lihc-cz closed 9 months ago

lihc-cz commented 9 months ago

Is there an existing issue for this?

[X] I have searched the existing issues

Current Behavior

File "torchsparse/nn/functional/conv/func/implicit_gemm.pyx", line 160, in torchsparse.nn.functional.conv.func.implicit_gemm.ImplicitGEMMConvolutionFuntion.backward RuntimeError: shape '[27, 32, 4]' is invalid for input of size 5184

Expected Behavior

No response

Environment

- GCC:
- NVCC:
- PyTorch:
- PyTorch CUDA:
- TorchSparse:

Anything else?

No response

ys-2020 commented 9 months ago

Hi @lihc-cz , could you please provide more details about the error (e.g., a sample code snippet that we could reproduce this bug)? From the information provided, it seems that your (kernel_volume, oc, ic) doesn't match the shape of output weight_gradient.