search

OpenGVLab / VideoMamba

VideoMamba: State Space Model for Efficient Video Understanding
https://arxiv.org/abs/2403.06977
Apache License 2.0
660 stars 47 forks source link

File "/xxx/selective_scan_interface.py", line 331, in backward dx, dconv1d_weight, dconv1d_bias = causal_conv1d_cuda.causal_conv1d_bwd( RuntimeError: weight must have shape (dim, width) #38

Open zyzcuczyu opened 2 months ago

zyzcuczyu commented 2 months ago

This is the line where the error code is located：

    dx, dconv1d_weight, dconv1d_bias = causal_conv1d_cuda.causal_conv1d_bwd(
        x, conv1d_weight, conv1d_bias, dconv1d_out, dx, None, None, None, True, True
    )

Before that, I print the shape of conv1d_weight: shape: torch.Size([1536, 4]) is_contiguous: True

It seems that the shape of conv1d_weight is right, but the code still goes wrong. I really don't know why. The image shows the details. image_1373 btw, these are versions of my packages: image

Hope someone can solve my problem :(

Andy1621 commented 2 months ago

Hi! I think you can open an issue in the author's repo.

zyzcuczyu commented 2 months ago

Hi! I think you can open an issue in the author's repo.

thanks, I will do that.