error when infer
If anyone has ever encountered this problem？
[TypeError: all() received an invalid combination of arguments - got (Tensor, keepdim=bool, dim=tuple), but expected one of:
(Tensor input, *, Tensor out)](TypeError: all() received an invalid combination of arguments - got (Tensor, keepdim=bool, dim=tuple), but expected one of:
(Tensor input, *, Tensor out)
didn't match because some of the keywords were incorrect: keepdim, dim
(Tensor input, int dim, bool keepdim, *, Tensor out)
(Tensor input, name dim, bool keepdim, *, Tensor out))
I had this problem, it was due to the torch version. I solved by upgrading to torch 2.4.1
I also solve it by transforming the code into this:
cfg = self.config
v_cfg = self.config.vision_backbone
B, T, N, D = images.shape
# mask = ~torch.all(images.view(B * T, N, D) == -1, dim=(1, 2), keepdim=True)
# Converts the shape of images from (B, T, N, D) to (B * T, N, D)
reshaped_images = images.view(B * T, N, D)
# Creates a Boolean tensor that indicates whether each pixel is equal to -1
equal_to_minus_one = reshaped_images == -1
# The first calculation is along dimension 2
intermediate_mask = torch.all(equal_to_minus_one, dim=2, keepdim=True)
# The second calculation is along dimension 1
all_minus_one = torch.all(intermediate_mask, dim=1, keepdim=True)
# Create a mask that indicates which views contain valid patches
mask = ~all_minus_one
It is not clear whether such code-switching has any effect on accuracy
I recreated this and solved by upgrading to torch 2.4.1. Feel free to reopen this issue if you need further help.