Skip to content

[Feature] remove an 'assert' statement in favor of a 'raise' statement to make MultiScaleDeformableAttention compatible with torch.compile() #3126

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
isaacfinberg wants to merge 2 commits into
base: main
Choose a base branch
from
Open

[Feature] remove an 'assert' statement in favor of a 'raise' statement to make MultiScaleDeformableAttention compatible with torch.compile() #3126

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7005676
switch assert for raise
isaacfinberg Jun 7, 2024
7ee1f9d
edit error msg
isaacfinberg Jun 7, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 4 additions & 1 deletion mmcv/ops/multi_scale_deform_attn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -332,7 +332,10 @@ def forward(self,

bs, num_query, _ = query.shape
bs, num_value, _ = value.shape
assert (spatial_shapes[:, 0] * spatial_shapes[:, 1]).sum() == num_value
if (spatial_shapes[:, 0] * spatial_shapes[:, 1]).sum() != num_value:
raise ValueError('The sequence length of `value` must be equal to'
'size of the flattened features maps summed over'
'all levels.')

value = self.value_proj(value)
if key_padding_mask is not None:
Expand Down