Skip to content

Replace .norm() with decomposed version for CLIP executorch export #32887

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

Merged
qubvel merged 2 commits into from
Aug 20, 2024
Merged

Replace .norm() with decomposed version for CLIP executorch export #32887

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b1baa0a
Replace .norm() with decomposed version for executorch export
qubvel Aug 19, 2024
07b6438
[run_slow] clip
qubvel Aug 19, 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
15 changes: 13 additions & 2 deletions src/transformers/models/clip/modeling_clip.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,17 @@ def clip_loss(similarity: torch.Tensor) -> torch.Tensor:
return (caption_loss + image_loss) / 2.0


def _get_vector_norm(tensor: torch.Tensor) -> torch.Tensor:
"""
This method is equivalent to tensor.norm(p=2, dim=-1, keepdim=True) and used to make
model `executorch` exportable. See issue https://github.com/pytorch/executorch/issues/3566
"""
square_tensor = torch.pow(tensor, 2)
sum_tensor = torch.sum(square_tensor, dim=-1, keepdim=True)
normed_tensor = torch.pow(sum_tensor, 0.5)
return normed_tensor


@dataclass
class CLIPVisionModelOutput(ModelOutput):
"""
Expand Down Expand Up @@ -1313,8 +1324,8 @@ def forward(
text_embeds = self.text_projection(text_embeds)

# normalized features
image_embeds = image_embeds / image_embeds.norm(p=2, dim=-1, keepdim=True)
text_embeds = text_embeds / text_embeds.norm(p=2, dim=-1, keepdim=True)
image_embeds = image_embeds / _get_vector_norm(image_embeds)
text_embeds = text_embeds / _get_vector_norm(text_embeds)

# cosine similarity as logits
logit_scale = self.logit_scale.exp()
Expand Down