Vectorize promo logit computation#2
Open
mcognetta wants to merge 3 commits into
Open
Conversation
mcognetta
commented
May 25, 2026
| bias = promo_biases[:, to_file, piece_idx] # (B,) | ||
| promotion_logits.append((base_score + bias).unsqueeze(1)) | ||
| promotion_logits = torch.cat(promotion_logits, dim=1) # (B, 256) | ||
| base = scores_base[:, self.rank7_indices][:, :, self.rank8_indices] # (B, 8, 8) |
Author
There was a problem hiding this comment.
The rank_7/rank_8 could also just be replaced directly with a slice like:
base = scores_base[:, 48:56][:, :, 56:65].
This avoids a new allocation, etc. since it is just a contiguous view. Doesn't give much speedup on my machine though, so maybe not worth the loss in clarity.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This replaces the nested loop in the promotion logit computation with a vectorized version that gives an ~18% speedup for the entire forward pass on a small local benchmark on my machine.
It also moves the rank7_indices and rank8_indices to the model definition, since these are static and don't need to be constructed each forward pass.