Fix batch embedding averaging for batch_size > 1#3839
Open
Chessing234 wants to merge 2 commits intolm-sys:mainfrom
Open
Fix batch embedding averaging for batch_size > 1#3839Chessing234 wants to merge 2 commits intolm-sys:mainfrom
Chessing234 wants to merge 2 commits intolm-sys:mainfrom
Conversation
Initialize x before the loop to prevent UnboundLocalError if generate_stream_gate yields no items. Fixes lm-sys#3786 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Compute per-sequence token counts instead of a single scalar across the entire batch. This fixes incorrect embeddings when batch_size > 1. Fixes lm-sys#3785 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
3 tasks
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
batch_size > 1inModelWorker.get_embeddings()token_numwas computed as a single scalar summing tokens across the entire batch (torch.sum(attention_mask).item()), butsum_embeddingsis per-sequence (shape[batch_size, hidden_dim]). Dividing a per-sequence tensor by a batch-wide scalar produces wrong averages for every sequence except whenbatch_size == 1.attention_mask.sum(dim=1, keepdim=True)so each sequence's embedding is divided by its own token count. Theret["token_num"]return value remains a scalar (total tokens) for API compatibility.Fixes #3785
Test plan
batch_size=1(no regression)batch_size=1vsbatch_size>1for the same inputs -- they should now matchembed_in_truncateand chunked (non-truncate) code pathsuse_cls_poolingenabled and disabled🤖 Generated with Claude Code