MPE: multi-prefix embeddings with chunked MaxSim retrieval - #229
Open
FarmersWrap wants to merge 61 commits into
Open
MPE: multi-prefix embeddings with chunked MaxSim retrieval#229FarmersWrap wants to merge 61 commits into
FarmersWrap wants to merge 61 commits into
Conversation
added 30 commits
December 10, 2025 03:36
The plaintext queries must never be committed; regenerate locally with examples/BrowseComp-Plus/decrypt_dataset.py --generate-tsv. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Closed
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Adds multi-prefix embedding (MPE) support to the retriever: passages can be chunked (fixed size, random size range, independently encoded chunks, or pre-chunked input), and query–document scores are computed via MaxSim over chunk embeddings. Scripts under
examples/mpe/cover data preparation, training, and evaluation on MLDR-en, BrowseComp-Plus, and LongEmbed.Relation to previous PRs
Supersedes #215 (and #211 before it). This is a fresh branch whose history never contained the decrypted BrowseComp-Plus
topics-qrels/queries.tsvflagged in the #215 review — the file was purged via history rewrite, and a.gitignoreentry now prevents it from being re-committed (the README documents regenerating it locally viadecrypt_dataset.py --generate-tsv).Status
The remaining review feedback from #215 is being addressed on this PR:
DenseModelsubclass so SPLADE/GradCache/subclass paths are untouched)maintrust_remote_code, removerich/debug leftoversmain(resolving thecollator.pyconflicts)🤖 Generated with Claude Code