docs(eval): record full-passage rerank re-measure (#236 confirmed)#237
Merged
Conversation
The #236 fix (rerank scores the full block body) flips the SciFact rerank result from a regression to an improvement. Re-measured on a 400-doc / 50-query mini split with full passages: rerank vs single_pass nDCG@10: snippet -0.175 -> full-passage +0.018 rerank vs single_pass MRR: snippet -0.19 -> full-passage +0.031 Quality is fixed (the cross-encoder now improves ranking). Latency is the remaining blocker — full passages are longer sequences, so rerank is ~63 s/query on CPU even at 50 candidates; it needs GPU + a small candidate pool. Adds the re-measure JSON + an Update section to baseline-scifact.md. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Follow-up to #236. The eval harness caught that the rerank stage regressed quality (it scored the 200-char snippet); #236 fixed it to score the full block body. This records the empirical confirmation in
docs/eval/baseline-scifact.md.Re-measured on a 400-doc / 50-query SciFact mini split with full passages:
The cross-encoder now improves ranking, as intended. Latency is the remaining blocker: full passages are longer token sequences, so rerank is ~63 s/query on CPU even at 50 candidates — it needs GPU and/or a small
rerank_candidates. Quality fixed; latency/default-on is the next lever.Docs + JSON only.
🤖 Generated with Claude Code
Need help on this PR? Tag
/codesmithwith what you need. Autofix is disabled.