Meta-eval: (2/7) Score sampled battles with accuracy and Cohen's kappa - #102
Open
ErlisLushtaku wants to merge 2 commits into
Open
Meta-eval: (2/7) Score sampled battles with accuracy and Cohen's kappa#102ErlisLushtaku wants to merge 2 commits into
ErlisLushtaku wants to merge 2 commits into
Conversation
The meta-eval runner now judges the stored A/B order with the task's default PairScore prompt (temperature 0.5) and writes annotations.parquet alongside the sample. Agreement is reported on all battles and on the subset that drops human ties, with bootstrap standard errors. Judge inputs are serialized to strings before they are stored, so later cache backends do not have to bind a LangChain prompt object.
This was referenced Aug 15, 2026
Two runs of the same task and judge under different prompt presets or swap modes wrote to the same result directory, so the second silently overwrote the first. Both now appear in the folder name, and the report records the preset the verdicts were parsed under. The tie boundary also moves back to the original one: a preference exactly eps above 0.5 is a tie, not a model_a win.
ErlisLushtaku
commented
Aug 17, 2026
|
|
||
| TIE_EPSILON = 0.01 | ||
| # Paper PairScore temperature for meta-eval; generate+judge keeps 0.3. | ||
| META_EVAL_PAIRSCORE_TEMPERATURE = 0.5 |
Collaborator
Author
There was a problem hiding this comment.
q: should we align it with the other cases where we use 0.3?
ErlisLushtaku
changed the title
ErlisLushtaku
marked this pull request as ready for review
August 17, 2026 09:21
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
Note
Packaged meta-eval split from closed PR #76. Meta-eval does not use the unified
do_inferencecache, and the cache stack (#94-#100) does not wire it up either: caching will be added to meta-eval once those PRs are merged.Judges the sampled battles in stored A/B order and reports accuracy and Cohen's kappa with bootstrap SE. Agreement is written for all battles and for the no-human-tie subset.
PairScore temperature is 0.5 here (paper / Figure 4); generate+judge stays 0.3.
This is stacked on #101.