WIP: (3/7) Judge both A/B orders with orientation-normalized rows - #103
Draft
ErlisLushtaku wants to merge 2 commits into
Draft
WIP: (3/7) Judge both A/B orders with orientation-normalized rows#103ErlisLushtaku wants to merge 2 commits into
ErlisLushtaku wants to merge 2 commits into
Conversation
swap_mode=both runs a second pass with the presented models reversed. winner_llm and pref_llm are inverted back onto the arena's stored A/B identity so overall accuracy and kappa can use both passes. Each row records orientation plus the presented model/completion pair, which is what later ranking metrics need to keep one forward-order battle.
This was referenced Aug 15, 2026
The swap_mode=both artifacts now land in the preset- and swap-keyed result directory, so the test reads them from there.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
WIP: packaged meta-eval split from closed PR #76. Caching is not in this stack; it stays with the unified
do_inferencecache PRs.--judge.swap_mode bothjudges each sampled battle in both orders. The reversed pass is inverted back onto the stored A/B identity. Overall accuracy and kappa use both passes; ranking later uses one forward-order row per battle.This is stacked on #102.