Repository navigation
WIP: Add a tune-judge task that searches judge configs with neps - #133
Draft
ErlisLushtaku wants to merge 5 commits into
Draft
ErlisLushtaku wants to merge 5 commits into
ErlisLushtaku wants to merge 5 commits into
Conversation
Judge tuning needs to select configurations on one partition of the human-labeled arena and report on another. Battles are assigned by a salted hash of their first user prompt, so repeated prompts never cross partitions and changing run.seed cannot move battles between them.
A tune_judge section on a meta-eval task expands a grid of judge overrides and runs each configuration as its own meta-eval subprocess on the validation split. After each rung, survivors are chosen by the non-dominated sort on judge cost and human agreement from Salinas et al. (ICML 2025). The best configuration per judge model is then scored once on the held-out test split.
tune-judge-* tasks reuse the meta-eval arena definitions and run each trial as the matching meta-eval task, so both share one judgement cache. The custom successive halving and Pareto selection give way to neps (priorband by default, with the base config as prior; hyperband and mo_hyperband), run via AskAndTell with validation battles per model as the fidelity.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
WIP. Draft work in progress, not ready for review.
Description
Adds a separate
tune-judge-*task that searches judge settings on the same human battles as meta-eval.meta-eval-*task on the validation split, so tuning and meta-eval share the judgement cache.--run.seeddoes not move battles between them.