(9/9) Fluency Task Declarations - #93
Open
kargibora wants to merge 2 commits into
Open
Conversation
One fluency task family with 43 language variants (fluency-<language>), loading pinned multilingual-fluency contexts through a registered dataset adapter. Adds the base_completion generation mode for base-model tasks.
Every benchmark now resolves through a packaged task definition, so this deletes the paths that existed only for unpackaged tasks: the fluency special cases in config and the pairwise runner, the monolithic judge-arena-dataset fallback in download_hf, the legacy prompt-preset fallbacks, and dataset_revisions.py (pins live in task YAML).
geoalgo
force-pushed
the
refactor/task-yaml-fluency
branch
from
August 5, 2026 13:32
f7d502b to
b46d5a4
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
This PR packages the fluency benchmark as a declarative task; the last benchmark to migrate. With every dataset now owned by a task definition, it also deletes all the legacy code paths that existed only for unpackaged tasks. This completes the task-YAML stack.
Fluency as a packaged task
fluencytask family with 43 language variants (fluency-french,fluency-mandarin-chinese, …), resolved through the same suffix-variant mechanism as m-ArenaHard.multilingual-fluencydataset source and revision are pinned in the task YAML and loaded through a registeredfluencydataset adapter.base_completiongeneration mode: the task protocol now declares that a base model continues raw text (instead of the runner inferring it from the task name).Legacy removal
Since every task is now packaged, the following are gone (net −260 lines):
config.pyand the pairwise runner — the runner has a single data path via the task's dataset adapter.judge-arena-datasetpattern-matching fallback indownload_hf; downloads go only through task-declared, revision-pinned sources.TASK_DEFAULT_PRESET, thefluency-name check) — presets come from the task YAML.dataset_revisions.pydeleted entirely; every revision pin lives in the task definition that uses it.Notes:
Entire stack is tested with running smoke tests for each of the defined task. No error is raised.