feat: collect multilingual responses for offline comparison - #5
Open
beamivalice wants to merge 1 commit into
Open
beamivalice wants to merge 1 commit into
beamivalice wants to merge 1 commit into
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Add
--language-onlyto collect comparable multilingual model responses for offline review. The suite includes 100 original prompts: eight matched task categories in each of 12 languages, plus four cross-language cases.The mode requires
--modeland a new--saveJSONL path. It records a manifest with the prompt hash and review rubric, complete responses and request settings, errors, and a summary. Defaults are concurrency 4, a 500-token output cap, greedy sampling, and thinking/speculation disabled. Existing logs are never overwritten, and request failures produce a nonzero exit status.The rubric defines five scoring dimensions, equal weighting per language, paired comparison guidance, and separate reporting for cross-language cases, missing answers, and truncation. This is response collection with an offline review protocol; it does not automatically score language quality, upload results, or feed the existing
--comparereport workflow. Endpoints must support the documented request controls.Validation against current upstream main:
npm test: 468 tests passed across 36 files, including prompt coverage, bounded concurrency, complete Unicode responses, error recording, and overwrite protection.npm run typecheck: passed.npx prettier --check . --ignore-unknown: passed.--helpsmoke check: passed.No live model evaluation was run for this PR.