docs(utils): add 'Submitting a new adapter' checklist#205
Closed
borgr wants to merge 1 commit into
Closed
Conversation
Converts openlifescienceai/results (lm-evaluation-harness medical-QA output) into EEE aggregate logs: one EvaluationLog per developer/model, with one EvaluationResult per medical benchmark (MedMCQA, MedQA-USMLE, PubMedQA, and 6 MMLU medical subjects). accuracy -> continuous [0,1], acc_stderr -> uncertainty. - source_type="documentation" + evaluator_relationship=third_party (the leaderboard re-hosts numbers it didn't produce), but eval_library names the unmistakable harness (lm-evaluation-harness). - each result's source_data points at that benchmark's own HF dataset repo. - evaluation_id keyed on the result file's parsed timestamp so re-ingest is idempotent; a stable path-derived fallback (never `now`) when unparseable. - skips hand-curated closed-model baselines (different provenance), hidden dirs, and Jupyter *-checkpoint.json files. Offline test (no optional deps -> runs in both core and full CI): builds logs from synthetic lm-eval result objects and asserts schema-validity, field mapping, per-benchmark dataset repos, and evaluation_id idempotency.
Collaborator
Author
|
Withdrawing this one. On review the per-field self-review rules here are low-level converter-authoring knowledge, not something that belongs in the human-oriented adapter README — and this repo already has docs/eval-converters/, docs/contributing/, and a per-field docs/data-structure/schema.md for public guidance. We'll keep this field-level guidance in our converter's own workflow notes instead. Sorry for the noise; happy to propose a targeted docs/ change if maintainers want one. |
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.
Adds a Submitting a new adapter section to
utils/README.md.Adapter output can be schema-valid and still misrepresent the data — several issues only surface in review, not in
validate. This section gives adapter authors a pre-submit routine that mirrors CI (ruff check,pytest tests,validateon the output, plus an offline shape test usingllm_statsas the reference), and a short self-review list of the mistakesvalidatecan't catch:input.rawmust not contain the model's answer (prompt = user/system turns only);output.rawis the full generation, not the extracted answer;metric_name(the metric) vsevaluation_name(the benchmark); nometric_type;eval_libraryis the harness, not the platform;source_typeevaluation_runvsdocumentation;continuousscore_type needsmin_score/max_score(no unbounded;infserializes to null);additional_detailsvalues must be strings;model_info.idcarries no effort/mode/quant suffix.Docs only; no code changes.