brainstorming
brainstorming
You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).
Explore skills
Development · SKILL.md
mmlu_recovery_harness
mmlu_recovery_harnessRun a bounded MMLU proxy recovery using generated item, prompting, and scoring skills.
Independent GitHub snapshot · Automated classification · Original documentation governs
Skill documentation
GitHubname: mmlu_recovery_harness description: Run a bounded MMLU proxy recovery using generated item, prompting, and scoring skills.
mmlu_recovery_harness
Use this skill when a recovery or evaluation task needs the MMLU mechanism represented by this module. Do not use it as a substitute for a full model evaluation unless the run explicitly declares a reduced proxy target.
Inputs
Inputs are JSON-compatible dictionaries following the module document contract. Required fields are validated by the script in scripts/.
Outputs
Outputs are JSON-compatible dictionaries with normalized labels, prompts, scores, or recovery artifacts. Downstream modules should consume these outputs instead of parsing free-form text.
Workflow
- Read the paper profile and module plan to confirm the selected target.
- Call the script or import the helper functions from
scripts/. - Preserve the boundary between item construction, prompt construction, scoring, and recovery logging.
- Run the validation command after changes:
python <distiller>/module-to-skill/scripts/validate_skill_tree.py <skill_dir> --run-tests.
Limitations
This skill captures the evaluation mechanism, not the full hosted GPT-3 runtime or the full private benchmark distribution. Soft-mode recovery must label reduced results as proxy evidence.