ViperEkura 2565755e45 refactor: switch eval datasets to HuggingFace source ...

- Replace GitHub/berkeley direct downloads with HF datasets API
- MMLU: cais/mmlu (all config), map val->validation split, write per-subject CSV
- HumanEval: openai/openai_humaneval
- IFEval: google/IFEval
- Enables HF_ENDPOINT mirror for faster downloads in CN

2026-07-18 00:09:31 +08:00

..

analyze_weights.py

fix: eval script bugs and add missing features

2026-07-17 14:02:58 +08:00

evaluate_humaneval.py

refactor: switch eval datasets to HuggingFace source

2026-07-18 00:09:31 +08:00

evaluate_ifd.py

fix: eval script bugs and add missing features

2026-07-17 14:02:58 +08:00

evaluate_ifeval.py

refactor: switch eval datasets to HuggingFace source

2026-07-18 00:09:31 +08:00

evaluate_mmlu.py

refactor: switch eval datasets to HuggingFace source

2026-07-18 00:09:31 +08:00

evaluate_ppl.py

fix: eval script bugs and add missing features

2026-07-17 14:02:58 +08:00

evaluate_rouge.py

feat : add ROUGE evaluation script (manual impl, no deps)

2026-07-05 01:15:01 +08:00