EEE Datastore
- Type
- dataset
- Venue
- Hugging Face / evaleval
- Year
- 2026
- Source
- huggingface
- Access
- free
- Language
- en
- Added
- 2026-08-14T18:28:00Z
- Verified
- 2026-08-14T18:28:00Z
Summary
evaleval/EEE_datastore stores per-run aggregate JSON plus optional instance-level JSONL. Legacy tree is data/<benchmark>/<developer>/<model>/; a generated flat/ UUID object store adds latest_manifest.json, checksummed entries.jsonl, and by_collection indexes (USAGE_EEE_datastore.md). Schema 0.2.2 records include source_metadata (e.g. inspect_ai / Arcadia Impact), model_info, evaluation_results, and hashed companion samples. 106 load_dataset configs (GAIA, IFEval, MMLU-Pro, SWE-Bench, TerminalBench, arena, ...). MIT. Eval-trace warehouse, not a pretraining corpus.
Keywords
evaleval · eee-datastore · evaluation · leaderboard · inspect-ai · gaia · mmlu-pro · huggingface
Topics
LLM evaluation, leaderboard traces, eval-result aggregation
Research notes
- Primary: HF dataset card (license:mit, modality:text, size_categories 1K<n<10K, downloads=11148 likes=39 checked 2026-08-14). USAGE_EEE_datastore.md is the layout doc; README is mostly YAML configs plus one-line description. Discord #random-papers posted the HF dataset pointer. Substantial hosted eval-trace corpus — see datasets_local.