Popular repositories Loading

1. mimir-korean-retrieval mimir-korean-retrieval Public

   Pre-registered, reproducible measurement of BM25 / dense / hybrid retrieval on public Korean benchmarks (MTEB-ko). The harness is validated against published baselines before any new number is repo…

   Python

2. themis-judge-reliability themis-judge-reliability Public

   Do LLM judges and reward models still agree with themselves when the same question is asked with the predicate inverted? Pre-registered measurement on RewardBench 2, run on one 24 GB GPU with open …

   Python

3. reward-bench reward-bench Public

   Forked from allenai/reward-bench

   RewardBench: the first evaluation tool for reward models.

   Python

4. mteb mteb Public

   Forked from embeddings-benchmark/mteb

   MTEB: State-of-the-art evaluation of embeddings across languages and modalities

   Python