HF open-sources Ettin rerankers: six sizes, SOTA, full training recipe
antoine_chaffin · x · 2026-11-13
Hugging Face researcher Tom Aarsen released six new Sentence Transformers CrossEncoder rerankers (17m to 1b), SOTA at each size, built on Ettin ModernBERT encoders, with training data and the full recipe open-sourced.
- Trained via pointwise MSE distillation from mixedbread mxbai-rerank-large-v2 scores on lightonai data subsets.
- Evaluated with google/embeddinggemma-300m and other embedders on MTEB(eng, v2) Retrieval.
- The recipe was bootstrapped with the new train-sentence-transformers Agent Skill in Sentence Transformers v5.5.0, letting coding agents fine-tune models on your data.
LightOn's Antoine Chaffin amplified the release: if you're still training on MSMARCO only, this open data now lets your ideas create SOTA for free — "open source creates a race to the top."
More from Models
- Leaderboards miss vision skills: why DeepSeek's chart-reading progress doesn't show up in rankings — TwoBlueFools · 2026-11-13
- OpenAI's 1147-page Yang-Mills proof faces its first expert assessment: not proven yet — mishig25 · 2026-11-13
- Terminal-Bench 4.0 Cost Math: GPT-6.1 Sol Matches Opus 5.5 at ~1/8 the Price — LeadershipNo2904 · 2026-11-13
- HF staffer spotlights basal-2.0-mini, a 1.5B decision-tuned model from Bielik — lhoestq · 2026-11-13
- Observer surprised OpenAI shipped no answer to rival's new model release — jdjohnson · 2026-11-13
- Google's Carbon-Argon-Hydrogen model naming breaks its own size logic — JeremyNguyenPhD · 2026-11-13