Medical retrieval: Fine-tuned Late Interaction hits 91.4 NDCG@10
tomaarsen · x · 2026-08-26
Evaluating 47 model configurations on 1,000 medical questions against 200k passages, the fine-tuned Late Interaction model achieved 91.4 NDCG@10, beating the best zero-shot model by 6.2 points.
Rank-1 accuracy improved to 84.9%, removing a third of the remaining error. Visual results confirm multi-vector/late-interaction dominance.
Related event: Fine-tuning ColBERT on a single RTX 3090 beats general-purpose retrievers(14 posts)→
More from Research
- OSVerify: LLM verification framework finds 18 issues across 11 OpenAI math preprints — james_y_zou · 2026-10-16
- Intelligence is proportional to data: reorder tasks by data volume and the jagged frontier becomes a spiral — sebkrier · 2026-10-16
- Doc2Harness: New benchmark finds best coding agent passes only 53% across 329 cases — _akhaliq · 2026-10-16
- Stanford's Fei-Fei Li team open-sources OpenWAM, a unified world-action model framework for robots — jiqizhixin · 2026-10-16
- NVIDIA's TacitVLA: robots reason without reading their own thoughts, 3.4x lower latency — arankomatsuzaki · 2026-10-16
- Cohere Labs world model course session 4: DreamerV1-V4 architectures in one go — Cohere_Labs · 2026-10-16