T-Search, 복잡한 다단계 검색을 위한 오픈 에이전트형 리트리버 공개
T-Search: An Open Agentic Retriever and Playground for Hard Multi-Step Search
- 매체
- Takara TLDR
- 게시
- 중요도
- 3/5
- 화제 지수
- —
한국어 요약
Qwen 3.6-35B 기반의 T-Search는 고난도 다단계 검색을 위한 오픈웨이트 에이전트형 리트리버다. 7개 영어·러시아어 벤치마크에서 단일 롤아웃 기준 Recall@10 56.0을 달성했으며, 모델과 하니스, 라이브 데모, 벤치마크를 공개했다.
원문 발췌 · English
We present T-Search, an open-weight agentic retriever for hard multi-step search. Given a question and a search tool over a fixed corpus, it runs a bounded multi-round search and returns a ranked list of evidence chunks with short justifications, leaving answer generation to a downstream model, so backend and generator can be swapped without retraining. T-Search is built on Qwen3.6-35B-A3B and trained on adversarially filtered synthetic search tasks with round-sliced supervised fine-tuning followed by GSPO on a recall reward. Averaged over seven English and Russian benchmarks with gold evidence annotations, it reaches 56.0 Recall@10 with one rollout, 14.4 points above its base, and 61.3 with three fused rollouts, outperforming larger open models. We release the model, harness, live demo, and three benchmarks, including TRuST, the first native-Russian hard-search benchmark.