Jev ニュース

用 SGLang 扩展类 JEV 决策模型:Score API 与多候选评分基准

LMSYS:Blog(Chatbot Arena 团队)AIHOT スコア 46/100

要約

SGLang 团队介绍如何用 /v1/score 接口服务类 JEV 决策模型,调用方通过 label_token_ids 显式指定标签 token 分数,而非依赖生成响应的 top-k logprobs。

元の記事の見出し

Blog Scaling JEV-like Decision Models with SGLang A customer asks whether an order has shipped. An agent already has the order ID and three possible next actions: query the order-status service, search general delivery-policy documentation, or ask th… Sundara Raman Ramachandran, Chuanrui Zhu, Qing Lan, Shri Rajamanikandan Vasudevan, Jian Sheng, I-Ting Chen, Fedor Borisyuk