Thought-Level Beam Search for Reasoning

  • 类型:arxiv
  • 标识:2608.08020
  • 链接:https://arxiv.org/abs/2608.08020
  • 主分类:llm-infra
  • 形态:method
  • 被引:0
  • 被引来源:Semantic Scholar
  • S2被引:0
  • 影响力被引:0
  • TLDR:By periodically pruning unpromising trajectories and immediately branching from high-quality prefixes, Gambit dynamically concentrates compute onto the most promising reasoning traces via a light-weight scorer probing hidden states while maintaining continuous high hardware utilization.
  • 待LLM分类:否
  • 标题中文:用于推理的思维级束搜索
  • TLDR中文:Gambit 通过周期性剪除低潜力轨迹并立即从高质量前缀分支,利用轻量 scorer 探测 hidden state,将算力动态集中到最有潜力的推理路径上,同时保持持续的高硬件利用率。
  • 成熟度:research
  • 场景:测试时计算、束搜索、推理
  • 来源文件
  • /inbox/tom/_candidates/2026-08-15-agent-rag-longcontext-candidates.json
  • /inbox/tom/_candidates/2026-08-16-agent-rag-longcontext-candidates.json
  • /inbox/tom/_candidates/2026-08-17-agent-rag-longcontext-candidates.json
  • /inbox/tom/_candidates/2026-08-17-agent-memory-tool-use-candidates.json
  • [S2 enrich]