Verified75% confidenceFactTime unknown
推理时计算扩展（Test-time Compute Scaling）通过在推理阶段投入更多计算资源来提升模型性能，研究表明在某些任务上比单纯增大模型参数更加高效
5
Sources
75%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
GitHub 8K星：最全LLM学习资源库深度解析
githubWangRongsheng
Related Entities
Related Claims
Verified推理时扩展（inference-time scaling）通过在推理阶段增加计算量提升模型性能，存在类似预训练的幂律关系84% similarUnverified推理时计算在模型参数固定的前提下，通过在推理阶段投入更多计算资源来改善输出质量80% similarUnverified推理时扩展在推理阶段增加计算量（如beam search、Best-of-N采样、蒙特卡洛树搜索），模型性能可随计算量增加持续提升并存在幂律关系79% similarUnverified推理时计算扩展通过延长思维链、多路径采样择优、自我验证反思等方式在推理阶段投入更多计算可提升复杂推理任务性能79% similarPartially VerifiedLarger model parameters bring stronger reasoning capabilities but also higher computational costs and longer response latency.79% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/33727API
curl https://kongchang.com/api/v1/knowledge/claims/33727MCP
get_claim(id=33727)