Unverified50% confidenceSolutionExact time
自洽性做法是对同一问题进行多次高温采样（temperature通常设为0.7-1.0），再用多数投票选出频率最高的结果
1
Sources
50%
Confidence
Long-term
Relevance
7/17/2026
First Seen
Sources
Related Claims
Unverified自洽性投票通常使用temperature 0.7-1.0进行多次高温采样，在GSM8K等数学推理基准上能带来高达17.9个百分点的准确率提升79% similarVerified温度（temperature）参数控制采样过程，温度越高输出越随机多样，温度越低越趋向最高概率的选择78% similarUnverifiedSelf-Consistency typically sets the temperature to 0.7 or above to increase sampling diversity.77% similarUnverified自洽性采样的temperature通常设为0.7-1.077% similarUnverifiedWhen testing citation behavior, temperature should be set to 0 or a low value with the top-p parameter fixed, and the same prompt should be run at least 20–50 times to use statistical distributions instead of single results.77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/547614API
curl https://kongchang.com/api/v1/knowledge/claims/547614MCP
get_claim(id=547614)