Unverified50% confidenceFactExact time
temperature 为 0 时退化为贪婪解码(Greedy Decoding),每步选取概率最高的 Token
1
Sources
50%
Confidence
Long-term
Relevance
7/8/2026
First Seen
Sources
并非一切都值得消耗Token:何时用AI,何时用确定性代码
hackernewshackernews7/6/2026
Related Claims
Verifiedtemperature为0时退化为贪婪解码,每步选取概率最高的Token,但仅能保证单次模型版本内的确定性85% similarVerifiedtemperature参数对Logits向量进行缩放,temperature为0时退化为贪婪解码76% similarUnverifiedDoom Loop在温度参数较低的采样策略下更容易出现,温度趋近0时模型几乎总选择概率最高的token(贪心解码)70% similarUnverified温度参数通过对 softmax 层的 logits 进行缩放来控制概率分布的尖锐程度,温度趋近于0时接近贪心解码70% similarUnverified温度是一个0到2之间的超参数,控制回答的随机性与创造性,低温(0.1-0.3)适合编程数学客服,高温(0.8-1.5)适合头脑风暴写作67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/252244API
curl https://kongchang.com/api/v1/knowledge/claims/252244MCP
get_claim(id=252244)