Unverified50% confidenceFactExact time
In the original example, with logits of 4.1, 3.4, and 2.3, raising the temperature from 0.5 to 2 lowers the probability of the highest-scoring token from 78.5% to 46.7%.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
Unverified复现官方基准分数时，温度从0.0调到0.6可能导致基准分数波动5-15个百分点69% similarUnverifiedWith temperature: 0.75 enabled alone, Llama 3.2 1B's pass rate was 19/24 (79%), higher than greedy's 18/24, so temperature is essentially harmless.69% similarUnverifiedIn knowledge distillation, the temperature coefficient T typically ranges from 3 to 20; the higher the temperature, the smoother the probability distribution.66% similarUnverifiedTemperature 0 does not equal deterministic output; three runs of the default model under identical conditions yielded 67.6%, 75.7%, and 83.8% respectively, meaning a single leaderboard figure carries approximately ±8 percentage points of noise.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1067599API
curl https://kongchang.com/api/v1/knowledge/claims/1067599MCP
get_claim(id=1067599)