Unverified90% confidenceFactTime unknown
Temperature=0（贪心采样）也不能保证模型输出的确定性，至少有三个原因：浮点运算非结合性、MoE路由不确定性、服务端配置漂移
1
Sources
90%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
AI Agent模型分层路由策略：Token成本、幻觉应对与选型实战
bilibili前端小卒
Related Entities
Related Claims
VerifiedEven when the temperature is set to 0, some inference frameworks still have non-determinism in underlying floating-point computations, causing output differences across hardware and across batches.79% similarUnverifiedEven when temperature is set to 0, some models still exhibit slight output fluctuations due to floating-point errors in parallel computation.77% similarVerifiedThe non-determinism of large language models stems from their autoregressive sampling process; even with temperature set to 0, differences in inference frameworks, batch sizes, and hardware floating-point operations can all cause output drift.77% similarUnverified加载MTP层后额外浮点运算会导致数值产生微小变化，可能导致模型走不同推理路径产生不同结果，即使温度设为0也会出现71% similarUnverifiedAt Temperature=0, greedy decoding is used, but due to the non-determinism of GPU floating-point operations (FP16/BF16 precision), completely consistent output still cannot be guaranteed in actual engineering.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/27306API
curl https://kongchang.com/api/v1/knowledge/claims/27306MCP
get_claim(id=27306)