Unverified50% confidenceCautionExact time
In multi-turn autonomous tasks, numerical non-determinism is systematically amplified, so one should not blindly trust the reproducibility of temperature=0.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTemperature=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80% similarUnverifiedUnder temperature=0 (greedy decoding) conditions, multiple runs of a coding agent on the same SWE-bench task still lead to different final states, meaning full reproducibility cannot be achieved.75% similarVerifiedEven 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.74% similarUnverifiedAgent behavior is non-deterministic; even using the same prompt and temperature=0 settings, the same model's outputs at different points in time may exhibit subtle differences, influenced by factors such as floating-point arithmetic precision, batching strategies, and model version updates.72% similarUnverifiedEven when temperature is set to 0, some models still exhibit slight output fluctuations due to floating-point errors in parallel computation.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/901349API
curl https://kongchang.com/api/v1/knowledge/claims/901349MCP
get_claim(id=901349)