Unverified50% confidenceFactExact time
Even when temperature is set to 0, different conversation histories and system prompts may cause minute shifts in the attention weights of the Transformer self-attention mechanism, thereby producing different numerical outputs.
1
Sources
50%
Confidence
Long-term
Relevance
9/8/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedEven with temperature set to 0 (greedy decoding), the behavior of large language models is still influenced by factors such as context length and positional bias in the attention mechanism.77% similarVerifiedThe attention mechanism of the Transformer architecture significantly reduces its focus on early content when processing very long sequences, causing the model to forget earlier settings76% similarUnverifiedAfter a model version update, its attention allocation mechanism, temperature sensitivity, and instruction parsing method may change, causing existing prompt strategies to become ineffective.70% similarUnverifiedDuring Transformer model inference, the KV Cache (key-value cache) expands continuously as the conversation grows, causing the attention score distribution to become increasingly flat and making it difficult for the model to precisely focus on information relevant to the current task.68% similarUnverifiedSetting the temperature parameter too high in decoding strategies leads to increased sampling randomness, which is associated with hallucination generation68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/875589API
curl https://kongchang.com/api/v1/knowledge/claims/875589MCP
get_claim(id=875589)