Unverified50% confidenceSolutionExact time
In reference-free DPO, since the base model is already frozen, its output logits can serve as the reference distribution, and the KL anchoring term directly constrains the corrected logit distribution from deviating excessively from the frozen base model's original output.
1
Sources
50%
Confidence
Long-term
Relevance
9/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLogits are the raw unnormalized scores output by the model's final layer; after softmax they become the probability distribution over each token, and they are the most observable quantity closest to "model confidence."61% similarUnverifiedLLM在推理阶段参数是冻结的，模型不会因用户交互而更新权重，因此本质上是无状态的58% similarUnverifiedLLM 的自回归生成特性决定了输出是逐 Token 的流式传输，在前端产生流式接收、增量 DOM 更新、不完整 Markdown 容错解析等工程需求57% similarUnverified当前的LLM一旦完成训练其参数即被冻结，只能通过微调或RAG等外部机制来更新57% similarUnverifiedlogit函数定义为log(p/(1-p))，将概率值从[0,1]映射到(-∞,+∞)，是广义线性模型中处理二分类因变量的核心技术56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/924667API
curl https://kongchang.com/api/v1/knowledge/claims/924667MCP
get_claim(id=924667)