Unverified50% confidenceFactExact time
Kirchenbauer方案在模型原始logits上为绿名单token加上一个通常取1-2之间的固定偏置δ，再进行softmax归一化后采样
1
Sources
50%
Confidence
Long-term
Relevance
8/23/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedKirchenbauer等人的水印算法在生成每个token时，根据前文的哈希值将词汇表划分为绿名单和红名单，并轻微提升绿名单token被选中的概率68% similarUnverifiedLLL算法主循环包含尺寸约简和Lovász条件交换两个核心操作，尺寸约简使Gram-Schmidt投影系数μᵢⱼ绝对值不超过1/259% similarUnverifiedThe model files used by Ollama are in GGUF format; the K_M suffix in the naming indicates the use of a grouped quantization strategy, which typically preserves better model accuracy at the same bit count.57% similarVerifiedFrom a Bayesian perspective, L1 regularization corresponds to a Laplace prior on the parameters, and L2 regularization corresponds to a Gaussian prior.56% similarUnverifiedWisprGemma 的不同功能模式本质上是给同一个模型换不同的提示词54% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/788622API
curl https://kongchang.com/api/v1/knowledge/claims/788622MCP
get_claim(id=788622)