Unverified50% confidenceFactExact time
In hard mode, the token logit matching a violating phrase is set to -inf, so the model cannot generate any part of that phrase at all, and if part has already been generated, it is forcibly ended with EOS.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedConstrained decoding allows only tokens conforming to grammar rules to be sampled by setting the logits of illegal tokens to negative infinity at each decoding step; FSM is applicable to regular languages, while CFG can handle nested structures such as JSON and XML.69% similarUnverifiedGBNF 通过在每个 Token 生成步骤对 logits 进行掩码过滤，强制模型只能生成符合预定义语法规则的 Token 序列66% similarUnverifiedOutlines pre-compiles constraints into an FSM before inference and performs an offline mapping between the tokenizer vocabulary and each state of the FSM; during inference, hard constraints are enforced by setting the logits of illegal tokens to negative infinity (logits masking).63% similarUnverifiedOutlines constructs a Finite State Machine (FSM) that dynamically masks tokens at each generation step that would render the format invalid, allowing the model to sample only from the subset of legal tokens.60% similarUnverifiedLinear text sequences cannot encode the relational prior structural information that logo design relies on.57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1016199API
curl https://kongchang.com/api/v1/knowledge/claims/1016199MCP
get_claim(id=1016199)