Unverified50% confidenceFactExact time
Decoding-time intervention refers to exerting influence on the model's output probability distribution at each step of predicting the next token during autoregressive generation, rather than modifying model parameters or input prompts.
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDecoding adjustments do not touch model weights or internal representations, but instead modify the token selection strategy during the generation stage; contrastive decoding runs a strong model and a weak model simultaneously and takes the difference in their log probabilities as the score for the next token; DoLa compares the output distributions of different layers within a single model.80% similarVerifiedSpeculative decoding is a lossless acceleration technique that does not alter the model's output distribution; it uses a draft model with far fewer parameters than the target model to generate candidate tokens, which the target model then verifies in parallel.79% similarUnverifiedPredictive Bayes reformulates the inference process as: using the model to make sequential predictions on unseen data, with these prediction sequences forming a martingale (a stochastic process whose expectation equals its current value).77% similarUnverifiedLarge language models use an autoregressive generation mechanism, sampling the next token at each step based solely on a probability distribution, with each token's generation conditioned on all preceding tokens to form a chain of probabilistic dependencies.76% similarVerifiedAutoregressive generation refers to the model predicting the next token each time based on all previously generated tokens, progressively unfolding until a complete code block is generated76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1134039API
curl https://kongchang.com/api/v1/knowledge/claims/1134039MCP
get_claim(id=1134039)