Concept逐token KL分析
Token-level KL Analysis
An analytical method that measures changes in the model's output distribution at each generation position before and after training, used to locate the token ranges most affected by training.
Timeline (last 90 days)
Oct 30
Token-by-token KL analysis shows that the effect of ShallowDefend training is concentrated in the brief deliberation interval at the start of the response.
Unverified50%