A probability comparison that decides whether to keep a draft token, chosen so the final output matches the target model's own distribution exactly.
Continue to AI University →