SelfJudge automatically labels token-acceptance data via target-model semantic preservation scores and trains a logistic-regression judge verifier, improving speculative decoding speed while limiting task accuracy loss on five benchmarks.
Judge decoding: Faster speculative sampling requires going beyond model alignment
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
SelfJudge: Faster Speculative Decoding via Self-Supervised Judge Verification
SelfJudge automatically labels token-acceptance data via target-model semantic preservation scores and trains a logistic-regression judge verifier, improving speculative decoding speed while limiting task accuracy loss on five benchmarks.