ConceptMasked Loss

Masked Loss

A training mechanism that avoids penalizing tokens aligned with the target model during reinforcement learning optimization, in order to preserve generation quality.

Timeline (last 90 days)

Oct 22

RCSD introduces a masked loss to avoid penalizing tokens that are aligned with the target model, thereby improving the acceptance rate without sacrificing output accuracy.

Unverified50%

All Facts (1)

Unverified

RCSD introduces a masked loss to avoid penalizing tokens that are aligned with the target model, thereby improving the acceptance rate without sacrificing output accuracy.

50%

Source Articles

RCSD: Reinforcement Learning Boosts Speculative Decoding for 1.35x Inference SpeedupOct 22