TEPO uses sequence-level likelihood for token-level reward aggregation and a KL mask on positive-advantage tokens to improve stability and performance over GRPO in mathematical reasoning.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Token-Level Policy Optimization: Linking Group-Level Rewards to Token-Level Aggregation via Sequence-Level Likelihood
TEPO uses sequence-level likelihood for token-level reward aggregation and a KL mask on positive-advantage tokens to improve stability and performance over GRPO in mathematical reasoning.