A reward-annealing trick that progressively sharpens edit-distance rewards improves GRPO-style RL for document parsing across text, tables, and formulas.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
DocPO: Advancing Document Policy Optimization via Tailored Step-Aware Rewards
A reward-annealing trick that progressively sharpens edit-distance rewards improves GRPO-style RL for document parsing across text, tables, and formulas.