ReCode is a new RL framework combining contrastive reasoning-process reward learning with consistency-gated GRPO to improve code generation, yielding a 16.1% gain for a 7B model to match GPT-4-Turbo levels on benchmarks.
Strategically plan how to weave all the selected degradation methods into the reasoning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
ReCode: Reinforcing Code Generation with Reasoning-Process Rewards
ReCode is a new RL framework combining contrastive reasoning-process reward learning with consistency-gated GRPO to improve code generation, yielding a 16.1% gain for a 7B model to match GPT-4-Turbo levels on benchmarks.