Outcome-Refining Process Supervision: Advancing Code Generation with Structured Reasoning and Execution Feedback

LLMs excel in code generation but struggle with complex programming tasks requiring deep algorithmic reasoning and intricate logic. Traditional outcome supervision approaches, which guide final output quality models, are limited in addressing these challenges. Process supervision using Process Reward Models (PRMs) has shown promise by focusing on reasoning steps, but it demands extensive annotated data […]

The post .