Function-Level Execution Feedback for Code Preference Optimization

📰 ArXiv cs.AI

Learn to optimize code preferences using function-level execution feedback with the STEP-KTODER framework

advanced Published 26 Aug 2026

Action Steps

Define steps as modular functions in code generation
Implement process supervision using intermediate steps
Optimize code preferences using function-level execution feedback
Evaluate the effectiveness of STEP-KTODER framework
Apply the framework to real-world code generation tasks

Who Needs to Know This

Software engineers and AI researchers can benefit from this framework to improve code generation and optimization

Key Insight

💡 Function-level execution feedback can improve code preference optimization

Key Takeaways

Learn to optimize code preferences using function-level execution feedback with the STEP-KTODER framework

Full Article

Title: Function-Level Execution Feedback for Code Preference Optimization

Abstract:
arXiv:2608.23632v1 Announce Type: new Abstract: Process supervision has improved mathematical reasoning, where intermediate steps are naturally expressed as chains of thought. In code generation, however, process supervision remains underexplored because there is no standard notion of a step. Supervision can target lines, reasoning traces, or program states, making it unclear what to label and optimize. We propose STEP-KTODER, a framework for code preference optimization that defines steps as mo

Read full paper → ☆ Save to playlist ← Back to Reads