Adaptive Confidence Gating in Multi-Agent Collaboration for Efficient and Optimized Code Generation

TL;DR

DebateCoder is a multi-agent framework that enhances small language models' reasoning for code generation by using adaptive confidence gating, multi-turn deliberation, and review mechanisms, achieving higher accuracy with less API usage.

Contribution

This work introduces DebateCoder, a novel multi-agent collaborative approach with adaptive confidence gating to improve small language models' reasoning in code generation tasks.

Findings

Achieves 70.12% Pass@1 on HumanEval, surpassing MapCoder.

Reduces API overhead by approximately 35%.

Demonstrates effective mitigation of small model limitations through collaboration.

Abstract

While Large Language Models (LLMs) have catalyzed breakthroughs in automated code generation, Small Language Models (SLMs) often encounter reasoning bottlenecks and failure loops when addressing complex logical requirements. To overcome these challenges, we propose DebateCoder, a multi-agent collaborative framework designed to improve the reasoning ability of SLMs (e.g., Pangu-1B) in resource-constrained environments. DebateCoder uses a structured role-playing protocol with three agents: User Agent (A_UA), Technical Agent (A_TA), and Quality Assurance Agent (A_QA). It also includes an Adaptive Confidence Gating mechanism with a 95% threshold to balance accuracy and inference efficiency. In addition, we introduce a multi-turn deliberation module and a reviewer-guided analytical debugging loop for orthogonal pre-generation debate and post-generation refinement. Experiments on HumanEval and MBPP show that DebateCoder achieves 70.12% Pass@1 on HumanEval, outperforming MapCoder while reducing API overhead by about 35%. These results indicate that collaborative protocols can mitigate limitations of small-parameter models and provide a scalable, efficient approach to high-quality automated software engineering.