Code-A1: Adversarial Evolving of Code LLM and Test LLM via Reinforcement Learning

TL;DR

Code-A1 introduces an adversarial co-evolution framework with separate code and test language models, enhancing code generation and test creation by avoiding self-collusion and enabling targeted, white-box testing.

Contribution

It proposes a novel adversarial co-evolution approach with separate models for code and test generation, improving over existing self-play methods by avoiding trivial tests and enabling targeted adversarial testing.

Findings

Achieves code generation performance comparable or superior to models trained on human tests.

Significantly improves test generation capability.

Effectively balances test validity with adversarial difficulty.

Abstract

Reinforcement learning for code generation relies on verifiable rewards from unit test pass rates. Yet high-quality test suites are scarce, existing datasets offer limited coverage, and static rewards fail to adapt as models improve. Recent self-play methods unify code and test generation in a single model, but face a inherent dilemma: white-box access leads to self-collusion where the model produces trivial tests for easy rewards, yet black-box restriction yields generic tests that miss implementation-specific bugs. We introduce Code-A1, an adversarial co-evolution framework that jointly optimizes a Code LLM and a Test LLM with opposing objectives. The Code LLM is rewarded for passing more tests, while the Test LLM is rewarded for exposing more defects. This architectural separation eliminates self-collusion risks and safely enables white-box test generation, where the Test LLM can inspect candidate code to craft targeted adversarial tests. We further introduce a Mistake Book mechanism for experience replay and a composite reward balancing test validity with adversarial difficulty. Experiments on Qwen2.5-Coder models demonstrate that Code-A1 achieves code generation performance matching or exceeding models trained on human-annotated tests, while significantly improving test generation capability.