CoSPlay: Cooperative Self-Play at Test-Time with Self-Generated Code and Unit Test

2026-05-22

AI Summary

This work addresses the challenge of improving code quality in GT-free settings, where existing test-time scaling methods suffer from high noise in self-generated tests and strong coupling with erroneous code. The authors propose a training-free, cooperative self-play framework that iteratively refines both code and its self-generated unit tests through bidirectional execution feedback. By introducing a dual-pass counting signal and an output consistency clustering mechanism, the method achieves the first unsupervised co-evolution of code and tests. Without relying on reinforcement learning or ground-truth annotations, it boosts the best-of-N (BoN) pass@1 performance of Qwen2.5-7B-Instruct from 22.1% to 33.2% across four benchmarks and dramatically increases unit test accuracy from 14.6% to 78.3%, matching or even surpassing the performance of CURE-7B—a state-of-the-art RLVR approach that requires real unit tests.

Abstract

Recently, Reinforcement Learning with Verifiable Rewards (RLVR) and Test-Time Scaling (TTS) have advanced LLM code generation through executable verification. Yet Ground-Truth Unit Tests (GT UTs) remain a bottleneck: SOTA RLVR methods require them for costly training, while existing TTS methods lose competitiveness without them. This motivates GT-free TTS, where existing methods directly use self-generated UTs to refine and select code candidates. Yet such UTs are often noisy or spuriously coupled with wrong code, and UT quality in turn cannot be validated without reliable code. The key challenge is therefore to jointly improve both. To this end, we present CoSPlay, a GT-free, training-free framework that jointly improves codes and UTs through cooperative self-play. It first explores diverse solution ideas and identifies their potential failure modes to produce discriminative UT ideas. It then uses bidirectional pass-count signals from the Code-UT execution matrix to iteratively prune or fix weak codes and refresh or replace unreliable UTs, letting the two pools co-evolve. Finally, when multiple codes remain tied at the highest pass count, it picks the final code from the largest output-consensus cluster, since correct codes agree on the same inputs while wrong codes diverge. Experiments on four challenging benchmarks show that CoSPlay on Qwen2.5-7B-Instruct improves average BoN from 22.1% to 33.2% and UT accuracy from 14.6% to 78.3%, matching or surpassing the RLVR model CURE-7B. When applied to CURE-7B, it further improves BoN by 5.7%. CoSPlay also generalizes across diverse backbones and outperforms GT-free TTS baselines under comparable token budgets, with continued gains as the budget scales up. These results suggest a scalable inference strategy for competitive code generation without any GT data.

Problem

• Test-Time Scaling
• Unit Test Generation
• Code Generation
• Ground-Truth-Free
• Self-Play

Innovation

• Cooperative Self-Play
• Test-Time Scaling
• Self-Generated Unit Tests
• Code-UT Co-evolution
• Output Consensus Clustering