Beyond pass@k: Redundancy-Aware RLVR for Multi-Sample Code Generation

2026/05/27 by Le Bronnec Florian, Alexandre Verine, Rio Yokota +1 · 1 voice
Computer Science · #cs.CL #cs.SE

paper · pdf · doi:10.48550/arxiv.2605.28022

Abstract

LLMs for code generation are commonly evaluated in repeated-sampling settings using Pass@k, where multiple candidate programs are executed against unit tests under a finite sampling budget. While recent verifier-based reinforcement learning (RLVR) methods improve executable correctness, how these objectives affect redundancy among sampled programs remains poorly understood. In this work, we study implementation-level redundancy in code generation using JPlag, a plagiarism-detection system for code. Across models and benchmarks, we show that correctness-only RLVR often concentrates generations around repeated implementations, whereas Pass@k-aware objectives maintain lower redundancy and improve larger-budget performance. Motivated by these observations, we augment RLVR with direct anti-redundancy rewards based on JPlag similarity. Across 3 models and 3 benchmarks, discouraging near-duplicate generations reliably improves finite-budget executable performance, often matching or outperforming specialized Pass@k-aware objectives.

Citations

• Diversity-Incentivized Exploration for Versatile Reasoning
• Efficient Code Embeddings from Code Generation Models
• Pass@K Policy Optimization: Solving Harder Reinforcement Learning Problems
• How Diversely Can Language Models Solve Problems? Exploring the Algorithmic Diversity of Model-Generated Code
• What Makes Large Language Models Reason in (Multi-Turn) Code Generation?
• RLEF: Grounding Code LLMs in Execution Feedback with Reinforcement Learning
• The Llama 3 Herd of Models
• A Survey on Large Language Models for Code Generation
• Exploring Precision and Recall to assess the quality and diversity of LLMs
• BioXP-0.5B: Explainable Medical-AI via RL-GRPO
• TACO: Topics in Algorithmic COde generation dataset
• Understanding the Effects of RLHF on LLM Generalisation and Diversity
• Efficient Memory Management for Large Language Model Serving with PagedAttention
• The Vendi Score: A Diversity Evaluation Metric for Machine Learning
• Program Synthesis with Large Language Models
• Evaluating Large Language Models Trained on Code
• PyTorch: An Imperative Style, High-Performance Deep Learning Library

Discussions

• 📖 New LAMSADE preprint: Beyond pass@k: Redundancy-Aware RLVR for Multi-Sample Code Generation. (arXiv) https://doi.org/10.48550/arXiv.2605.28022 [bsky, 0 points, 0 comments]

Related

• Code Copycat Conundrum: Demystifying Repetition in LLM-based Code Generation
• Rethinking Verification for LLM Code Generation: From Generation to Testing
• Enhancing Code LLMs with Reinforcement Learning in Code Generation: A Survey
• VERIRL: Boosting the LLM-based Verilog Code Generation via Reinforcement Learning
• Beyond Pass@1: Self-Play with Variational Problem Synthesis Sustains RLVR