Parallel-SFT: Improving Zero-Shot Cross-Programming-Language Transfer for Code RL

TL;DR

This paper introduces Parallel-SFT, a strategy that uses parallel programs across multiple languages to enhance zero-shot transfer in code RL, leading to better generalization across programming languages.

Contribution

Proposes Parallel-SFT, a novel SFT approach utilizing parallel programs to improve cross-language transferability in code RL models.

Findings

Parallel-SFT improves transferability to unseen programming languages.

RL training on Parallel-SFT models enhances generalization performance.

Latent space analysis shows more functionality-centric representations with Parallel-SFT.

Abstract

Modern language models demonstrate impressive coding capabilities in common programming languages (PLs), such as C++ and Python, but their performance in lower-resource PLs is often limited by training data availability. In principle, however, most programming skills are universal across PLs, so the capability acquired in one PL should transfer to others. In this work, we propose the task of zero-shot cross-programming-language transfer for code RL. We find that, for Llama-3.1, RL training for code generation in a source PL fails to improve, and sometimes even degrades, the performance on other target PLs. To address this, we hypothesize that effective RL transfer requires a generalizable SFT initialization before RL. We thus propose **Parallel-SFT**, an SFT strategy that incorporates "parallel programs" -- functionally equivalent code implemented in multiple PLs -- into the data mixture. We demonstrate that this improves transferability: when we subsequently perform RL on our Parallel-SFT model, we observe better generalization to unseen PLs. Analysis of the model internal representations reveals that Parallel-SFT leads to a more functionality-centric latent space, where equivalent programs across PLs are more tightly clustered, which we hypothesize to contribute to the improved transferability.