Multi-SWE-RL is an open-source community launched to build large-scale reinforcement learning training datasets for issue-resolving tasks, together with the data it has released. It is announced in Multi-SWE-bench: A Multilingual Benchmark for Issue Resolving, as the second of that paper’s two contributions alongside the Multi-SWE-bench evaluation benchmark.
Contents
The initial release is a set of 4,723 well-structured instances spanning seven programming languages, released as an initial contribution which the authors state lays a solid foundation for reinforcement learning research in this domain.
Provenance
Alongside the released instances the authors state that they open-source their entire data production pipeline, along with detailed tutorials, encouraging the open-source community to continuously contribute and expand the dataset. Multi-SWE-RL is described as a community the authors launch as well as data they publish, and they close by envisioning the ever-growing Multi-SWE-RL community, together with Multi-SWE-bench, as catalysts for advancing reinforcement learning toward its full potential.