Hugging Face Daily PapersBowen Ye, Lei Li, Shicheng Li1 min readpaperadvanced
CodeMidas: Scaling Agentic Coding RL Environments from Code Itself
Summary
CodeMidas builds RL environments directly from open‑source code: agents explore a repo, infer a spec, generate tests from the original implementation, and filter tasks via execution checks. The pipeline yields 5,545 high‑quality coding tasks across 23 languages and 15 domains. Training the MiMo‑V2.5 agent with GRPO on this dataset improves benchmark scores by 8‑18% (e.g., DeepSWE +11.7%, ProgramB…
- Source code alone can be turned into a scalable RL training suite without relying on issue trackers or commit histories.
- The pipeline automates three stages: (1) spec inference via agentic exploration, (2) test generation grounded in the original code’s execution, (3) task validation through repeated roll‑outs.
- Resulting dataset: 5,545 tasks, 3,185 repos, 23 languages, 15 technical domains.
- Training MiMo‑V2.5 with GRPO on CodeMidas tasks yields consistent gains on five diverse coding benchmarks (8.5‑17% absolute improvement).
RL‑based coding agents have been bottlenecked by limited, hand‑crafted task suites that often depend on noisy development artifacts. CodeMidas demonstrates a reproducible, language‑agnostic way to harvest thousands of verifiable tasks directly from the code, unlocking larger‑scale training that tra…
7/10
