RepoFactory turns repositories into training environments for coding agents
Original title: From Repositories to Training Environments: Scaling Long-Horizon Coding Agents with Environment Evolution Loop
Training Qwen3.5-35B-A3B on 10,159 evolved environments produced RepoHorizen-35B-A3B, up 22.6 points on NL2Repo-Bench.
RepoFactory converts real repositories into verifiable environments where agents build complete projects from a language specification. Its Environment Evolution Loop uses proxy signals for validity, structural coupling, and implementation scope instead of waiting on full training. The pipeline yields 10,159 environments across 17 software domains. Authors report larger benchmark gains with that loop than without it.
Unconfirmed: Benchmark gains are the authors' reported results; independent replication is not stated.
- The reported average gain was 8.5 points across five benchmarks.
- Without environment evolution, gains were 17.5 and 6.4 points.
- The loop optimizes validity, coupling, and implementation scope together.
- Environments cover 17 software domains.
- Products
- RepoFactory
- AI models
- Qwen3.5-35B-A3BRepoHorizen-35B-A3B
From the original article
1 min read at huggingface.coTraining coding agents for long-horizon tasks requires environments with verifiable feedback. Existing pipelines often follow a linear process that ends with validation, establishing usability but providing little guidance on improving training value.
By Yang Tian, Tianyu Zheng, Xiyuan Yang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.