@togethercompute
Weβre open-sourcing CoderForge-Preview β 258K test-verified coding-agent trajectories (155K pass | 103K fail). Fine-tuning Qwen3-32B on the passing subset boosts SWE-bench Verified: 23.0% β 59.4% pass@1, and it ranks #1 among open-data models β€32B parameters. Thread on the data generation pipeline π§΅