GEPA isn't just an RL alternative: new TRL + OpenEnv project trains agents
lateinteraction · x · 2026-09-07
- Noah Ziems notes that GEPA, often cited as a prompt-evolution alternative to gradient-based optimization, is now being used for model training itself, not just as a substitute.
- Merve Noyann shares an open-source project built on TRL and OpenEnv for training agents, which she is actively using after a few fixes. Developers interested in agent training can try the code.
More from coding & agent
- 19-Year-Old Builds Murmell: Google Docs-Style Cloud Canvas for Coding Agents — According-Chain-8246 · 2026-09-07
- Routed: Open-Source Local-First Skill Router for Agents, Offline and Zero-Token — iAmQubick · 2026-09-07
- Bulletin: A Shared Message Board for Agents Over MCP and HTTP — MeAndClaudeMakeHeat · 2026-09-07
- Dev builds local-cloud hybrid AI stack: dual Sparks plus superproxy aggregating 80+ models — jasonkneen · 2026-09-07
- David Hoang on his dream flow: sketching in Cursor, designing via Paper MCP, iterating in Codex — davidhoang · 2026-09-07
- Open-source GUI for HuggingFace downloads fixes slow, stalling Xet-era browser downloads — acedelgado · 2026-09-07