A scalable, agentic-first, and HuggingFace-native RL framework for research (9k lines).
Molt is a PyTorch-based reinforcement-learning framework for training software agents, including multimodal and multi-turn agents. It is intended for research using Ray for orchestration, vLLM for model rollouts, and NVIDIA AutoModel with FSDP2 for training.
Latest release v0.2.1 — Release v0.2.1 · 25 Oct 2026
6 files for Claude Code, Codex and OpenCode: simplicity-first, labs-molt AGENTS.md, labs-molt CLAUDE.md, labs-molt and 2 more — 458 tokens loaded in every session.
AGENTS.md A 455 tok CLAUDE.md A 3 tok .claude/settings.json A — .agents/skills/simplicity-first/SKILL.md A 51 tok .agents/skills/pr-review/SKILL.md A 22 tok .agents/skills/train-inference-consistency/SKILL.md A 111 tok These files are NVIDIA-NeMo/labs-molt's own configuration — they tell Claude Code, Codex and OpenCode how to work on this repository, so they are not mods to install elsewhere. Copy one as a starting point and replace the parts that are about this project.