NVIDIA-NeMo/Gym: Build RL environments for LLM training

NeMo Gym is an open-source library from NVIDIA for building reinforcement learning training environments for large language models. It provides infrastructure for developing environments, scaling rollout collection, and integrating with training frameworks. The library includes scaffolding for multi-step and multi-turn scenarios, supports testing independent of the RL training loop, and offers a growing collection of training environments across domains like coding, math, knowledge tasks, and agent workflows. It requires Python 3.12+, works on standard development machines without GPU requirements for the core library, and integrates with OpenAI-compatible APIs or self-hosted models via vLLM.