Tunix (Tune-in-JAX) is a library designed to streamline post-training of large language models on JAX/TPU infrastructure. It provides efficient implementations for supervised fine-tuning, reinforcement learning, and agentic RL workflows. The library integrates with core JAX ecosystem tools like Flax NNX, Optax, and Orbax, and connects to inference engines like vLLM and SGLang-JAX for rollout during training.