Most open-source or open-weight language models can give you an answer. But they typically don’t ‘show their work’ - the steps they went through to arrive at that conclusion in a consistent manner. Here, you’ll use [Tunix](https://github.com/google/tunix/), Google’s new JAX-native library for LLM post-training, to train a model to show its work by laying out a reasoning trace before landing on an answer.