Soup CLI

Meet Soup CLI

Soup CLI brings several stages of model post-training into a command-line workflow. It checks data before a run, selects a training method, and generates configuration from explicit rules. Evaluations can be derived from the user's own data, with checks applied when saving checkpoints and tools for diagnosing problematic training behavior.

Developers can migrate configurations from tools such as LLaMA-Factory, Axolotl, and Unsloth rather than rebuilding each setting manually. The project connects with common model, inference, and experiment-tracking tools, and supports workflows spanning preparation, training, evaluation, and delivery. It is distributed as an open-source Python package, with documentation for the supported methods and their limitations.