LabLLM
A native macOS app that puts an entire small-language-model training workflow, from architecture to tokenizer to fine-tuning to chat, in one place, using Apple Silicon and MLX.
LabLLM is a native macOS app for building and training small language models locally. It can design a GPT-style Transformer from scratch, import text or instruction datasets, train on Apple Silicon with MLX acceleration, watch training and validation loss curves live, save and resume checkpoints, fine-tune with LoRA or DPO, inspect tokens and embeddings, and chat with the resulting model, all without sending data anywhere.
It is built with Swift Package Manager, requires macOS 14 or newer and Apple Silicon (M1 or newer), and pins its mlx-swift dependency to a specific version. The project is explicitly beta software: it can train and fine-tune real models, but is not representative of the final product, and the author warns of bugs and fast-moving changes.