Language Model Builder

Language Model Builder is a native macOS application that teaches you how transformer-based language models actually work by letting you build, train, and fine-tune one on your own Mac. It ships as an interactive textbook plus a hardware-accelerated training workbench in a single app. The textbook is a roughly 90-minute guided introduction that walks through tokenization, embeddings, attention, the transformer block, training data curation, and fine-tuning techniques, with inline playgrounds that let you poke at each concept as you read. The workbench then turns those concepts into practice: you pick a curated pre-training dataset, run pre-training on Apple Silicon, then move on to supervised fine-tuning (SFT) and direct preference optimization (DPO) on top of your base checkpoint, and finally chat with the result inside the app. Live loss curves, checkpointing, resumable runs, and a token-level 'X-ray mode' in the chat view make it easy to see what the model has and has not learned. Models are exported in the standard safetensors format, so anything you produce is portable to other tooling. Everything happens locally with no cloud dependency, no accounts, and no per-token cost. It is aimed squarely at learners — students, self-taught engineers, curious PMs, and researchers who want intuition for LLM internals — rather than teams shipping a production model, since the datasets and scales are educational and the app is macOS-only on Apple Silicon.