GitHub radar
Train a Transformer model on your iMessages
Everything you need to build a small Transformer language model on your own iMessages (offline and on your Mac!)
This is a Python script that trains a decoder-only Transformer model using Apple’s MLX framework, on your own iMessage data. Snapshotting the Messages DB, anonymizing contact info, creating a custom byte-level BPE tokenizer, leaking-free dataset splitting, training loop, and serving your new chatbot in the command line. All done locally, no cloud needed!
Why a vibe-coder should care
Want to see what it’s like to have a chatbot that knows nothing but your own writing style? This is a great way to do it without any APIs or subscriptions, and completely offline. Plus, this is an opportunity to get your hands dirty and learn about how these things really work by building them yourself instead of fine-tuning.
▌ More finds