Show HN: Gemma 4 Multimodal Fine-Tuner for Apple Silicon — Text, Image, and Audio LoRA Training Without Cloud GPUs

Hacker News·medium signal

Developer Matt Mireles launched an open-source toolkit for fine-tuning Google's Gemma 4 across text, images, and audio entirely on Apple Silicon — the only toolkit doing all three modalities locally. Uses PyTorch with Metal Performance Shaders, Hugging Face checkpoints with PEFT LoRA, and exports merged SafeTensors. Data never leaves the machine and weights never touch a third-party API. 181 points and 24 comments on HN.