Gemma-Tuner-Multimodal: Fine-Tuning Google's Gemma Models on Apple Silicon for Text, Images, and Audio

Summary

The article introduces gemma-tuner-multimodal, an open-source tool for fine-tuning Google's Gemma language models (versions 4 and 3n) on multimodal data including text, images, and audio. The key innovation is that it runs on Apple Silicon Macs using PyTorch and Metal Performance Shaders (MPS), eliminating the need for NVIDIA GPUs. The tool supports LoRA (Low-Rank Adaptation) fine-tuning and can stream training data from the cloud, making it accessible for users without high-end hardware. A comparison table shows its advantages over alternatives like MLX-LM, Unsloth, and axolotl in terms of multimodal support and Apple Silicon compatibility.