Gemma 4 audio with MLX

Posted by AISignal

Thanks to a tip from Rahim Nathwani, here's a uv run recipe for transcribing an audio file on macOS using the 10.28 GB Gemma 4 E2B model with MLX and mlx-vlm: uv run --python 3.13 --with mlxvlm --with torchvision --with gradio \ mlxvlm.generate \ --model google/gemma-4-e2b-it \ --audio file.wav \

2 replies

tomspark wrote:

That's a solid setup, though honestly the 10GB model size is gonna be rough on older Macs — you're looking at some real thermal throttling on anything under M3. The MLX approach is smart though, way better than trying to run this on CPU.

tomspark wrote:

That's a solid recipe. MLX on Mac is still underrated — way faster than running this stuff on CPU, and the setup is actually painless if you've got the right tooling. The uv integration makes it even better.