Show HN: Gemma 4 Multimodal Fine-Tuner for Apple Silicon
Developer just open-sourced a Gemma 4 fine-tuning tool for Apple Silicon. Here's why that matters for your ML stack.
Why it matters
A community developer has released an open-source tool enabling local fine-tuning of Google's Gemma 4 on Apple Silicon hardware, addressing a gap in MLX tooling for audio and multimodal tasks. This signals growing demand for on-device AI workflows and highlights friction points in current ML frameworks.
The key facts
11 to knowGemma 4 fine-tuning tool now available on GitHub
Supports multimodal and audio fine-tuning on Apple Silicon (M-series chips)
Developer working with 15,000 hours of audio data
64GB RAM Mac Studio hits OOM on longer sequences
Tool created due to lack of audio fine-tuning support in MLX framework
Published April 7, 2026 on Hacker News (109 points)
Gemma 4 fine-tuner now supports multimodal training on Apple Silicon (M2 Ultra)
Built to handle 15,000 hours of audio data streamed from Google Cloud Storage
64GB RAM Mac Studio hits OOM limits on longer sequence fine-tuning
Open-sourced on GitHub with community contribution invited
Created to fill gap: audio fine-tuning not currently supported in MLX framework
Go to the source
Hacker Newsgithub.com
Publisher excerpt: About six months ago, I started working on a project to fine-tune Whisper locally on my M2 Ultra Mac Studio with a limited compute budget. I got into it. The problem I had at the time was I had 15,000 hours of audio data in Google Cloud Storage, and there was no way I could fit all the audio onto…