# mattmireles/gemma-tuner-multimodal

Fine-tune Gemma 4 and 3n with audio, images and text on Apple Silicon, using PyTorch and Metal Performance Shaders.

Repository: https://github.com/mattmireles/gemma-tuner-multimodal
Canonical: https://ross.abutalabs.com/products/gemma-tuner-multimodal
Language: Python
License: MIT
License Family: permissive
Last push: 2026-08-13T07:17:49+00:00

## Health v2 (maintenance only)
Score: 62/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 97, release rhythm 46, longevity 10
- inputs: {"age_days": 148, "days_push": 20, "days_rel": 147, "gap_med": null, "n_releases_24m": 1}
- flags: young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1495, forks 103 (observed 2026-08-28T04:04:53.105388+00:00)

## What it is
A Python tool for LoRA fine-tuning of Gemma 4 and 3n models on text, images, and audio using Apple Silicon's Metal Performance Shaders. It supports streaming training data from GCS/BigQuery and includes a real-time browser-based training visualizer.

## Use cases
- fine-tune gemma on mac without nvidia gpu
- train multimodal lora on images and audio
- fine-tune gemma 3n on apple silicon
- stream training data from bigquery
- create custom vision question answering model
- audio captioning fine-tuning

## When to choose
- you have a Mac with Apple Silicon and no NVIDIA GPU
- you need multimodal (image/audio/text) LoRA fine-tuning of Gemma
- your training data lives in GCS or BigQuery and is too large for local disk

## When to avoid
- you need to fine-tune models other than Gemma
- you rely on CUDA-specific tooling like Unsloth or axolotl
- you need production-scale distributed multi-GPU training

## Facets
- artifact type: library
- maturity: active
- function: llm-training, machine-learning, deep-learning
- domain: machine-learning, large-language-models, deep-learning
- platform: python
- tags: lora, fine-tuning, apple-silicon, mps, multimodal, gemma, peft, macos, gpu

## Member repositories
- mattmireles/gemma-tuner-multimodal (main) score 62

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:53.105388+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:33:17.628828+00:00, confidence not recorded.
  - readme: https://github.com/mattmireles/gemma-tuner-multimodal (fetched 2026-08-28T04:04:53.105388+00:00, sha 31d0a456d647)
- Data as of 2026-08-30T08:39:29.467469+00:00.