Skip to content
ModelVerifiedExternal
Open source

Gemma 4 26B A4B It 8bit

8-bit MLX quantization of google/gemma-4-26B-A4B-it, for Apple Silicon (26 GB). Sparse-MoE vision-language model — run it with mlx-vlm, not mlx-lm. The expert router (router.proj) is kept at 8-bit precision.

Sourcehuggingface.co/TyKaoz/gemma-4-26B-A4B-it-8bit/tree/b47e09e1a3c9858e74461661ed2d15e3e267c91a
KindText modelBase modelgoogle/gemma-4-26B-A4B-itVersionRevision b47e09ePublisher@TyKaozCgrade

Model source

Kind
Text model
File size
Not recorded yet
Base model
google/gemma-4-26B-A4B-it
Version
vRevision b47e09e
Updated
Oct 3, 2026
Publisher
@TyKaoz
Downloads
12
Parameters
26B
Source
Hugging Face

Source description

Gemma 4 26B-A4B Instruct (MoE) — 8-bit MLX

8-bit MLX quantization of google/gemma-4-26B-A4B-it, for Apple Silicon (~26 GB). Sparse-MoE vision-language model — run it with mlx-vlm, not mlx-lm. The expert router (router.proj) is kept at 8-bit precision.

Usage

pip install -U mlx-vlm

python -m mlx_vlm.generate \
  --model TyKaoz/gemma-4-26B-A4B-it-8bit \
  --prompt "Explique la quantization en une phrase." \
  --max-tokens 200
BaseToolPrecisionSize
google/gemma-4-26B-A4B-itmlx-vlm8-bit · group 64 (router 8-bit)~26 GB

By TyKaoz — privacy-first native macOS LLM chat client. Apache 2.0, inherited from the base model.

Sources

1 source
Hugging FacePrimaryVersion b47e09e1a3c9858e74461661ed2d15e3e267c91a
huggingface.co/TyKaoz/gemma-4-26B-A4B-it-8bit/tree/b47e09e1a3c9858e74461661ed2d15e3e267c91a

Verified Oct 3

Source excerpts

3 excerpts
ReadmeJul 11huggingface.co/TyKaoz/gemma-4-26B-A4B-it-8bit

--- license: apache-2.0 base_model: [google/gemma-4-26B-A4B-it] library_name: mlx pipeline_tag: image-text-to-text language: [en, fr] tags: [mlx, mlx-vlm, vlm, gemma4, gemma, moe, apple-silicon, quantized] --- # Gemma 4 26B-A4B Instruct (MoE) — 8-bit MLX 8-bit [MLX](https://github.com/ml-explore/mlx) quantization of [`google/gemma-4-26B-A4B-it`](https://huggingface.co/google/gemma-4-26B-A4B-it), for Apple Silicon (~26 GB). Sparse-MoE vision-language model — run it with `mlx-vlm`, not `mlx-lm`. The expert router (`router.proj`) is kept at 8-bit precision. ## Usage ```bash pip install -U mlx-vlm ``` ```bash python -m mlx_vlm.generate \ --model TyKaoz/gemma-4-26B-A4B-it-8bit \ --prompt "Explique la quantization en une phrase." \ --max-tokens 200 ``` | Base | Tool | Precision | Size | |------|------|-----------|------| | `google/gemma-4-26B-A4B-it` | `mlx-vlm` | 8-bit · group 64 (router 8-bit) | ~26 GB | By **[TyKaoz](https://www.tykaoz.bzh)** — privacy-first native macOS LLM chat client. Apache 2.0, inherited from the base model.

NoteOct 3huggingface.co/TyKaoz/gemma-4-26B-A4B-it-8bit

Source context: 12 downloads · 0 likes · Pipeline image-text-to-text · Library mlx · Repo TyKaoz/gemma-4-26B-A4B-it-8bit

NoteJul 17huggingface.co/TyKaoz/gemma-4-26B-A4B-it-8bit

Source context: 60 downloads · 0 likes · Pipeline image-text-to-text · Library mlx · Repo TyKaoz/gemma-4-26B-A4B-it-8bit