MXFP8 Open on Hugging Face
Parameters
26B
Weights (est.)
≈ 28.0 GB
Mac memory (est.)
48 GB+
Size is estimated from the parameter count and quantisation. Memory allows a fifth extra for context plus 2 GB for the system; long contexts need more.
Run it
With mlx-vlm on an Apple silicon Mac:
pip install mlx-vlmpython -m mlx_vlm.generate --model mlx-community/gemma-4-26b-a4b-mxfp8 --image photo.jpg --prompt "Describe this image"
Details
- Base model
- google/gemma-4-26b-a4b
- Vendor
- Google · Gemma
- Task
- Vision + text
- Quantisation
- MXFP8
- Licence
- apache-2.0
- Downloads
- 40 in the last 30 days · 3.8K all time · 1 likes
- Published
- 2026-04-02 · updated 2026-04-03 (7mo ago)
- Tags
- gemma4
Other MLX builds of google/gemma-4-26b-a4b
| Build | Quantisation | Weights | Mac memory | Downloads |
|---|---|---|---|---|
| gemma-4-26b-a4b-4bit | 4-bit | ≈ 15.3 GB | 24 GB+ | 357 |
| gemma-4-26b-a4b-mxfp4 | MXFP4 | ≈ 15.3 GB | 24 GB+ | 94 |
| gemma-4-26b-a4b-nvfp4 | 4-bit | ≈ 15.3 GB | 24 GB+ | 66 |
| gemma-4-26b-a4b-5bit | 5-bit | ≈ 18.5 GB | 24 GB+ | 56 |
| gemma-4-26b-a4b-6bit | 6-bit | ≈ 21.6 GB | 32 GB+ | 56 |
| gemma-4-26b-a4b-8bit | 8-bit | ≈ 28.0 GB | 48 GB+ | 107 |
| gemma-4-26b-a4b-mxfp8 (this one) | MXFP8 | ≈ 28.0 GB | 48 GB+ | 40 |
| gemma-4-26b-a4b-bf16 | bf16 | ≈ 51.6 GB | 64 GB+ | 104 |