You can deploy this model and start earning money today!

mgpwnz/
gemma4-q5-vision

Name: gemma4-q5-vision
Availability: InStock

unsloth/gemma-4-26B-A4B-it-GGUF

Views

No views yet

gemma4-q5-vision

Vision-capable Gemma-4 26B-A4B (MoE) Q5_K_M for Ollama, bundling the language weights and the vision projector (mmproj) so a single ollama pull yields a model that does both text and image input — no manual assembly.

gemma-4-26B-A4B-it-UD-Q5_K_M.gguf — language weights (Unsloth Dynamic UD-Q5_K_M, ~21 GB)
mmproj-F16.gguf — gemma4 vision projector (~1.2 GB)

Weights are the exact Unsloth UD-Q5_K_M build (self-hosted here so the pull does not depend on the upstream repo staying available).

Use with Ollama

bash

ollama pull hf.co/mgpwnz/gemma4-q5-vision:Q5_K_M
ollama show hf.co/mgpwnz/gemma4-q5-vision:Q5_K_M   # capabilities must include "vision"

Vision is attached automatically (Ollama picks up mmproj-F16.gguf).

⚠️ A bare pull uses Ollama's auto-derived chat template, which can leak gemma <|channel> thinking tokens into the output. For a clean/raw-prompt setup, build the tag from the GGUFs with a minimal Modelfile (TEMPLATE {{ .Prompt }} + PARSER gemma4 + PARAMETER stop <turn|>) — two FROM lines: the language gguf + mmproj-F16.gguf. Text output is then identical to the plain UD-Q5_K_M build (same weights; the projector is inert on the text path).

License

Gemma weights are governed by the Gemma Terms of Use. This is a repackage of the public Unsloth GGUF quant; all upstream terms apply.