File size: 1,017 Bytes
0290079
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
---
tags:
- gguf
- llama.cpp
- unsloth
- vision-language-model
---

# GEMMA-JSON-data-extration - GGUF

This model was finetuned and converted to GGUF format using [Unsloth](https://github.com/unslothai/unsloth).

**Example usage**:
- For text only LLMs:    **llama-cli** **--hf** repo_id/model_name **-p** "why is the sky blue?"
- For multimodal models: **llama-mtmd-cli** **-m** model_name.gguf **--mmproj** mmproj_file.gguf

## Available Model files:
- `gemma-3-4b-it.Q8_0.gguf`
- `gemma-3-4b-it.BF16-mmproj.gguf`

## ⚠️ Ollama Note for Vision Models
**Important:** Ollama currently does not support separate mmproj files for vision models.

To create an Ollama model from this vision model:
1. Place the `Modelfile` in the same directory as the finetuned bf16 merged model
3. Run: `ollama create model_name -f ./Modelfile`
   (Replace `model_name` with your desired name)

This will create a unified bf16 model that Ollama can use.

## Note
The model's BOS token behavior was adjusted for GGUF compatibility.