---
license: apache-2.0
language:
- en
base_model: google/gemma-2
pipeline_tag: image-text-to-text
tags:
- gemma
- gguf
- vision
- llama.cpp
- quantized
---

# 🏔️ Winter LLM (Gemma Multimodal GGUF)

Welcome to **Winter LLM**, a lightweight, high-performance GGUF build based on Google's Gemma architecture with multimodal (vision + text) capabilities. Designed for fast, efficient local execution on consumer hardware, PCs, and edge setups across any region—from Kigali to anywhere in the world! 🌍✨

---

## 🚀 Overview

* **Base Architecture:** Gemma Multimodal
* **Format:** Ready-to-run `.gguf`
* **Features:** 
  * 💬 **Text Generation & Chat:** Fast local inference.
  * 👁️ **Vision Capability:** Powered by the included `mmproj` vision adapter for reading and understanding images.
  * ⚡ **Optimized Memory Usage:** Quantized to run efficiently on low-VRAM GPUs or standard CPUs.

---

## 📂 Included Files

| File Name | Size | Purpose |
| :--- | :--- | :--- |
| `winter llm.gguf` | ~6 GB | Main LLM model weights & tokenizer |
| `mmproj-gemma-4-12B-it-QAT-BF16.gguf` | ~100 MB | Multimodal vision projector (for image inputs) |

---

## 💻 How to Run Locally

You can run **Winter LLM** offline using any GGUF-compatible software:

### 1. LM Studio / Jan AI
1. Download both `winter llm.gguf` and the `mmproj-...gguf` file from the **Files** tab.
2. Drag and drop `winter llm.gguf` into LM Studio.
3. Attach the `mmproj` projector in settings if you wish to upload images to the chat!

### 2. Ollama / llama.cpp
Run via terminal using `llama.cpp`:
```bash
./main -m "winter llm.gguf" --mmproj "mmproj-gemma-4-12B-it-QAT-BF16.gguf" -p "User: Describe this image: [image] Assistant:"