---
license: apache-2.0
tags:
  - gabforge
  - vision
  - coding
  - screenshot-to-code
  - qwen3.5
  - gguf
base_model: Qwen/Qwen3.5-9B
pipeline_tag: image-text-to-text
---

# GabForge Mini v1 — Vision + Coding GGUF

The first GabForge model: a fine-tuned **Qwen3.5-9B** with built-in vision capability, optimized for screenshot→code generation.

## Model Details

| Property | Value |
|----------|-------|
| Base Model | Qwen3.5-9B (multimodal) |
| Fine-tuning | QLoRA (rank 32), 7,419 WebSight screenshot→code examples |
| Training | 2.15 epochs, final loss 0.15 |
| Quantization | Q4_K_M |
| File Size | 5.3 GB |
| Min VRAM | 8 GB |
| License | Apache 2.0 |

## Capabilities

- **Screenshot→Code**: Give it a UI screenshot, get HTML/CSS/JS back
- **General Coding**: Inherits Qwen3.5-9B's strong coding ability
- **Vision Understanding**: Reads UI layouts, diagrams, charts, error screenshots
- **Chat**: Standard instruction-following conversational model

## Usage

Works with any llama.cpp-compatible inference engine:

```bash
# With llama-server
llama-server -m GabForge-Mini-v1-Q4_K_M.gguf --port 8766

# With GabForge AI Studio (automatic)
# Download via Settings → AI Models → Local tab
```

## Made for GabForge AI Studio

This model is the default local model in [GabForge AI Studio](https://gabforge.ai) — the privacy-first AI coding IDE. Vision runs entirely on your machine.

## Training Data

- [HuggingFaceM4/WebSight](https://huggingface.co/datasets/HuggingFaceM4/WebSight) — screenshot→HTML/CSS pairs
- Additional coding data from Qwen3.5-9B's base knowledge