---
title: Grug-9B Demo
emoji: 🪨
colorFrom: red
colorTo: yellow
sdk: gradio
sdk_version: 6.15.1
app_file: app.py
short_description: Vision-language reasoning model with concise thinking
python_version: "3.12"
startup_duration_timeout: 1h
---

# 🪨 Grug-9B Vision-Language Demo

A Gradio demo for [ProCreations/grug-9b](https://huggingface.co/ProCreations/grug-9b) —
a 9B-parameter reasoning VLM fine-tuned from Ornith-1.0-9B (Qwen3.5 architecture)
that produces concise reasoning traces ("grug think small") instead of verbose chain-of-thought.

## Features

- **Image + text input**: Upload an image and ask questions about it
- **Thinking mode**: Toggle the model's reasoning traces (default on)
- **Adjustable generation**: Max tokens, temperature, top-p sampling

## Model details

- Architecture: Qwen3.5 (`Qwen3_5ForConditionalGeneration`)
- Parameters: ~9.4B (dense, bf16)
- Base model: [deepreinforce-ai/Ornith-1.0-9B](https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B)
- License: MIT