---
title: ForgeAI
emoji: "🔥"
colorFrom: purple
colorTo: red
sdk: static
pinned: false
---

# ForgeAI

**Making AI models accessible for consumer hardware.**

We quantize large AI models, fix compatibility issues, and document everything so the community can benefit.

## What We Do

- **Quantize large models** to run on GPUs with 16 GB VRAM or less
- **Fix compatibility issues** with modern PyTorch, transformers and torchtune versions
- **Provide ComfyUI integration** for creative AI workflows
- **Document all fixes** with copy-paste code examples

## Our Models

| Model | What it does | VRAM | Link |
|---|---|---|---|
| **HeartMuLa-3B-4bit** | AI music generation from lyrics | 16 GB | [Download](https://huggingface.co/PavonicAI/HeartMuLa-3B-4bit) |

## Tech Stack

- PyTorch + CUDA
- bitsandbytes (NF4 / INT8 quantization)
- Hugging Face Transformers
- ComfyUI custom nodes

## Get In Touch

Open an issue or discussion on any of our model repos.