Hidream I1 Full Nf4

🚀 HiDream-I1 4Bit Quantized Model

This repository offers a 4-bit quantized version of HiDream-I1, enabling the full model to operate with less than 16GB of VRAM.

The original repository can be accessed here.

HiDream-I1 is a novel open-source image generative foundation model with 17B parameters, capable of achieving state-of-the-art image generation quality within seconds.

🚀 Quick Start

Installation

Simply run the following command to install the package:

pip install hdi1 --no-build-isolation

⚠️ Important Note

It's advisable to create a new Python environment for this package to prevent dependency conflicts. You can use conda create -n hdi1 python=3.12 and then conda activate hdi1. Alternatively, you can use python3 -m venv venv and then source venv/bin/activate on Linux or venv\Scripts\activate on Windows.

Usage

Command Line Interface

After installation, you can generate images using the following commands:

python -m hdi1 "A cat holding a sign that says 'hello world'"

# or you can specify the model
python -m hdi1 "A cat holding a sign that says 'hello world'" -m fast

⚠️ Important Note

The inference script will attempt to automatically download the meta-llama/Llama-3.1-8B-Instruct model files. You must agree to the license of the Llama model on your Hugging Face account and log in using huggingface-cli login to use the automatic downloader.

Web Dashboard

We also offer a web dashboard for interactive image generation. Start it with the following command:

python -m hdi1.web

✨ Features

We provide both the full version and distilled models. Although they have the same parameter size and thus require the same amount of GPU memory, the distilled models are faster due to the reduced number of inference steps.

📦 Hardware Requirements

• GPU Architecture: NVIDIA >= Ampere (e.g., A100, H100, A40, RTX 3090, RTX 4090)
• GPU RAM: >= 16 GB
• CPU RAM: >= 16 GB

📄 License

The code in this repository and the HiDream-I1 models are licensed under the MIT License.