Cosmos 1.0 Prompt Upsampler 12 B Text2 World LLM Model

Introduction

Cosmos-1.0-Prompt-Upsampler-Text2World is a large language model created by NVIDIA, designed to enhance text prompts by adding detail, improving coherence, and structuring descriptions consistently for text-to-world applications. It supports commercial use and aims to produce higher quality outputs by refining input prompts.

Architecture

The model is based on a Transformer architecture, specifically utilizing the Mistral-NeMo-12B-Instruct network. It accepts text inputs of up to 512 tokens and generates similarly constrained text outputs. It is compatible with NVIDIA's Ampere and Hopper microarchitectures and runs on Linux operating systems.

Training

The model was trained using NVIDIA's proprietary systems and methodologies. The initial release, version 1.0, leverages the Cosmos framework and integrates with NVIDIA's NeMo library for model development and deployment.

Guide: Running Locally

Environment Setup: Ensure you have Python and necessary dependencies installed. Clone the Cosmos repository from GitHub.
```
git clone https://github.com/NVIDIA/Cosmos
cd Cosmos
```

Install Dependencies: Use a virtual environment to manage dependencies.

python3 -m venv env
source env/bin/activate
pip install -r requirements.txt

Model Download: Download the model weights from Hugging Face.

git lfs install
git clone https://huggingface.co/nvidia/Cosmos-1.0-Prompt-Upsampler-12B-Text2World

Run the Model: Use the provided scripts in the repository to run the model with your inputs.
```
python run_model.py --input "Your input text here"
```

Cloud GPUs: To handle the computational demands, consider using cloud services that offer NVIDIA GPUs, such as AWS EC2 with NVIDIA GPUs or Google's Cloud Platform.

License

The Cosmos-1.0-Prompt-Upsampler-Text2World model is released under the NVIDIA Open Model License. The license allows for commercial use, creation, and distribution of derivative models. However, it enforces compliance with certain ethical guidelines and technical limitations. Bypassing these limitations results in automatic termination of license rights. For custom licensing arrangements, contact cosmos-license@nvidia.com.

More Related APIs