Setup and System Requirements
This guide covers the setup requirements for running Cosmos Transfer 1 for warehouse multi-view inference.
System Requirements
Minimum Hardware Requirements
- GPU: Single A100, H100, or later GPU (recommended minimum)
- Memory: Sufficient VRAM for model inference
- Storage: Adequate space for model weights
Software Requirements
The setup requires both the Cosmos Transfer 1 repository and the Cosmos Cookbook to be properly installed and configured.
Installation
Cosmos Cookbook Setup
Then set up the Cosmos Cookbook repository and its dependencies by following the comprehensive setup tutorial:
The getting started guide provides detailed instructions for:
- Repository cloning and setup
- Environment configuration
- Essential tool installation (uv, Hugging Face CLI, AWS CLI, etc.)
- Development dependencies
Cosmos Transfer 1 Setup
Begin by installing and setting up the Cosmos Transfer 1 repository by following the official guide:
Cosmos Transfer 1 Installation Guide
The installation guide provides comprehensive steps for:
- Repository cloning and setup
- Environment configuration
- Model weight downloads
- Dependency installation
- Inference configuration
After completing the installation, copy the example cookbook assets into your Cosmos Transfer 1 repository. Assuming your Cosmos Transfer 1 repository root is available as $COSMOS_TRANSFER_ROOT, run:
Replace $COSMOS_TRANSFER_ROOT with the real path to the Cosmos Transfer 1 directory path, or export the environment variable with the path.
Install ffmpeg (required for video preprocessing/postprocessing)
Install ffmpeg into your active Conda environment:
Next Steps
Once the setup is complete, proceed to the inference tutorial to learn how to use Cosmos Transfer 1 for warehouse multi-view inference.