Installation¶

Prerequisites¶

Python 3.12
uv (recommended package manager)
CUDA-compatible GPU (for training)

Install with uv (Recommended)¶

We recommend using uv for environment management. Once uv is installed:

# Clone the repository
git clone https://github.com/TRI-ML/vla_foundry.git
cd vla_foundry

# Create environment and install dependencies
uv sync
uv pip install -e .

Running Scripts¶

The recommended workflow is to run scripts directly with uv:

uv run <script> <args>

Alternatively, activate the virtual environment:

source .venv/bin/activate

Note

Even when using the activated venv, prefer uv for package and dependency management.

Optional Dependency Groups¶

VLA Foundry organizes optional dependencies into groups for specific workflows:

PreprocessingSageMakerInference

uv sync --group=preprocessing

Required for data preprocessing scripts (Ray, img2dataset, etc.)

uv sync --group=sagemaker

Required for launching training jobs on AWS SageMaker.

uv sync --group=inference

Required for model inference and deployment.

Verify Installation¶

Run the essential test suite to verify everything is working:

uv run pytest tests/essential -v

Tip

If tests fail with Hugging Face errors, see the FAQ for troubleshooting HF token setup.

AWS Credentials Setup¶

If you need access to S3 datasets, configure AWS credentials. See the FAQ for detailed instructions.