Setu

Global Tensor Exchange for Distributed Deep Learning Workloads

Setup

Option 1: Using VS Code Devcontainer

Setu now supports development using VS Code devcontainers, which provides a consistent, pre-configured development environment:

1. Install Docker and VS Code
2. Install the Dev Containers extension for VS Code
3. Clone the repository:

   git clone https://github.com/project-vajra/setu

4. Open the repository in VS Code
5. VS Code will detect the devcontainer configuration and prompt you to reopen the project in a container. Click "Reopen in Container".
6. If you want to use a subset of GPUs, update the --gpus flag in .devcontainer/devcontainer.json.
7. The devcontainer will set up the environment with all dependencies automatically
8. Use VS Code's built-in build tasks (Terminal > Run Build Task...) to easily run common Setu commands like build, test, and lint directly from the IDE

Option 2: Using Conda/Mamba

Prerequisites

• CUDA: Setu has been tested with CUDA 12.6 on A100 and H100 GPUs
• Git: For cloning the repository

Setup

First, ensure you have mamba installed,

# Install mamba (if not already available)
wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh
bash Miniforge3-Linux-x86_64.sh

Next clone the repository,

# Clone repository
git clone https://github.com/project-vajra/setu
cd setu

Now create a mamba environment,

# Create environment
make setup/environment

Now activate the environment,

# Activate environment
mamba activate ./env

Finally, build the project:

# Install dependencies and build
make build

Development

Building

# Intelligent build (automatically selects optimal strategy)
make build

# Advanced build options (for specific needs)
make build/native              # Native extension only (full build)
make build/native_incremental  # Incremental native build (fastest)
make build/editable            # Install editable package (legacy)
make build/wheel               # Build Python wheel

Testing

# Build tests intelligently (automatically detects what needs rebuilding)
make build/test

# Run all tests
make test

# Run specific test types
make test/unit           # All unit tests (Python + C++)
make test/integration    # Integration tests
make test/ctest          # C++ tests only

# Rerun only failed tests (for faster iteration)
make test/unit-failed-only     # Failed unit tests (Python + C++)
make test/pyunit-failed-only   # Failed Python unit tests only
make test/ctest-failed-only    # Failed C++ tests only

Code Quality

# Lint code
make lint

# Format code
make format

# Check system dependencies
make setup/check

Environment Management

# Create environment (conda)
make setup/environment

# Update environment
make setup/update-environment

Utilities

# Clean build artifacts
make clean

# Show ccache statistics
make ccache/stats

# Show all available targets
make help

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft’s Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies.