SpectraFormer

Installation ​

This guide covers how to install SpectraFormer for both inference and training.

Requirements ​

  • Python 3.11
  • CUDA 12 (optional, for GPU acceleration)

Quick Install ​

bash
# Clone the repository (--depth 1 for faster download)
git clone --depth 1 https://github.com/pietronvll/SpectraFormer.git
cd SpectraFormer

# Install with uv (recommended)
uv sync

# Or with pip
pip install -e .

GPU Support ​

For GPU-accelerated inference and training with CUDA 12:

bash
uv sync --extra cuda12

Optional Dependencies ​

SpectraFormer provides optional dependency groups for different use cases:

Training ​

Install additional dependencies needed for model training:

bash
uv sync --extra train

This includes:

  • tensorboardX for logging
  • gpustat for GPU monitoring

Dashboard ​

Install the Streamlit dashboard for interactive visualization:

bash
uv sync --extra dashboard

Development ​

Install development tools:

bash
uv sync --extra dev

All Dependencies ​

Install everything:

bash
uv sync --extra all

Verify Installation ​

After installation, verify that SpectraFormer is correctly installed:

bash
# Check if the CLI is available
spectraformer-unmix --help

For GPU users, verify JAX can see your GPU:

python
import jax
print(jax.devices())
# Should show your GPU(s), e.g., [cuda(id=0)]

Troubleshooting ​

JAX Platform Errors ​

If you encounter JAX platform errors, you can explicitly set the platform:

bash
# Force CPU
JAX_PLATFORMS=cpu spectraformer-unmix ...

# Force GPU
JAX_PLATFORMS=cuda spectraformer-unmix ...

CUDA Version Mismatch ​

Ensure your CUDA version matches JAX's requirements. For CUDA 12 support:

bash
uv sync --extra cuda12

Python Version ​

SpectraFormer requires Python 3.11. Check your version:

bash
python --version