pynq-torch

Automated PyTorch wheel builds for Pynq-Z2 (ARMv7l) architectures. Builds are triggered by git tags and are available in the Releases page.

Supported Versions

When a tag (e.g., v2.10.0) is pushed, wheels are automatically built for:

• Python 3.10
• Python 3.11
• Python 3.12
• Python 3.13

Installation

Download the wheel for your Python version from the latest release and install it:

# Example for Python 3.12
pip install torch-X.Y.Z-cp312-cp312-linux_armv7l.whl

Warning

Pynq-Z2 has 512MB RAM. You MUST use at least 2GB swap space to avoid out-of-memory errors during usage.

Build Configuration

• Platform: linux/arm/v7
• Base Image: python:slim-bookworm
• Optimizations: NEON, VFPv4 (Hard Float)
• Flags:
  • NO_CUDA=1
  • NO_DISTRIBUTED=1
  • USE_QNNPACK=1
  • Use_PYTORCH_QNNPACK=1

To manually build locally:

docker build --platform linux/arm/v7 -t pynq-torch-builder .