executorch

ExecuTorch

Executorch is a PyTorch inference API for Edge Devices.

Instructions

A prebuilt tar is included.

Build

To build executorch you will require cmake and python3. You will also require torch.

# get executorch git
git clone --branch viable/strict https://github.com/pytorch/executorch.git /tmp/executorch
cd /tmp/executorch && git submodule sync && git submodule update --init

# copy missing files (monkey patch)
# We require these files for building execu
ls /usr/local/lib/python3.10/dist-packages/torchgen/packaged/ATen/native/
# This is a workaround for now
mkdir -p /usr/lib/python3/dist-packages/torchgen/packaged/ATen/native/
cp /usr/local/lib/python3.10/dist-packages/torchgen/packaged/ATen/native/* /usr/lib/python3/dist-packages/torchgen/packaged/ATen/native/

Run this Python Code

import sysconfig
lib_path = sysconfig.get_paths()["purelib"]
print(lib_path)

Copy the LIB PATH into this CMAKE_PREFIX_PATH value

cd /tmp/executorch; CMAKE_PREFIX_PATH={lib_path} EXECUTORCH_BUILD_XNNPACK=ON bash ./install_executorch.sh --pybind xnnpack; echo "build completed"

Wait for build to complete.