tensorflow-aarch64

TensorFlow wheels (whl) and docker images for aarch64 / ARMv8 / ARM64

GitHub | Web | Torch

Install

pip install tensorflow -f https://tf.kmtea.eu/whl/stable.html

To use linaro's build, replace tensorflow with tensorflow-aarch64 for newer versions (v2.6.0 and later), and tensorflow-cpu for older versions.

Backup link: pip install tensorflow -f https://cf.tf.kmtea.eu/whl/stable.html

To pick the whl files manually, please check the releases.

---

Build

Please check the tutorial.

---

More Info

Note on v2.5.0

Custom build on v2.5.0 always failed, so this version has switched to linaro's build, which might be more stable.

You can still fetch the archived wheels here.

RuntimeError (v2.4)

If you see RuntimeError: module compiled against API version 0xe but this version of numpy is 0xd, it is because tensorflow requires numpy~=1.19.2 but I built it using numpy~=1.20.2.

You may try pip install -U numpy.

Building Environment

Host: Raspberry Pi 4 Model B

SoC: BCM2711 (quad-core A53)

Architecture: ARMv8 / ARM64 / aarch64

OS: CentOS 7

GCC: v8.3.0

Virtualization: Docker

Linaro's wheels

There are official wheels for Python 3.6 - 3.9, you could fetch them here.