TensorFlow Lite C++ for Raspberry Pi and x86 Platforms

Precompiled TensorFlow Lite 2.18.0 binaries for Raspberry Pi 3, 4 (64-bit) and x86 (64-bit) systems.

Supported Features

• NEON optimization (Raspberry Pi)
• XNNPACK delegate (Raspberry Pi, x86)

Supported Platforms

Raspberry Pi

• Boards:
  • Raspberry Pi 3 Model A+
  • Raspberry Pi 3 Model B+
  • Raspberry Pi 4 Model B
  • Raspberry Pi 5 (Not tested)
• OS:
  • Raspberry Pi OS Bookworm 64-bit
• Tested On:
  • Raspberry Pi 3 Model B+
  • Raspberry Pi 4 Model B (8 GB)

x86

• Note:
  • No support for avxnnint8. This flag is disabled as it causes build failure of the library.
• Tested On:
  • x86 64-bit machine running Ubuntu 22.04.

Installation

Download and install the appropriate .deb package for your platform:

Raspberry Pi (64-bit)

wget https://github.com/tflite-build/releases/latest/download/tensorflowlite-elinux_aarch64.deb
sudo apt install -y ./tensorflowlite-elinux_aarch64.deb

x86 (64-bit)

wget https://github.com/tflite-build/releases/latest/download/tensorflowlite-amd64.deb
sudo apt install -y ./tensorflowlite-amd64.deb

Uninstallation

To uninstall TensorFlow Lite, run:

sudo apt purge --autoremove -y tensorflowlite

Debian Package Contents

The debian package contains the following:

Library	Description
libtensorflowlite.so	C++ API to access TensorFlow Lite interpreter and perform an inference
Header Files	Includes headers for TensorFlow Lite, enabling development with C++ APIs

Reference

1. TensorFlow Lite repository
2. Bazel