Installation Guide

Quick Installation

Install TensorFlow using our custom wheels for optimized performance on various architectures:

pip install tensorflow==2.8.3 -f https://diyor28.github.io/wheels

Supported Versions

Ensure compatibility with these supported versions before installation:

• Python: 3.7, 3.8, 3.9, 3.10, 3.11
• TensorFlow: 2.7, 2.8, 2.9, 2.10
• TensorFlow Data Validation (TFDV): 1.4 to 1.14

Note:

• TensorFlow versions 2.11 and above provide prebuilt wheels for aarch64.
• TFDV 1.14 can be manually built, but a prebuilt wheel is currently not available.
• Custom builds for other versions can be requested.

Docker Images

Use our prebuilt Docker images available on Docker Hub:

FROM diyor28/tensorflow:2.7.3-py37
RUN python -c "import tensorflow; print(tensorflow.__version__)"

All images are based on the official Python Docker images.

TFDV Docker images will be available soon.

Web Server Setup with Docker

Basic Authentication

Create a directory and password for basic authentication:

mkdir basic_auth
htpasswd -c basic_auth/.htpasswd admin

Workbench Directory

Set up a directory to serve as a workbench:

mkdir /tmp/tf_aarch64

Configuration

Configure environment settings such as build concurrency and caching:

1. Edit the environment settings:

   vim .env

2. Update or add the following configurations:

   BUILDER_THREADS=1  # Max of 2 for systems with <32GB RAM
   USE_CACHE=True     # Disable if using BuildKit or if caching is not desired

3. Start the Docker container:

   docker compose up -d --build

Visit http://localhost to access the web interface, which includes version selection tools for TensorFlow and TFX.

Development Setup

Python Backend Server

cd back/
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

UI Server

cd front/
npm install
npm run serve

Manual Wheel Building

Generate Dockerfile

Create a Dockerfile for building a specific TensorFlow version:

python gen.py tensorflow -v 2.7.3 -py 3.7 ./Dockerfile_tf

Build Docker Image

Build the Docker image based on the generated Dockerfile:

docker build -t tensorflow:2.7.3-py3.7 -f ./Dockerfile_tf ./build_templates/context/

Extract Wheels

Copy the built TensorFlow wheels from the Docker container to your host machine:

docker run -v /host/machine/path:/builds tensorflow:2.7.3-py3.7 cp -a /wheels/. /builds

Future Updates

Details on adding and building custom versions will be provided soon.