Instructions for compilation

Install right version of bazel:

wget https://releases.bazel.build/3.7.2/release/bazel-3.7.2-installer-linux-x86_64.sh
chmod +x bazel-3.7.2-installer-linux-x86_64.sh
sudo ./bazel-3.7.2-installer-linux-x86_64.sh
source /usr/local/lib/bazel/bin/bazel-complete.bash

Install python dependencies:

pip install numpy==1.19.5 
pip install keras_preprocessing==1.1.2

Build package tool (the actual tensorflow):

bazel build //tensorflow/tools/pip_package:build_pip_package

Build the python wheel NOTE: wheel will be placed in the folder, change it to a valid path

./bazel-bin/tensorflow/tools/pip_package/build_pip_package <dst-folder>

After compilation

Install python wheel (move to another folder or create a virtual environment )

python3 -m venv tf_env
source tf_env/bin/activate
pip install protobuf==3.20.1
pip install <wheel-file>

Use tensorposit in python scripts:

# This may generate some warnings, you can ignore it
import tensorflow as tf
from tensorflow.python import keras

# Set default floatx
keras.backend.set_floatx("posit160")

# Or create a tensor of posits
tp = tf.ones((10,10), dtype=tf.posit160)

NOTE: posit160 is just a name for the internal posit type (I hope to change it to "posit" in the near future). The default posit configuration is Posit16,2.

In the releases you can find prebuilt Posit16,2 and Posit8,2 wheels as separate installation. In the future I may want to put them together as posit16 and posit8 dtypes inside tensorflow.