Winding Notes

TensorFlow Install

Dependencies and steps through which Tensorflow (TF) gets installed on your development machine.

Intro

As of Dec 2019, the instructions to install Tensorflow (TF) are here instructions and one should run:

$  pip install tensorflow

To learn more about pip see the (pip note)[https://winding-lines.github.io/notes/pip/,

wheels

For python 3.7 TF is provided through 3 files, one per operating system:

  • tensorflow-2.0.0-cp37-cp37m-macosx_10_11_x86_64.whl
  • tensorflow-2.0.0-cp37-cp37m-manylinux2010_x86_64.whl
  • tensorflow-2.0.0-cp37-cp37m-win_amd64.whl

The manylinux2010 is platform tag which allows binary wheels to be installed on compatible linux libraries. This is probably the reason why TF requires pip version 19, as per the changelog this platform support first appeared in this version.

Structure of the macos wheel

The macos wheel contains the following files:

Ext Files Largest file Size [kB]
.py 1510 tensorflow_core/python/ops/gen_nn_ops.py 738.56
.h 2219 tensorflow_core/include/Eigen/src/misc/lapacke.h 1033.56
.dylib 3 tensorflow_core/libtensorflow_framework.2.0.0.dylib 27937.71
.so 7 tensorflow_core/python/_pywrap_tensorflow_internal.so 321594.8
.inc 25 tensorflow_core/include/google/protobuf/test_util.inc 123.58
.c 75 tensorflow_core/include/external/boringssl/src/third_party/fiat/p256.c 80.71

These files are split amongst the following top-level folders:

  • tensorflow: just an __init__.py to load the actual code
  • tensorflow_core: most of the python and pre-compiled libraries
  • tensorflow-2.0.0.data:
  • tensorflow-2.0.0.dist-info: metadata required by the wheel

The wheel contains the following pre-compiled libraries:

  • tensorflow_core/python/_pywrap_tensorflow_internal.so, biggest file at 32MB
  • tensorflow_core/libtensorflow_framework.dylib
  • tensorflow_core/compiler/tf2tensorrt/_wrap_py_utils.so
  • tensorflow_core/compiler/tf2xla/ops/_xla_ops.so
  • tensorflow_core/lite/experimental/microfrontend/python/ops/_audio_microfrontend_op.so
  • tensorflow_core/lite/python/interpreter_wrapper/_tensorflow_wrap_interpreter_wrapper.so
  • tensorflow_core/lite/python/optimize/_tensorflow_lite_wrap_calibration_wrapper.so
  • tensorflow_core/python/framework/fast_tensor_util.so

It has the following outside dependencies:

  • absl-py (>=0.7.0)
  • astor (>=0.6.0)
  • gast (==0.2.2)
  • google-pasta (>=0.1.6)
  • keras-applications (>=1.0.8)
  • keras-preprocessing (>=1.0.5)
  • numpy (<2.0,>=1.16.0)
  • opt-einsum (>=2.3.2): optimize einsum chains
  • six (>=1.10.0)
  • protobuf (>=3.6.1)
  • tensorboard (<2.1.0,>=2.0.0)
  • tensorflow-estimator (<2.1.0,>=2.0.0)
  • termcolor (>=1.1.0)
  • wrapt (>=1.11.1)
  • grpcio (>=1.8.6)
  • wheel (>=0.26)
  • backports.weakref (>=1.0rc1); python_version < “3.4”
  • enum34 (>=1.1.6); python_version < “3.4”

Tensorflow internals: https://github.com/horance-liu/tensorflow-internals