https://github.com/BlackHC/tfpyth

Putting TensorFlow back in PyTorch, back in TensorFlow (differentiable TensorFlow PyTorch adapters).
https://github.com/BlackHC/tfpyth

machine-learning pytorch tensorflow

Last synced: about 2 months ago
JSON representation

Putting TensorFlow back in PyTorch, back in TensorFlow (differentiable TensorFlow PyTorch adapters).

• Host: GitHub
• URL: https://github.com/BlackHC/tfpyth
• Owner: BlackHC
• License: mit
• Created: 2019-07-05T15:08:50.000Z (about 5 years ago)
• Default Branch: master
• Last Pushed: 2020-11-30T19:46:39.000Z (almost 4 years ago)
• Last Synced: 2024-07-21T21:43:31.913Z (2 months ago)
• Topics: machine-learning, pytorch, tensorflow
• Language: Python
• Homepage:
• Size: 14.6 KB
• Stars: 641
• Watchers: 25
• Forks: 97
• Open Issues: 4
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# TfPyTh

[![Build Status](https://travis-ci.com/BlackHC/tfpyth.svg?branch=master)](https://travis-ci.com/BlackHC/tfpyth) [![codecov](https://codecov.io/gh/BlackHC/tfpyth/branch/master/graph/badge.svg)](https://codecov.io/gh/BlackHC/tfpyth)

> Putting TensorFlow back in PyTorch, back in TensorFlow (with differentiable TensorFlow PyTorch adapters).

Do you have a codebase that uses TensorFlow and one that uses PyTorch and want to train a model that uses both end-to-end?

This library makes it possible without having to rewrite either codebase! 

It allows you to wrap a TensorFlow graph to make it callable (and differentiable) through PyTorch, and vice-versa, using simple functions.

The only caveat is that tensors have to be copied and routed through the CPU until TensorFlow supports `__cuda_array_interface` (please star the [GitHub issue](https://github.com/tensorflow/tensorflow/issues/29039)).

## Install

```

pip install tfpyth

```

### Example

```python

import tensorflow as tf

import torch as th

import numpy as np

import tfpyth

session = tf.Session()

def get_torch_function():

    a = tf.placeholder(tf.float32, name='a')

    b = tf.placeholder(tf.float32, name='b')

    c = 3 * a + 4 * b * b

    f = tfpyth.torch_from_tensorflow(session, [a, b], c).apply

    return f

f = get_torch_function()

a = th.tensor(1, dtype=th.float32, requires_grad=True)

b = th.tensor(3, dtype=th.float32, requires_grad=True)

x = f(a, b)

assert x == 39.

x.backward()

assert np.allclose((a.grad, b.grad), (3., 24.))

```

## What it's got

### `torch_from_tensorflow`

Creates a PyTorch function that is differentiable by evaluating a TensorFlow output tensor given input placeholders.

### `eager_tensorflow_from_torch`

Creates an eager Tensorflow function from a PyTorch function.

### `tensorflow_from_torch`

Creates a TensorFlow op/tensor from a PyTorch function.

## Future work

- [ ] support JAX

- [ ] support higher-order derivatives