Skip to main content

EagerPy is a thin wrapper around PyTorch, TensorFlow Eager, JAX and NumPy that unifies their interface and thus allows writing code that works natively across all of them.

Project description

EagerPy: Writing Code That Works Natively with PyTorch, TensorFlow, JAX, and NumPy

EagerPy is a Python framework that lets you write code that automatically works natively with PyTorch, TensorFlow, JAX, and NumPy. EagerPy is also great when you work with just one framework but prefer a clean and consistent API that is fully chainable, provides extensive type annotions and lets you write beautiful code.

🔥 Design goals

  • Native Performance: EagerPy operations get directly translated into the corresponding native operations.

  • Fully Chainable: All functionality is available as methods on the tensor objects and as EagerPy functions.

  • Type Checking: Catch bugs before running your code thanks to EagerPy’s extensive type annotations.

📖 Documentation

Learn more about EagerPy in the documentation.

🚀 Quickstart

pip install eagerpy

EagerPy requires Python 3.6 or newer. Besides that, all essential dependencies are automatically installed. To use it with PyTorch, TensorFlow, JAX, or NumPy, the respective framework needs to be installed separately. These frameworks are not declared as dependencies because not everyone wants to use and thus install all of them and because some of these packages have different builds for different architectures and CUDA versions.

🎉 Example

import torch
x = torch.tensor([1., 2., 3., 4., 5., 6.])

import tensorflow as tf
x = tf.constant([1., 2., 3., 4., 5., 6.])

import jax.numpy as np
x = np.array([1., 2., 3., 4., 5., 6.])

import numpy as np
x = np.array([1., 2., 3., 4., 5., 6.])

# No matter which framwork you use, you can use the same code
import eagerpy as ep

# Just wrap a native tensor using EagerPy
x = ep.astensor(x)

# All of EagerPy's functionality is available as methods
x = x.reshape((2, 3))
# or just: x.flatten(1).norms.l2()

# and as functions (yes, gradients are also supported!)
loss, grad = ep.value_and_grad(loss_fn, x)
ep.clip(x + eps * grad, 0, 1)

# You can even write functions that work transparently with
# Pytorch tensors, TensorFlow tensors, JAX arrays, NumPy arrays

def my_universal_function(a, b, c):
    # Convert all inputs to EagerPy tensors
    a, b, c = ep.astensors(a, b, c)

    # performs some computations
    result = (a + b * c).square()

    # and return a native tensor
    return result.raw

🗺 Use cases

Foolbox Native, the latest version of Foolbox, a popular adversarial attacks library, has been rewritten from scratch using EagerPy instead of NumPy to achieve native performance on models developed in PyTorch, TensorFlow and JAX, all with one code base.

EagerPy is also used by other frameworks to reduce code duplication (e.g. GUDHI) or to compare the performance of different frameworks.

📄 Citation

If you use EagerPy, please cite our paper using the this BibTex entry:

  title={{EagerPy}: Writing Code That Works Natively with {PyTorch}, {TensorFlow}, {JAX}, and {NumPy}},
  author={Rauber, Jonas and Bethge, Matthias and Brendel, Wieland},
  journal={arXiv preprint arXiv:2008.04175},

🐍 Compatibility

We currently test with the following versions:

  • PyTorch 1.4.0

  • TensorFlow 2.1.0

  • JAX 0.1.57

  • NumPy 1.18.1

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

eagerpy-0.30.0.tar.gz (23.0 kB view hashes)

Uploaded source

Built Distribution

eagerpy-0.30.0-py3-none-any.whl (31.2 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page