PyTorch Adds New APIs
Written by Kay Ewbank   
Tuesday, 10 November 2020

PyTorch has been updated with several new APIs including support for NumPy-Compatible FFT operations, profiling tools and major updates to both distributed data parallel (DDP) and remote procedure call (RPC) based distributed training.

PyTorch is an optimized tensor library for deep learning using GPUs and CPUs. It aims to offer a replacement for NumPy that make use of the power of GPUs, while providing a deep learning research platform that provides maximum flexibility and speed.


There are several new front end APIs, all in beta format. These include torch.fft, which  implements FFT-related functions with the same API as NumPy. PyTorch has historically supported a few FFT-related functions, but the new API adds more functions for use in scientific fields like signal processing.

A second API adds C++ support for nn.transformer modules and allows developers to use the nn.transformer module abstraction from the C++ Frontend. It also means you no longer need to save a module from python/JIT and load into C++ as it can now be used it in C++ directly.

The third new API is torch.set_deterministic. This may help identify errors when debugging or testing a program. Its function directs PyTorch operators to select deterministic algorithms when available, and to throw a runtime error if an operation may result in nondeterministic behavior.

Along with the new APIs, several features have been moved to a 'stable' level, including custom C++ Classes, the memory profiler, extensions via custom tensor-like objects, user async functions in RPC and a number of other features in torch.distributed.

The developers say that one highlight is that CUDA 11 is now officially supported with binaries available at This release also includes updates and additions to profiling and performance for RPC, TorchScript and Stack traces in the autograd profiler. Elsewhere, Transforms now support Tensor inputs, batch computation, GPU, and TorchScript, and PyTorch now has native image I/O for JPEG and PNG formats.


More Information

PyTorch Website

PyThorch On GitHub

Related Articles

PyTorch Scholarship Challenge

Pyro Now On Watson Machine Learning

More Efficient Style Transfer Algorithm

ONNX For AI Model Interoperability

Microsoft Cognitive Toolkit Version 2.0

NVIDA Updates Free Deep Learning Software

TensorFlow - Googles Open Source AI And Computation Engine

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin.


OpenSilver 2.2 Adds LightSwitch Compatibility Pack

OpenSilver 2.2 has been released with the addition of a LightSwitch Compatibility Pack designed to provide a way to run legacy Visual Studio LightSwitch applications on modern browsers. The open-sourc [ ... ]

Digital Play Shown To Be Good For Kids

When designed with their needs in mind, video games can benefit children’s well-being. This finding comes research from  UNICEF in partnership with LEGO and the University of Sheffield.

More News

raspberry pi books



or email your comment to:

Last Updated ( Tuesday, 10 November 2020 )