July 18, 2019
PyTorch Adds New Ecosystem Projects for Encrypted AI and Quantum Computing, Expands PyTorch Hub
The PyTorch ecosystem includes projects, tools, models and libraries from a broad community of researchers in academia and industry, application developers, and ML engineers. The goal of this ecosystem is to support, accelerate, and aid in your exploration with PyTorch and help you push the state of the art, no matter what field you are exploring. Similarly, we are expanding the recently launched PyTorch Hub to further help you discover and reproduce the latest research.
June 10, 2019
Towards Reproducible Research with PyTorch Hub
Reproducibility is an essential requirement for many fields of research including those based on machine learning techniques. However, many machine learning publications are either not reproducible or are difficult to reproduce. With the continued growth in the number of research publications, including tens of thousands of papers now hosted on arXiv and submissions to conferences at an all time high, research reproducibility is more important than ever. While many of these publications are a...
May 22, 2019
torchvision 0.3: segmentation, detection models, new datasets and more..
PyTorch domain libraries like torchvision provide convenient access to common datasets and models that can be used to quickly create a state-of-the-art baseline. Moreover, they also provide common abstractions to reduce boilerplate code that users might have to otherwise repeatedly write. The torchvision 0.3 release brings several new features including models for semantic segmentation, object detection, instance segmentation, and person keypoint detection, as well as custom C++ / CUDA ops sp...
May 08, 2019
Model Serving in PyTorch
PyTorch has seen a lot of adoption in research, but people can get confused about how well PyTorch models can be taken into production. This blog post is meant to clear up any confusion people might have about the road to production in PyTorch. Usually when people talk about taking a model “to production,” they usually mean performing inference, sometimes called model evaluation or prediction or serving. At the level of a function call, in PyTorch, inference looks something like this:
May 01, 2019
Optimizing CUDA Recurrent Neural Networks with TorchScript
This week, we officially released PyTorch 1.1, a large feature update to PyTorch 1.0. One of the new features we’ve added is better support for fast, custom Recurrent Neural Networks (fastrnns) with TorchScript (the PyTorch JIT) (https://pytorch.org/docs/stable/jit.html).
May 01, 2019
PyTorch adds new dev tools as it hits production scale
This is a partial re-post of the original blog post on the Facebook AI Blog. The full post can be viewed here
April 29, 2019
Stochastic Weight Averaging in PyTorch
In this blogpost we describe the recently proposed Stochastic Weight Averaging (SWA) technique [1, 2], and its new implementation in torchcontrib. SWA is a simple procedure that improves generalization in deep learning over Stochastic Gradient Descent (SGD) at no additional cost, and can be used as a drop-in replacement for any other optimizer in PyTorch. SWA has a wide range of applications and features:
