Blog

1. Introduction PyTorch supports two execution modes : eager mode and graph mode. In eager…

tl;dr Multi-Objective Optimization in Ax enables efficient exploration of tradeoffs (e.g. between model performance and…

Introduction In recent years, scaling model sizes has become a promising area of research. In…

We are announcing TorchMultimodal Beta, a PyTorch domain library for training SoTA multi-task multimodal models…

On May 25, 2021, we announced the PyTorch Enterprise Support Program (ESP) that enabled providers to develop…

Note: A previous version of this post was published in November 2022. We have updated…

Summary We are bringing a number of improvements to the current PyTorch libraries, alongside the…

We are excited to announce the release of PyTorch® 1.13 (release note)! This includes Stable versions…

Introduction TL;DR: It can be challenging to run PyTorch on mobile devices, SBCs (Single Board…

Introduction The research community has witnessed a lot of successes with large models across NLP,…

1. Meta’s AI Performance Profiling (MAIProf) Figure 1: A simplified illustration of the Meta’s AI…

We are excited to announce that the PyTorch Conference returns in-person as a satellite event…

Today, I am proud to announce that PyTorch is moving to the Linux Foundation (LF) as a…

Beam search decoding with industry-leading speed from Flashlight Text (part of the Flashlight ML framework) is now available with…

nvFuser is a Deep Learning Compiler for NVIDIA GPUs that automatically just-in-time compiles fast and…

Overview Memory formats has significant impact on performance when running vision models, generally Channels Last…

TorchVision now supports listing and initializing all available built-in models and weights by name. This…

Overview Recent years, the growing complexity of AI models have been posing requirements on hardware…

In December, we announced PyTorch Live, a toolkit for building AI-powered mobile prototypes in minutes.…

Efficient training of modern neural networks often relies on using lower precision data types. Peak…