1642739890_maxresdefault.jpg
12
09:43

WHY AND HOW OF SCALING LARGE LANGUAGE MODELS | NICHOLAS JOSEPH

Anthropic is an AI safety and research company that’s working to build reliable, interpretable, and steerable AI systems. Over the past decade, the amount of compute used for the largest training runs has increased at an exponential pace. We’ve also seen in many domains that larger models are able to attain better performance following precise […]
1642653389_maxresdefault.jpg
19
16:54

PYTORCH, TODAY AND THE FUTURE | LIN QIAO

In this keynote, Lin Qiao (Engineering Director, Meta AI) discusses how the PyTorch team has focused on building features and libraries to accelerate the speed of iteration of this wisdom cycle in the past year. She also goes into detail about continuous obsessions on improving usability and empowering community collaborations. Lin further shares her vision […]
1642393924_maxresdefault.jpg
23
16:28

REAL-WORLD RESEARCH TO PRODUCTION AT FIDELITY | AUSTIN HUANG

Learn first-hand about real-world approaches to taking machine learning from research to production at Fidelity. In this talk, Austin Huang (Vice President, AI & Machine Learning, Fidelity) explains how machine learning use cases have changed – evolving from batch prediction pipelines to real-time consumers of unstructured data. These use cases have also given rise to […]
1642307176_hqdefault.jpg
4
09:20

TUTEL-MoE-STACK OPTIMIZATION FOR MODERN DISTRIBUTED TRAINING | RAFAEL SALAS & YIFAN XIONG

The Mixture-of-Experts (MoE) is a sparsely activated deep learning model architecture that has sublinear compute costs with respect to their parameters. MoE is one of the few scalable approaches for training trillion-parameter scale deep learning models. This talk will present Tutel, an open-source project built with the Pytorch framework. Tutel is being actively developed by […]
1642220727_maxresdefault.jpg
4
10:41

MODEL SERVING IN PYTORCH | GEETA CHAUHAN

Deploying ML models in Production and scaling your ML services still continue to be big challenge. TorchServe, the model serving solution for PyTorch solves this problem and has now evolved into a multi-platform solution that can run on-prem or on any cloud with integrations for major OSS platforms like Kubernetes, MLflow, Kubeflow Pipelines, KServe. This […]
1642059668_maxresdefault.jpg
29
29:08

Proximal Policy Optimization is Easy with Tensorflow 2 | PPO Tutorial

Proximal Policy Optimization (PPO) has emerged as a powerful on policy actor critic algorithm. You might think that implementing it is difficult, but in fact tensorflow 2 makes coding up a PPO agent relatively simple. We’re going to take advantage of my PyTorch code for this, as it serves as a great basis to expand […]
1642047772_maxresdefault.jpg
18
09:14

STATE OF PYTORCH 2021 | DMYTRO DZHULGAKOV

It takes a village to build an open-source framework. PyTorch has added many great features and has seen explosive growth in 2021, all thanks to our awesome community of contributors. In this talk, Dmytro Dzhulgakov (Software Engineer, Meta AI) gives a run through of who builds PyTorch, new and upcoming improvements to the framework and […]
1641961330_maxresdefault.jpg
4
08:18

TORCHX: PRODUCTION ML WITH PYTORCH | KIUK CHUNG

TorchX is an SDK for quickly building and deploying ML applications from R&D to production. It offers various built-in components that encode MLOps best practices and make advanced features like distributed training and hyperparameter optimization accessible to all. In this talk, Kiuk Chung (Software Engineer, Meta AI) explains what TorchX is and how it facilitates […]
1641874792_maxresdefault.jpg
10
09:00

FUNCTORCH | RICHARD ZOU & HORACE HE

functorch (https://github.com/pytorch/functorch) is a library of JAX-like composable function transforms for PyTorch. It aims to provide composable vmap (batching) and grad (autodiff) transforms that work with PyTorch modules and PyTorch autograd with good performance with or without a JIT compiler. In this talk, Richard Zou (Software Engineer, Meta AI) and Horace He (Software Engineer, Meta […]
1641788278_maxresdefault.jpg
5
11:20

RADIOLOGY AI @MARS VETERINARY HEALTH | MICHAEL FITZKE

The global veterinary imaging market size was valued at 2.01 billion in 2018 and increase in utilization for veterinary diagnostics is expected to be driven largely by rising demand for pet insurance and growing animal healthcare expenditure, increasing companion animal population, and growth in the number of veterinary practitioners globally. Currently, while medical imaging use […]
1641701798_maxresdefault.jpg
3
07:04

TORCHAUDIO | MOTO HIRA

TorchAudio is the building blocks for research and production. It provides powerful audio i/o functions, preprocessing transforms and dataset to provide a seamless path from research prototyping to production deployment with GPU support. In this talk, Moto Hira (Software Engineer, Meta AI) discusses the most recent releases on TorchAudio, the growing community, and a look […]
1641615321_maxresdefault.jpg
1
05:29

PYTORCH ON iOS | TAO XU

PyTorch Mobile aims to combine a best-in-class experience for ML developers with high-performance execution on all mobile hardware. The support for Core ML is essential to meeting that goal since it is the only way to leverage Apple’s Neural Engine on iOS devices. The Core ML delegate is a prototype feature released in PyTorch for […]
1641528854_maxresdefault.jpg
2
13:26

EAGER IN PRODUCTION | MICHAEL SUO

With TorchScript, PyTorch provides ease-of-use and flexibility in eager mode, while seamlessly transitioning to graph mode for speed, optimization, and functionality in C++ runtime environments. In this talk, Michael Suo (Software Engineer, Meta AI) covers our journey enabling PyTorch eager mode execution in production at Meta. He’ll also discuss the technical challenges involved in packaging, […]
1641442417_maxresdefault.jpg
9
08:31

OPACUS | ALEX SABLAYROLLES & IGOR SHILOV

Opacus is a library that enables training PyTorch models with differential privacy. It supports training with minimal code changes required on the client, has little impact on training performance and allows the client to online track the privacy budget expended at any given moment. In this talk, Igor Shilov (Research Engineer, Meta AI) and Alex […]
1641408884_maxresdefault.jpg
13
14:54

Basic Hyperparameter Tuning in DeepMinds ACME Framework

In today’s ACME deep reinforcement learning framework tutorial, I will showy ou how to do some basic hyperparameter tuning in their built in Deep Q Learning agent. Learn how to turn deep reinforcement learning papers into code: Deep Q Learning: https://www.udemy.com/course/deep-q-learning-from-paper-to-code/?couponCode=DQN-AUG-2021 Actor Critic Methods: https://www.udemy.com/course/actor-critic-methods-from-paper-to-code-with-pytorch/?couponCode=AC-AUG-2021 Natural Language Processing from First Principles: https://www.udemy.com/course/natural-language-processing-from-first-principles/?couponCode=NLP1-AUG-2021 Reinforcement Learning Fundamentals […]
1641355987_maxresdefault.jpg
5
10:25

MODEL INTERPRETABILITY WITH CAPTUM | NARINE KOKHLIKYAN

In this talk, Narine Kokhlikyan (Research Scientist, Artificial Intelligence, Meta AI) discusses further developments of the Captum library beyond feature attributions. She dives deeper into adversarial perturbations, counterfactuals, concept-based model interpretability and shows how those approaches can be used together to increase models’ transparency and trust. Narine also demonstrates different applications of concept-based model interpretability […]
1641278680_maxresdefault.jpg
22
25:46

Getting Started with Encryption in 2022

When you think of encryption you probably think of highly comjplex algorithms like SHA-256, but you can actually get unbreakable encryption with only a few lines of python. We’ll see how in this tutorial. We are going to cover 3 of the fundamental algorithms in encryption: the Caesar cipher, the Vignere cipher, and the one […]
1641269250_maxresdefault.jpg
8
06:00

SCIENTIFIC COMPUTING IN PYTORCH | MIKE RUBERRY

Scientific computing helps us understand the natural world, and PyTorch lets you research, create, and quickly run the latest deep learning and scientific computing programs available today. In this talk, Mike Ruberry (Software Engineer, Meta AI) discusses how PyTorch has improved its support for scientific computing over the past year, with a focus on support […]
1641182808_maxresdefault.jpg
7
09:51

INTRODUCING PYTORCH LIVE | RAZIEL ALVAREZ GUEVARA & ROMAN RÄDLE

PyTorch’s mission is to accelerate the path from research prototyping to production deployment. With the growing mobile ML ecosystem, this has never been more important than before. With the aim of helping reduce the friction for mobile developers to create novel ML-based solutions, we introduce PyTorch Live: a tool to build, test and (in the […]
1641096096_maxresdefault.jpg
8
12:08

PROFILING AND OPTIMIZING PYTORCH APPLICATIONS WITH THE PYTORCH PROFILER | SABRINA SMAI

PyTorch Profiler is a tool that allows the collection of the performance metrics during the training and inference. The Profiler’s context API can be used to better understand what model operators are the most expensive, examine their input shapes and stack traces, study device kernel activity and visualize the execution trace. In this talk, Sabrina […]
1641083607_maxresdefault.jpg
45
26:44

How to Code RL Agents Like DeepMind

DeepMind is known for leading the way in deep reinforcement learning research. Creating novel agents to conquer the most advanced environments requires the use of some sophisticated infrastructure. Fortunately for us mere mortals, they’ve open sourced their framework for designing deep reinforcement learning agents: ACME. In ACME, you’ll find everything from deep Q learning all […]
1641009654_maxresdefault.jpg
10
12:11

THE TRITON LANGUAGE | PHILIPPE TILLET

Triton is a new language that enables researchers with no CUDA experience to write highly efficient GPU code-most of the time on par with what an expert would be able to produce. Since its inception as a graduate research project, Triton has helped provide fast and maintainable kernels across several organizations (e.g., OpenAI, Facebook, Microsoft). […]