优化 ARM CPU 上的专家混合 LLM 推理
大家好，我是来自英特尔印度研究院的 Majid，我很高兴能在今年 10 月的 PyTorch 北美会议上发表演讲。我将讨论使用 VLLM 和 OpenVINO 的专家 LLM 推理的有效组合，以及我们如何在 ARM CPU 上高效运行这些模型。如果…
YouTube — PyTorch Official
Video
1596 статей
不断发展的 AI Argo 工作流程：新功能、调整和多租户隔离 - S. Tian 和 Y. Su
不断发展的 AI Argo 工作流程：新功能、调整和多租户隔离 - S. Tian 和 Y. Su
YouTube — PyTorch Official
项目闪电演讲：KubeEdge Everywhere：最新项目更新和工业案例 - H.Zhang
项目闪电演讲：KubeEdge Everywhere：最新项目更新和工业案例 - H.Zhang
YouTube — PyTorch Official
ENSelf-Healing Rollouts: Automating Production Fixes With Agentic AI - Kevin Dubois & Daniel Oh
All right. Uh, thanks for joining today and um, so my name is Daniel uh, from Redhead and also IBM from Boston. It's a long way. Uh, we got here a couple days ago and then but we still a little bit jet lag. Anyway, so so we're going to talk about little bit about the…
YouTube — PyTorch Official
ENIntegrating RCCL GPU-Initiated Networking into TorchTitan’s MoE Communication
Hi, I'm Rishi from AMD. Within distributed training, communication is one of the primary bottlenecks, and limiting the time spent on communication can massively boost end-to-end performance. We investigate this hypothesis by integrating work from the Rickle team to use…
YouTube — PyTorch Official
ENInside the PyTorch Ambassador Model: How Ambassadors Grow PyTorch Adoption & How You Can Become One
Hello everyone, I'm Sadiq Allah. I'm a PyTorch Ambassador and an open source developer at IBM. I'll be at the PyTorch conference 2026 in San Jose and I'm excited about many things. About learning, meeting fellow ambassadors, catching up with friends in the community and…
YouTube — PyTorch Official
ENWhat You Cannot Profile, You Cannot Optimize: Learning to Read PyTorch Traces
Hello, my name is Souvik and I work at hugging face. I'm going to be at the PyTorch conference 2026 and I will be talking about how you can get started with using the PyTorch profiler on your models. And how you can extract the maximum efficiency from them. Come join us on…
YouTube — PyTorch Official
ENExtending TorchInductor with FlyDSL: A New MLIR-Native Backend for High-Performance GEMMs
Здравствуйте, я Лиз Ли из AMD. >> Здравствуйте, я Джек Хуай из AMD. >> Современные рабочие нагрузки ИИ зависят от растущего спектра специализированных ядер GPU, и ни один бэкенд генерации ядер не является оптимальным для всех форм операторов и архитектур. Мы исследуем эту…
YouTube — PyTorch Official
ENPyTorch-Native LLM Serving on TPU: SGLang and vLLM
Hi, I'm G from Google. >> I'm Colin from Meta. >> I'm Angela from Meta. >> We are going to be speaking at the PyTorch conference San Jose about new VLM and SG learn engines built on a new PyTorch native TPU backend torch TPU. >> Rather than build a new serving engine or porting…
YouTube — PyTorch Official
ENMaking Weather Forecasts More Accurate
So we know the Earth is getting warmer. We understand climate change at a big picture, but we're struggling to predict the details of what that means for the Earth. NeuralGCM is a new kind of atmospheric model. We feel it has potential to transform how we understand climate.…
YouTube — Google Research
ENProject Vaani
- In India, we've got 1,300+ languages. There's a saying here, every mile the taste of water changes and every four miles the language changes. - If you travel from one state to another, even one district to another district, you find it difficult to communicate. What will…
YouTube — Google Research
ENPyTorchCon is where the open source Al community gathers to talk about what we're building together
Hi there. I'm Priya Nagpurkar, VP of AI Native Systems at IBM Research. I'll be at the PyTorch Conference in San Jose, where the open source AI community will gather to talk about what we're all building together. >> [music] >> On Wednesday, October 21st, I'll be giving a…
YouTube — PyTorch Official
ENDeepSpeed Is Not Just ZeRO: Tensor, Sequence, and Expert Parallelism with ZeRO-Level Usability
Masahiro Tanaka will present “DeepSpeed Is Not Just ZeRO: Tensor, Sequence, and Expert Parallelism with ZeRO-Level Usability” at PyTorch Conference North America 2026. The session will cover DeepSpeed’s AutoTP, AutoSP, and AutoEP capabilities for tensor, sequence, and expert…
YouTube — PyTorch Official
ENInvolved in the AI ecosystem? There's something for everyone at PyTorch Conference North America
Hey folks, my name's Steve. Um, I work at Red Hat in the open source AI programs office and I really want to encourage you to come join us at PyTorch conference in San Jose this year. There's a lot of reasons to go, everything from PyTorch in um, we're going to be doing a…
YouTube — PyTorch Official
ENLanguage AI at Google
- Language is core to what it means to be human. It's how we make sense of the world around us and how we relate to one another. It's an expression of identity. Language is not only how we communicate with one another, it's also how we communicate with technology. And while most…
YouTube — Google Research
ENA Conversation on Language AI
- I wanna start with Google's sort of broader mission of making information more accessible and useful. And for 25 years or so, we've done that, impacting billions of people across the world. What do you see as the future of making information universally accessible and what…
YouTube — Google Research
ENCan we hear disease before we see it?
Интересно представить, что будущее клинической помощи может заключаться в том, что мы будем слышать, прежде чем видеть. В исследовательском центре Google мы разработали акустические модели для здравоохранения, известные как Here, которые являются базовой моделью. С помощью Here…
YouTube — Google Research
ENHigh Performance Serving Framework for Offline and Real-time Diffusion Models - SGLang-Diffusion
SGLang-Diffusion is a high-performance serving framework for diffusion models, designed for both large-scale offline generation and latency-sensitive real-time inference. As diffusion workloads expand from image and video generation to diffusion-based world models, serving…
YouTube — PyTorch Official
ENHigh-Velocity GPU Kernel Authoring with CUTLASS Python
Hi everyone. I am Guray, compiler engineer in Nvidia, and I'm going to share our recent improvements to CUTLASS for Python. CUTLASS for Python is our technology stack for writing speed of light kernels for the latest generations of GPUs and can be found under the hood today's…
YouTube — PyTorch Official
ENMoRI + vLLM: Wide Expert Parallelism and RDMA KV-Cache Transfer for Disaggregated MoE Serving on AMD
Running large Mixture-of-Experts models across pods creates two major challenges: expert dispatch and combine traffic, and moving the KV cache between prefill and decode. At North America 2026, Ravi Gupta, Rishi Madduri, and Shiksha Patel of AMD will present “MoRI + vLLM: Wide…
YouTube — PyTorch Official
ENAI for Eye Care: A fireside chat on AI in clinical workflows
- So, a few words of introduction. Dr. Kim is the Chief Medical Officer and Director of IT and Systems at Aravind Eye Care System. He pioneer teleophthalmology in India and developed an EMR system used in many eye hospitals. He has been one of our very close collaborators for…
YouTube — Google Research
ENUsing AI to track wildfire boundaries | Chile, February 2024
Using AI to track wildfire boundaries | Chile, February 2024
YouTube — Google Research
ENOpen Buildings: Using AI to put everyone on the map
(gentle music) - [Abigail] Maps tell us where to go, but they are also a lifeline to the things we need. - [Abdoulaye] It's very important to be visible on Maps. We need to know where people live, where are the buildings, especially in informal settlements. That helps with…
YouTube — Google Research
ENExchange ideas w/ other devs | on-device AI, inference, voice agents, machine learning compilers...
Hi, I'm Masahiro Hiramoto from Mitsubishi Electric. I'm a member of the program committee for PyTorch conference North America and a PyTorch ambassador. I'm especially looking forward to seeing what people across the community are building with PyTorch and exchanging ideas with…
YouTube — PyTorch Official
ENOpen Source Inference at Production Scale
Hello everyone, I'm Carlos Costa, a distinction engineer here at IBM Research and co-founder of the LMD project. I'll be heading soon to San Jose for PyTorch Conference North America, where I'll have the privilege of sharing some exciting news about LMD. How the project changed…
YouTube — PyTorch Official
ENInterchangeable KV Importance Signals under One Compiled FlexAttention Budget with FlexBudget
When you run a model over a long context, the KV cache gets huge, so you have to throw some of it away. There are competing rules for choosing what to drop: keep the most recent tokens, keep the ones that get the most attention, and so on. The problem is that every method ships…
YouTube — PyTorch Official
ENBringing Atmospheric Data Down to Earth | Mapping the Ionosphere with Android Devices
the northern and southern lights have been a source of Fascination and wonder for thousands of years but in 2024 people in some parts of the world noticed that the Aurora had come to them and if they checked their phones they might have also noticed their GPS thought they were…
YouTube — Google Research
ENHow AI can help policymakers improve access to emergency obstetric care
(relaxed guitar music) The labor started late in the night around 1:00 AM. I endured the pain 'til 5:00 AM in the morning because of the journey. We had to travel for so long and my wish for every woman is that this should not happen again. (relaxed guitar music) (crickets…
YouTube — Google Research
ENGeospatial Reasoning: Unlocking insights with generative AI and multiple foundation models
From Maps and Trends, to weather, floods, and wildfires, Google has studied the geospatial world for decades. And we’ve made that information accessible through AI models and real-time services. But synthesizing across these models and combining your data with ours can be…
YouTube — Google Research
ENBefore Rivers Rise | Using AI to Make Global Flood Forecasts
(rain pattering) - [Narrator] When it comes to a flood crisis, advanced warnings make a big difference. Knowing a flood is coming, even an hour in advance, gives people time to prepare and take action. So imagine if hours became days. (notifications chiming) AI makes that…
YouTube — Google Research