IndustryMulti-source10月20日 08:45

PyTorchCon 北美 2026 明日在圣何塞开幕，设七大技术议题

官方账号PyTorch@PyTorch

Post· 11 sources

Why we picked it

PyTorch 官方大会明天开幕，vLLM、Ray、DeepSpeed 这些项目的开发者都会到场，做工程的朋友可以看看七个议题的议程。

PyTorchCon 北美 2026 于 10 月 20–21 日在 San Jose 举行，内容覆盖 PyTorch、vLLM、DeepSpeed、Ray、Helion 和 Safetensors 等开源项目。大会设 Applications、Core PyTorch、Inference、Kernel Engineering、Responsible AI 等七个技术议题。主题演讲嘉宾来自 Meta、Google Cloud、NVIDIA、Qualcomm、Shopify、IBM、Amazon 等公司，包括 PyTorch Foundation 的 Sparky Collier 和 UC Berkeley 的 Ion Stoica。现场还安排海报展示、Meet the Developers 问答和 Inference Engineering 作者 Philip Kiely 的签书活动。

Original · PyTorch

#PyTorchCon North America 2026 starts tomorrow in San Jose, bringing the open source AI community together to accelerate open intelligence.

Across PyTorch, @vllm_project, @DeepSpeedAI, @raydistributed, Helion, and Safetensors, our program explores the full AI stack, from bare metal infrastructure to applications and agent-based systems, through technical talks, hands-on workshops, and candid conversations.

Tomorrow and Wednesday, October 20–21, the program spans seven tracks: Applications, Core PyTorch, Inference, Introduction, Kernel Engineering, Responsible AI, and Training.

Keynote speakers include @sparkycollier (PyTorch Foundation); @cmazzaanthony (@Shopify); @albanDesmaison, Edward Yang, Jana van Greunen, and Zongwei Zhou (@Meta); @clattner_llvm (@Qualcomm); @istoica05 (@UCBerkeley / @anyscalecompute); Robert Hundt (@amazon); @sarahookr (@adaption_ai); @simon_mo_ (@inferact); Carlos Costa and Priya Nagpurkar (@IBM); Abdullah Gharaibeh and Bill Jia (@googlecloud); Stephen Watt (@RedHat); Ujval Kapasi (@nvidia); Mazin Gilbert (@AgenticAIFdn); Zhipeng Wang (@Google); Rachad Alao (@cohere); Yiwei Song (@CrusoeAI); and @marksaroufim (Core Automation).

Beyond the talks, exchange ideas in Birds of a Feather discussions, ask questions in Meet the Developers sessions, and watch sponsor presentations in the Demo Theater.

Tuesday offers more opportunities for conversation: discuss presenters’ work at the evening poster session, connect at the Women & Non-Binary in PyTorch Lunch, or meet Inference Engineering author Philip Kiely at his book signing.

Explore the program: https://t.co/hrXYyxK37G

Join us at #PyTorchCon tomorrow and Wednesday: https://t.co/x5F9b2veWG

Nathan Lambert10-20 17:11原文
Thomas Wolf10-18 14:34原文
rohanpaul_ai10-20 19:31原文
TestingCatalog10-19 10:37原文
elvis10-19 13:46原文
berryxia10-20 11:34原文
Artificial Analysis10-20 15:20原文
techcrunch10-20 18:38原文
Blognone10-21 13:33原文
vLLM10-18 01:55原文

#PyTorch #vLLM #Ray #开源模型 #PyTorchCon

Keep reading

Recent stories with the same tag or section

Industry BuildersMulti-source05:21

Chinese content2026 PyTorch Contributor Award 获奖者名单公布

PyTorch 年度贡献者奖揭晓，华人开发者拿下不少名额，做开源框架的可以看看名单认认人。

#PyTorch #vLLM #DeepSpeed #开源模型

Event

官方账号PyTorch@PyTorch11 sourcesSource

Products BuildersMulti-sourceSelected04:07

PyTorchCon Demo Theater: 13 Sessions on Training, Inference and Agents

The lineup reveals where the PyTorch ecosystem is concentrating effort: inference speed for video models, MoE expert parallelism on NVL72, and fault-tolerant training that survives link and node failures live. Engineers choosing accelerator backends can use these demos to compare AWS Trainium, Intel AI PCs, TPUs and Rebellions hardware on real workloads. The Huawei CRCR CI relay is also a practical pattern for teams maintaining forks against upstream PyTorch.

#PyTorch #PyTorchCon #Wan 2.2 #vLLM

Event

官方账号PyTorch@PyTorch11 sourcesSource

Products BuildersMulti-source04:01

Chinese contentTorchTitan 更新：支持 DeepSeek-V4 等 5D 并行，新增 TitanRL 框架

PyTorch 官方的 TorchTitan 大更新，DeepSeek-V4 和 Qwen3.5 都能跑 5D 并行了，Muon 优化器还能提速 79%，训大模型的得看看。

#TorchTitan #PyTorch #DeepSeek-V4 #Qwen3.5

Event

官方账号PyTorch@PyTorch5 sourcesSource

How-to BuildersSelected03:43

PyTorch Examines Linear Algebra Kernels Behind Faster Optimizers

The speed of an optimizer depends as much on kernel-level linear algebra as on step count, a factor most training guides ignore. Engineers tuning large-scale training can apply the QR, Cholesky and eigendecomposition implementation patterns to cut data movement on GPUs. The talk ties directly to GPU MODE's competition, giving concrete reference implementations to study.

#PyTorch #optimizers #GPU #linear algebra

官方账号PyTorch@PyTorchSource

Products BuildersSelected02:54

vLLM-TorchTPU brings native PyTorch serving to Google TPUs

The practical effect is that vLLM's optimizations — prefix caching and speculative decoding — now apply to TPU hardware, giving teams a second accelerator option beyond NVIDIA GPUs for self-hosted inference. Organisations already locked into the PyTorch and vLLM stack can evaluate TPU economics without rewriting their serving code. It also signals Google's continued push to open TPU inference beyond JAX.

#vLLM #TPU #PyTorch #Google

官方账号vLLM@vllm_projectSource

Products BuildersSelected02:01

FlyDSL kernel DSL now supports PyTorch on AMD MI350 GPUs

AMD MI350 support matters because GPU kernel tooling has historically centred on NVIDIA's stack, and this gives teams serving open models a Python-native path to tune kernels without CUDA-only tools. Benchmark and deployment engineers running vLLM on AMD accelerators can adopt the DSL directly from PyTorch's ecosystem and measure the 3.01x serving gain against their own baselines.

#FlyDSL #PyTorch #AMD #vLLM

官方账号PyTorch@PyTorchSource