"One thing that's special about the PyTorch Conference is that it really does focus on cutting-edge research and tools and trends that we see in AI."
Jana van Greunen, Director of Engineering, @Meta
At PyTorch Conference North America, our community gathered in person to gence across the entire loop. From keynotes to technical deep dives, researchers, maintainers, and engineers came together to solve open bottlenecks and decide what comes next.
Read the full event recap: bit.ly/481GKe8
Catch the community highlights and hear directly from attendees: bit.ly/4zGi2vR
Jana van Greunen, Director of Engineering, @Meta
At PyTorch Conference North America, our community gathered in person to gence across the entire loop. From keynotes to technical deep dives, researchers, maintainers, and engineers came together to solve open bottlenecks and decide what comes next.
Read the full event recap: bit.ly/481GKe8
Catch the community highlights and hear directly from attendees: bit.ly/4zGi2vR
5
8
21
5,973
PyTorch Foundation, @CloudNativeFdn, and the @AgenticAIFdn are hosting an evening reception alongside NeurIPS in Sydney on December 8.
Meet researchers, engineers, and open source practitioners working across AI and cloud native infrastructure, and swap ideas over drinks and tion runs from 6 to 9 PM Sydney time on Tuesday, December 8, at Darling Pavilion, 1 Harbour St. Space is limited.
Find event details and registration at pytorch.org/event/cncf-pyt…
Meet researchers, engineers, and open source practitioners working across AI and cloud native infrastructure, and swap ideas over drinks and tion runs from 6 to 9 PM Sydney time on Tuesday, December 8, at Darling Pavilion, 1 Harbour St. Space is limited.
Find event details and registration at pytorch.org/event/cncf-pyt…
1
2
12
6,054
PyTorch operations can be performance optimized for MoE models applications.
MoE models replace a single dense feed-forward block with multiple expert networks. A naive implementation, however, can trigger excessive kernel launches. For example, the Hugging Face baseline over all experts in a Python loop, with each expert triggering separate kernel launches.
Grouped execution preserves the individual expert matrices but submits their work together, reducing launch and scheduling overhead. This blog from @nvidia uses biological foundation models as an example for efficient MoE training.
Read the full post:
bit.ly/4z2ulkW
MoE models replace a single dense feed-forward block with multiple expert networks. A naive implementation, however, can trigger excessive kernel launches. For example, the Hugging Face baseline over all experts in a Python loop, with each expert triggering separate kernel launches.
Grouped execution preserves the individual expert matrices but submits their work together, reducing launch and scheduling overhead. This blog from @nvidia uses biological foundation models as an example for efficient MoE training.
Read the full post:
bit.ly/4z2ulkW
11
7
84
10,811
Submit your proposal for PyTorch Day China 2026 before Tuesday, November 10, at 23:59 China Standard Time.
Propose a session presentation or lightning talk on model training and post-training, efficient inference and deployment, PyTorch Core and hardware acceleration, Physical m tools and applications.
Share your work with PyTorch developers, machine learning engineers, AI researchers, and open source AI leaders in Beijing on December 12.
Read more: pytorch.org/blog/call-for-…
Propose a session presentation or lightning talk on model training and post-training, efficient inference and deployment, PyTorch Core and hardware acceleration, Physical m tools and applications.
Share your work with PyTorch developers, machine learning engineers, AI researchers, and open source AI leaders in Beijing on December 12.
Read more: pytorch.org/blog/call-for-…
2
1
10
5,336
What is driving the unprecedented acceleration in AI? @sparkycollier summed it up in one word: OPEN during his PyTorch Conference North America 2026 keynote.
At the center of this movement is the PyTorch Foundation serving as the home for the open source intelligence layer ning, optimizing, serving, and running models on any chip in any cloud for any agent.
Watch Mark's full keynote that sets the stage for the whole event: youtube.com/watch?v=n33zCp…
+ follow along with us as we share all of the recordings from #PyTorchCon over the next few weeks
At the center of this movement is the PyTorch Foundation serving as the home for the open source intelligence layer ning, optimizing, serving, and running models on any chip in any cloud for any agent.
Watch Mark's full keynote that sets the stage for the whole event: youtube.com/watch?v=n33zCp…
+ follow along with us as we share all of the recordings from #PyTorchCon over the next few weeks
3
3
18
7,681
Bring your questions about PyTorch 2.15 to our live Q&A TOMORROW, November 5, at 11 AM PT.
Tristan Rice (@rice_fry), Natalia Gimelshein, Huy Do, and Jane Xu (@janeyx99) (@Meta), with moderator Chris Gottbrath (Gottbrath Tech), will share an overview of the release and answer commyTorch and the new capabilities in the release.
PyTorch 2.15 makes the rewritten NCCL backend the default, adds one-sided transports with a NIXL backend, and gives FSDP2 and pipelining finer memory and communication control. The release also brings faster 4-bit (NVFP4) LLM decoding on NVIDIA Blackwell GPUs, FP8 and faster decode on Apple Silicon, and broader platform support.
Register: pytorch.org/event/pytorch-…
Tristan Rice (@rice_fry), Natalia Gimelshein, Huy Do, and Jane Xu (@janeyx99) (@Meta), with moderator Chris Gottbrath (Gottbrath Tech), will share an overview of the release and answer commyTorch and the new capabilities in the release.
PyTorch 2.15 makes the rewritten NCCL backend the default, adds one-sided transports with a NIXL backend, and gives FSDP2 and pipelining finer memory and communication control. The release also brings faster 4-bit (NVFP4) LLM decoding on NVIDIA Blackwell GPUs, FP8 and faster decode on Apple Silicon, and broader platform support.
Register: pytorch.org/event/pytorch-…
0
1
12
5,459
#PyTorchDayChina 2026 comes to Beijing on December 12, and the call for proposals is OPEN!
We're accepting proposals for session presentations and Lightning Talks across the full spectrum of open source AI, PyTorch, and AI/ML.
Submit by Tuesday, November 10, at 23:59 China //t.co/szlAXuZiFV
We're accepting proposals for session presentations and Lightning Talks across the full spectrum of open source AI, PyTorch, and AI/ML.
Submit by Tuesday, November 10, at 23:59 China //t.co/szlAXuZiFV
4
4
27
7,905
🌏 From Learners to Contributors: Growing the PyTorch Ecosystem through PyTorchKR
bit.ly/4ixkw9A
Join Junghwan Park from PyTorchKR & Jiho Kim, PyTorch Foundation Ambassador, during PyTorch Day Japan 2026 to discover how eight years of community building have helped deved students become open source contributors.
Register by November 11 and save ¥3,000. bit.ly/4zMCquV
bit.ly/4ixkw9A
Join Junghwan Park from PyTorchKR & Jiho Kim, PyTorch Foundation Ambassador, during PyTorch Day Japan 2026 to discover how eight years of community building have helped deved students become open source contributors.
Register by November 11 and save ¥3,000. bit.ly/4zMCquV
0
2
16
6,601
This blog post from @nvidia is intended to develop a clearer way of reasoning about performance optimization for PyTorch models and other representations.
It explores questions such as: What makes one compiled engine better than another? Why can individually promising decisions ould accuracy constrain performance optimization?
It introduces the Global Performance Tuner, a feature for automatic performance tuning released in TensorRT 11.1, and explores these questions through an intuitive example and a mathematical framework.
Read the full blog post here: bit.ly/4zl9sCB
It explores questions such as: What makes one compiled engine better than another? Why can individually promising decisions ould accuracy constrain performance optimization?
It introduces the Global Performance Tuner, a feature for automatic performance tuning released in TensorRT 11.1, and explores these questions through an intuitive example and a mathematical framework.
Read the full blog post here: bit.ly/4zl9sCB
2
6
50
9,730
At the recent KVCache Meetup Korea, co-hosted by PyTorchKR and XCENA, speakers from XCENA, SK hynix, FuriosaAI, and LMCache dove deep into the architectural trade-offs powering large language model serving.
Talks spanned CXL memory pooling with Maru, roofline analysis of GPU th during prefill and decode, batch scheduling on RNGD, and process-isolated cache serving with LMCache MP Mode.
Real-world optimization requires collaboration across every tier of the stack, from accelerators and memory architectures to runtime engines. Thank you to PyTorch Foundation Ambassador Junghwan Park, co-host XCENA, and sponsor @tensormesh for bringing this local developer community together.
Catch the full event recap here: bit.ly/4sLT8bG
Talks spanned CXL memory pooling with Maru, roofline analysis of GPU th during prefill and decode, batch scheduling on RNGD, and process-isolated cache serving with LMCache MP Mode.
Real-world optimization requires collaboration across every tier of the stack, from accelerators and memory architectures to runtime engines. Thank you to PyTorch Foundation Ambassador Junghwan Park, co-host XCENA, and sponsor @tensormesh for bringing this local developer community together.
Catch the full event recap here: bit.ly/4sLT8bG
6
1
38
9,030
Bring your questions about PyTorch 2.15 to our live Q&A this Thursday, November 5, at 11 AM PT.
Tristan Rice (@rice_fry), Natalia Gimelshein, Huy Do, and Jane Xu (@janeyx99) (all @Meta), with moderator Chris Gottbrath (Gottbrath Tech), will share an overview of the release and answabout PyTorch and the new capabilities in the release.
PyTorch 2.15 makes the rewritten NCCL backend the default, adds one-sided transports with a NIXL backend, and gives FSDP2 and pipelining finer memory and communication control. The release also brings faster 4-bit (NVFP4) LLM decoding on NVIDIA Blackwell GPUs, FP8 and faster decode on Apple Silicon, and broader platform support.
Register: pytorch.org/event/pytorch-…
Tristan Rice (@rice_fry), Natalia Gimelshein, Huy Do, and Jane Xu (@janeyx99) (all @Meta), with moderator Chris Gottbrath (Gottbrath Tech), will share an overview of the release and answabout PyTorch and the new capabilities in the release.
PyTorch 2.15 makes the rewritten NCCL backend the default, adds one-sided transports with a NIXL backend, and gives FSDP2 and pipelining finer memory and communication control. The release also brings faster 4-bit (NVFP4) LLM decoding on NVIDIA Blackwell GPUs, FP8 and faster decode on Apple Silicon, and broader platform support.
Register: pytorch.org/event/pytorch-…
0
4
24
8,333
Today, PyTorch Foundation project, Ray is announcing the Special Interest Group (SIG) for Ray on Accelerators.
Ray already supports more than a dozen accelerator types, from NVIDIA and AMD GPUs to TPUs and Trainium. The SIG brings hardware vendors, cloud providers, and the Ray nd a vendor-neutral roadmap focused on:
· Development and testing infrastructure on real hardware
· Hardware-aware scheduling that understands topology
· Direct device-to-device transport with Ray Direct Transport
Join us on the Ray Slack channel, at community meetings, or on GitHub.
Learn more here: bit.ly/47V3QCZ
@raydistributed
Ray already supports more than a dozen accelerator types, from NVIDIA and AMD GPUs to TPUs and Trainium. The SIG brings hardware vendors, cloud providers, and the Ray nd a vendor-neutral roadmap focused on:
· Development and testing infrastructure on real hardware
· Hardware-aware scheduling that understands topology
· Direct device-to-device transport with Ray Direct Transport
Join us on the Ray Slack channel, at community meetings, or on GitHub.
Learn more here: bit.ly/47V3QCZ
@raydistributed
5
12
51
8,526
Validate your real-world PyTorch skills with the PyTorch Certified Associate (PTCA) program.
Designed for early-stage practitioners, PTCA helps you demonstrate foundational ability across model design, training, and deployment while building credibility across the open source re and enroll here: bit.ly/4do6qEx
Designed for early-stage practitioners, PTCA helps you demonstrate foundational ability across model design, training, and deployment while building credibility across the open source re and enroll here: bit.ly/4do6qEx
1
1
31
9,486
We ❤️ our ambassadors. The ambassador meetup at #PyTorchCon North America was a good time to connect and make plans for the year ahead.
PyTorch Foundation Ambassadors help grow our projects (PyTorch, vLLM, DeepSpeed, Ray, Helion, & Safetensors) around the world by organizing evenl content, mentoring new users, and contributing to the open source ecosystem.
Meet all of your PyTorch Foundation ambassadors: pytorch.org/programs/ambas…
PyTorch Foundation Ambassadors help grow our projects (PyTorch, vLLM, DeepSpeed, Ray, Helion, & Safetensors) around the world by organizing evenl content, mentoring new users, and contributing to the open source ecosystem.
Meet all of your PyTorch Foundation ambassadors: pytorch.org/programs/ambas…
2
3
55
14,333
🚀 Unlock frontier-level AI training performance with DeepSpeed!
bit.ly/4zLzUoS
Join Masahiro Tanaka of DeepSpeed at PyTorch Day Japan 2026 as he presents “DeepSpeed at the Frontier: Community-Driven Optimizations for State-of-the-Art Training Performance,” exploring advancing the performance and scalability of distributed AI training.
Register by November 11 and save ¥3,000. bit.ly/4zMCquV
bit.ly/4zLzUoS
Join Masahiro Tanaka of DeepSpeed at PyTorch Day Japan 2026 as he presents “DeepSpeed at the Frontier: Community-Driven Optimizations for State-of-the-Art Training Performance,” exploring advancing the performance and scalability of distributed AI training.
Register by November 11 and save ¥3,000. bit.ly/4zMCquV
2
3
33
8,924
PyTorch 2.15 is here, with 3,117 commits from 583 contributors and updates across distributed training, NVGEMM, Apple Silicon, and accelerator platform support.
The release makes the rewritten NCCL backend the default and adds one-sided transports with a NIXL backend. FSDP2 and emory and communication control. It also brings faster 4-bit (NVFP4) LLM decoding on NVIDIA Blackwell GPUs, FP8 and faster decode on Apple Silicon, and broader platform support.
On Thursday, November 5, at 11 AM PST, join Tristan Rice (@rice_fry), Natalia Gimelshein, Huy Do, and Jane Xu (@janeyx99) (@Meta) for a live PyTorch 2.15 Q&A, moderated by Chris Gottbrath (Gottbrath Tech). The speakers will share an overview of the release and answer community questions about PyTorch and the new capabilities in this release.
🔗 Read the release blog and register for the live Q&A: pytorch.org/blog/pytorch-2…
The release makes the rewritten NCCL backend the default and adds one-sided transports with a NIXL backend. FSDP2 and emory and communication control. It also brings faster 4-bit (NVFP4) LLM decoding on NVIDIA Blackwell GPUs, FP8 and faster decode on Apple Silicon, and broader platform support.
On Thursday, November 5, at 11 AM PST, join Tristan Rice (@rice_fry), Natalia Gimelshein, Huy Do, and Jane Xu (@janeyx99) (@Meta) for a live PyTorch 2.15 Q&A, moderated by Chris Gottbrath (Gottbrath Tech). The speakers will share an overview of the release and answer community questions about PyTorch and the new capabilities in this release.
🔗 Read the release blog and register for the live Q&A: pytorch.org/blog/pytorch-2…
10
20
170
18,260
On Oct 19 at PyTorch Conference North America 2026 in San Jose, attendees spent a full day in our instructor-led PyTorch Associate Training preparing for the PyTorch Certified Associate (PTCA) exam. Here’s what they had to say:
💬 “From basics to advanced, the instructors did a ”
💬 “Everything was hands-on... with powerful GPUs.”
💬 “I used to think of PyTorch as training models, but it was so much more: benchmarking and profiling.”
💬 “Made me confident for the exam.”
💬 “Absolutely recommend to friends.”
Thank you to our teaching staff from @NVIDIAAI: Allison Ding, Krishna Kalyan, Lakshay Garg, Santosh Bahir, Ian Wang, and Faradawn Yang.
Learn more about getting certified: bit.ly/4do6qEx
Check out what the attendees had to say here: bit.ly/4Bvhb2j
#PyTorchCon
💬 “From basics to advanced, the instructors did a ”
💬 “Everything was hands-on... with powerful GPUs.”
💬 “I used to think of PyTorch as training models, but it was so much more: benchmarking and profiling.”
💬 “Made me confident for the exam.”
💬 “Absolutely recommend to friends.”
Thank you to our teaching staff from @NVIDIAAI: Allison Ding, Krishna Kalyan, Lakshay Garg, Santosh Bahir, Ian Wang, and Faradawn Yang.
Learn more about getting certified: bit.ly/4do6qEx
Check out what the attendees had to say here: bit.ly/4Bvhb2j
#PyTorchCon
2
1
27
11,513
Faster streaming speech recognition on @Arm edge devices: see how ExecuTorch, SME2, FP16 and QC8W balance accuracy and latency—delivering up to 2.12× faster processing on Pixel 11.
Read our latest blog: pytorch.org/blog/optimizin…
✍ Josiah Davis, Kshitij Sisodia
Read our latest blog: pytorch.org/blog/optimizin…
✍ Josiah Davis, Kshitij Sisodia
14
11
79
13,444