Key highlights from Ultralytics at PyTorch Conference 2025

Last week, the global AI community gathered in San Francisco for PyTorch Conference 2025, held from October 22nd to 23rd at Moscone West. The two-day event brought together researchers, developers, and industry leaders to explore the latest advancements in deep learning, edge AI, and innovation in action.

For Ultralytics, it was an exciting opportunity not only to sponsor but to connect with the PyTorch community and share our latest work in Vision AI. One of the highlights was releasing our new ExecuTorch integration, a PyTorch-native solution that enables efficient on-device deployment of Ultralytics YOLO models on mobile and edge devices. 

With ExecuTorch, users can run real-time computer vision tasks like object detection directly on hardware such as smartphones, Raspberry Pi, and other embedded systems. This makes high-performance Vision AI more accessible to everyone.

Throughout the event, our team shared practical applications of Ultralytics YOLO models, met with partners across the AI ecosystem, and celebrated the creativity of our growing global community. From live demos and interviews to meaningful conversations on the expo floor, PyTorch Conference 2025 captured the spirit of cutting-edge ideas and community that drives progress in AI.

So let’s revisit some of our favorite moments and key highlights from an inspiring week in San Francisco. Let's get started!

Arriving in San Francisco and getting ready for the conference

Francesco Mattioli, our Lead Partnership Engineer, and Nuvola Ladi, our Digital Content Manager, arrived in San Francisco on October 20th, ready for a week of progress, collaboration, and connection. Before the conference began, they took time to explore the city and visit the Golden Gate Bridge, enjoying a quiet moment before the excitement of PyTorch Conference 2025.

‍

The following day, preparations began at Moscone West as the team set up the Ultralytics booth, arranged live demos, and made final checks for two full days of conversations with the AI community. Later that evening, Francesco and Nuvola attended Arm’s Connect, Chat, Chill event, where AI practitioners and partners from across the PyTorch ecosystem came together to exchange ideas and insights.

The Arm team’s sessions and demos showcased how their collaboration with PyTorch is enabling developers to bring efficient, high-performance AI from cloud to edge. It was an uplifting start to the week and a great way to kick off PyTorch Conference 2025.

An overview of PyTorch Conference 2025

Before we dive into Ultralytics’ experience at PyTorch Conference 2025, let’s take a quick look at the event itself.

First launched in 2022, PyTorch Conference has quickly grown into the world’s premier event dedicated to the framework powering today’s most groundbreaking AI innovations. This year’s edition brought together over 1,500 researchers, developers, and innovators from around the world.

As part of Open Source AI Week, the conference featured keynotes, technical sessions, and workshops covering everything from deep learning compilers and distributed training to edge AI and large language models. The overall focus was open collaboration and real-world impact.

Speakers included leaders from Meta, NVIDIA, AMD, UC Berkeley, Caltech, and fast.ai. Meanwhile, partners and sponsors such as Intel, Qualcomm, Arm, and Ultralytics showcased their latest innovations in the Expo Hall.

Attendees also joined casual meet-the-experts sessions, Birds of a Feather discussions, poster presentations, and community events such as Llama Palooza and the Flare Party. The atmosphere was full of energy and curiosity!

Key themes at PyTorch Conference 2025

Here are some of the key themes that stood out at PyTorch Conference 2025, capturing the excitement of the open-source AI community:

• Expanding the PyTorch ecosystem: The conference introduced exciting updates to PyTorch, including faster training, better performance, and new integrations such as vLLM, DeepSpeed, and Ray. These improvements are helping developers build smarter and more efficient AI systems.
  ‍
• AI at the edge: The launch of ExecuTorch 1.0 showed how AI can now run directly on devices like phones and embedded systems. This brings real-time, on-device intelligence to more people and makes advanced AI more accessible.
  ‍
• The rise of large language models: Speakers shared how large language models are becoming more open and collaborative through shared tools, data, and research that push innovation forward across the community.
  ‍
• Responsible and inclusive AI: Many sessions emphasized the importance of creating AI that is ethical, transparent, and inclusive. The message was clear that real progress in AI depends on openness, collaboration, and shared responsibility.

Showcasing the impact of Ultralytics YOLO models

At the Ultralytics booth, Francesco Mattioli and Nuvola Ladi welcomed attendees to experience live demos of Ultralytics YOLO models in action. The hands-on setup drew a steady stream of visitors curious to see what’s possible with real-time Vision AI.

Conversations quickly turned into deep dives on model performance, deployment efficiency, and accessibility. The feedback we received reflected a shared excitement about how Ultralytics YOLO is making computer vision more practical and scalable for AI enthusiasts everywhere.

‍

The booth soon became a meeting point for AI professionals, researchers, and enthusiasts, all eager to exchange ideas and share how they’re using Vision AI in their own work. It was especially rewarding to hear how Ultralytics YOLO models are being applied across industries, from manufacturing and retail to smart cities, agriculture, and healthcare, driving real-world advancements in so many different ways.

A spotlight on the new ExecuTorch integration

A key moment from day one of the conference was the spotlight on the new PyTorch ExecuTorch integration supported by Ultralytics, a runtime designed to enable efficient on-device AI across mobile and edge platforms. ExecuTorch makes it easy to deploy Ultralytics YOLO models directly on devices like smartphones, tablets, and embedded systems, delivering fast, real-time inference without relying on the cloud.

During the event, Francesco Mattioli, our Lead Partnership Engineer, was interviewed about Ultralytics and had the chance to explain how the ExecuTorch integration can expand on-device AI capabilities for developers. He explained that the integration enables models to be exported, optimized, and deployed seamlessly across platforms, making high-performance computer vision accessible even on resource-constrained devices.

‍

Ultralytics was also recognized as a PyTorch Executor success story, reflecting our early support for ExecuTorch and continued contributions to the PyTorch ecosystem. This collaboration stems from a shared goal of making AI more open, efficient, and accessible.

Connecting with partners and the AI community

Another highlight of the week was reconnecting with partners and meeting new faces from across the AI community. It was great to catch up with teams like Intel, share ideas, and talk about what’s next in Vision AI.

‍

We also loved hearing from developers and organizations already using Ultralytics YOLO in their projects. Their stories made it clear why it’s so important to stay connected with the community, to listen, learn, and keep pushing Vision AI to be even more impactful in the real world.

Celebrating the Ultralytics community

The second day of the event was all about connecting with our community. We’re always excited to meet users and partners in person, but this time was extra special as we announced the winner of our first Ultralytics giveaway. The prize included a Raspberry Pi 500 desktop kit, a CS mount camera, and an 8GB RAM Raspberry Pi 5.

‍

We’re constantly energized by the creativity and dedication of the global Ultralytics community. From open-source contributors to engineers building real-world solutions, every project shared helps move Vision AI forward. 

A huge thank you to everyone who continues to innovate, collaborate, and inspire us. Celebrating our shared progress is what makes this community so special and keeps us moving forward together.

Key takeaways

PyTorch Conference 2025 was a trailblazing two days in San Francisco filled with innovation, collaboration, and community. We had the chance to showcase Ultralytics YOLO models, highlight the new ExecuTorch integration, and share how Vision AI is continuing to evolve. 

On top of that, it was a good reminder of how meaningful it is to connect in person with the people and partners shaping the future of AI. As we look ahead, we’re more motivated than ever to continue pushing the boundaries of what’s possible with Vision AI.

Join our community and explore the latest in Vision AI on our GitHub repository. Visit our solution pages to learn how applications like AI in manufacturing and computer vision in healthcare are driving progress, and check out our licensing options to power your next AI solution.