AMD at PyTorch Conference Europe

The inaugural PyTorch Conference Europe held April 7-8, 2026 at Station F in Paris, marked a significant milestone for the global AI community. As the first time a full-scale PyTorch Conference took place outside North America, the conference underscored the rapid expansion of the ecosystem and the role of hardware-software co-design in scaling today’s AI workloads.

Bridging Research and Production

The event featured over 100 technical sessions drawn from a pool of over 300 submissions, highlighting a decisive industry shift toward bridging the gap between cutting-edge research and production-grade efficiency. Key themes included distributed training scalability, privacy-first fine-tuning, and the optimization of inference serving, areas where AMD continues to lead through performance-per-watt innovation and open-source contributions.

AMD Contributions: From Silicon to Software

AMD was well represented across multiple layers of the stack: from distributed training and inference to model compression and developer tools. Collectively, our sessions reflected a clear focus: enabling performant, efficient, and open AI development on AMD platforms using PyTorch. Key sessions led by AMD include:

• Keynote: From One Node to Distributed Training and Inference  by Ramine Roane, Corporate VP of AI Product Management

• Training Embedding Model Resiliently for Multimodal Inference Routing, by Haichen Zhang, Senior AI Software Engineer
• Privacy-First Fine-Tuning on AI PCs, by Daniel Holanda Noronha, Solutions Architect & ML Engineer
• Brevitas Quantization Library, by Pablo Monteagudo Lago, Research Scientist
• Bringing PyTorch Monarch to AMD GPUs, by Liz Li, Principal AI Engineer and Zachary Streeter, Senior Member of Technical Staff
• Near-Lossless MXFP4 Compression for Accelerated LLM Serving, by Felix Marty, Senior Software Engineer

Demos and Developer Engagement

Beyond technical presentations, AMD hosted a series of live demos in the Community Expo area. These hands-on demos gave attendees an opportunity to see the combination of PyTorch with AMD hardware and software in action and seamless model migration for developers.

Looking Ahead: The Power of In-Person Collaboration

While digital collaboration remains essential, the high level of engagement at Station F demonstrated the unique value of in-person technical exchange. These interactions foster the collaborative problem-solving necessary to advance complex AI frameworks.

As a member of the PyTorch Foundation Governing Board, AMD remains committed to fostering this growth globally. Through upstream contributions and open tooling, we are helping enable a more performant and accessible AI ecosystem. This momentum continues throughout 2026, with AMD participating in:

• PyTorch Conference China (Sep 8-9, Shanghai)
• PyTorch Conference (Oct 20-21, San Jose)

For those who couldn’t attend in person, selected content will be available on the PyTorch YouTube Channel.

As the AI landscape continues to evolve, AMD is focused on enabling developers to efficiently build, train, and deploy models at scale, working closely with the community to push the boundaries of what’s possible with PyTorch.