Featured projects
As deep learning models move rapidly from research prototypes into core enterprise infrastructure, the demand for practical, end-to-end PyTorch expertise has never been higher. Building robust neural networks requires more than just high-level understanding. It demands hands-on experience in training, optimizing, and deploying models effectively.
To support developers, researchers, and engineers looking to deepen their technical skills, the PyTorch Foundation is running a dedicated Introduction Track during the PyTorch Conference North America, as well as an official, full-day PyTorch Associate Training session on Monday, October 19, 2026, 9am – 5pm in San Jose, California.
In this blog we share what developers can expect from the technical sessions and training.
What to Expect from the Introduction Track
All attendees, from students, first time attendees all the way to seasoned AI engineers that want to learn about new areas are invited to check out the brand new Introduction Track at PyTorch Conference NA. Learn about areas of the PyTorch library, PyTorch ecosystem or AI stack that you are not familiar with. Explore foundational concepts, common workflows, project orientation, and approachable ways to begin building with the PyTorch ecosystem. Sessions include:
From Math Panic to PyTorch Confidence
Yashasvi Misra, 3:25 PM PDT, Room LL20AB
Designed for Python developers, students, and educators new to machine learning, this session uses visual explanations and interactive exercises to demystify tensors, gradients, and dynamic computation graphs. Attendees will build intuitive mental models before implementing a foundational neural network and training loop in PyTorch.
Practical GPU Programming with Triton for PyTorch Developers
Suman Debnath, JanakiRam Goteti, 11:45 AM PDT, Room LL20CD
PyTorch developers rely on tools like torch.compile to generate GPU kernels, but writing custom kernels can feel daunting. This session introduces Triton, an open-source Python-embedded programming model for authoring high-performance GPU code without CUDA or C++. Attendees will learn GPU memory movement fundamentals, explore thread parallelization, and write custom vector addition and matrix multiplication kernels.
What You Cannot Profile, You Cannot Optimize: Learning to Read PyTorch Traces
Aritra Roy Gosthipaty, Suvaditya Mukherjee, 2:50 PM PDT, Room LL20AB
Modern profiling tools generate complex traces that can be difficult to interpret. This talk builds practical mental models around CPU-to-GPU dispatch chains, compute-bound versus overhead-bound workloads, and kernel timing variations. Attendees will evaluate workloads using torch.compile, fused Triton kernels, and Liger kernels to make accurate performance comparisons.
From Scratch to PyTorch: Demystifying ML Frameworks by Building Your Own
Andrea Mattia Garavagno, Vijay Janapa Reddi, 2:15 PM PDT, Room LL20AB
Understanding framework internals is critical for debugging complex logic, optimizing hardware, and contributing to backend infrastructure. This session introduces TinyTorch, an open-source CLI project that guides developers through rebuilding core framework components, including tensors, autograd, optimizers, and transformers, in pure Python.
Understanding Modern Vision Language Models
Aastha Jhunjhunwala, Mark Moyou, 12:20 PM PDT, Room LL20AB
Vision-language models (VLMs) present unique serving and scaling challenges. Deconstructing five open source architectures, this talk examines image-to-token encoding, vision-language fusion strategies, multi-GPU fine-tuning, and production serving overheads such as KV-cache pressure and image-token expansion.
A Developer’s Guide to Attention in vLLM
Lucas Wilkinson, Matthew Bonanni, 11:45 AM PDT, Room LL20AB
Modern state-of-the-art models rely on hybrid, sliding-window, sparse, and linear attention mechanisms to manage KV-cache memory over long context windows. This talk details how the vLLM engine represents and optimizes these diverse patterns through attention backends, KV-cache connectors, and hybrid memory allocators.
Elevate your AI Skills: PyTorch Associate Training
The PyTorch Foundation is excited to host an official, full-day, instructor-led PyTorch Associate Training on Monday, October 19, 2026.
What to expect from the training
The PyTorch Associate Training is an in-person, instructor-led program, designed to equip participants with foundational and practical skills identified across the key domains of the PyTorch Certified Associate exam. Delivered through a structured series of sequential modules, the curriculum bridges the gap between core concepts and production-ready implementation.
Participants in this full-day workshop will work with:
- Interactive Lectures & Live Demos: Gain clear technical insights directly from experienced instructors as they demonstrate real-time model construction and training.
- Guided Jupyter Notebook Labs: Build, train, and optimize deep learning models from scratch in interactive lab environments.
- Comprehension Checkpoints: Solidify your understanding of key PyTorch ecosystem components through structured quizzes and reviews.
- Practical Industry Projects: Work on real-world use cases that mirror production scenarios and deployment challenges.
By learning alongside peers in an interactive setting, attendees receive direct feedback and real-time guidance from instructors.
To get the most out of this hands-on course, attendees should have:
- Proficiency in Python programming
- Familiarity with Jupyter Notebooks and basic Google Colab environment workflows
- A basic understanding of machine learning concepts
- An active Google Account for accessing practical exercises
Meet your Instructor
Faradawn Yang is an engineer on the AI platform software team at NVIDIA, where he focuses on AI inference products. Faradawn holds a master’s degree in computer science and a bachelor’s degree in mathematics and computer science from the University of Chicago. Before joining NVIDIA, he worked as a data engineer at a marketing measurement company.
Receive a Certification Voucher
To support your long-term professional development, every attendee who completes the training will receive a voucher (valued at $250) for the official PyTorch Certified Associate (PTCA) certification exam. Taking the PTCA exam provides a clear, industry-recognized credential to validate your practical skills in deep learning workflows using PyTorch.
Register now to elevate your AI skills
Pre-registration is required to secure your seat in the classroom, as space is limited to maintain an optimal student-to-instructor ratio.You can add the PyTorch Associate Training directly to your PyTorch Conference North America registration through Tuesday, October 13th.
Learn more and register for the PyTorch Associate Training at PyTorch Conference NA.