Overview
Py Torch Day Japan 2026, hosted by the Py Torch Foundation, Hugging Face, IBM, and Mitsubishi Electric, is a pivotal event for community-driven learning, networking, and open source AI innovation. It's a premier gathering for Py Torch enthusiasts, machine learning engineers, AI researchers, and industry professionals. The event provides an unparalleled platform for insightful technical talks and interactive discussions, fostering knowledge exchange and collaboration on cutting-edge advancements in AI.
Circle
Meet the others goingBe the first to join · open until 30 days after the eventSpeakers
23Schedule
PyTorch Day Japan
Py Torch Day Japan 2026, hosted by the Py Torch Foundation, Hugging Face, IBM, and Mitsubishi Electric, is a pivotal event for community-driven learning, networking, and open source AI innovation.
Coffee Break
Registration & Badge Pick-Up
Opening Remarks - Mark Collier, Executive Director, PyTorch Foundation
Raghu Ganti, Distinguished Engineer, IBM
Lysandre Debut, Chief Open-Source Officer, Hugging Face
Tsubasa Masui, Senior Manager & Masahiro Hiramori, Head Engineer, Mitsubishi Electric
Coffee Break
Beyond GPUs: A Practical Playbook for Bringing ExecuTorch to a New Edge Accelerator
<p>Edge and on-prem AI inference increasingly runs on silicon that isn't a GPU. This talk tells the story of bringing ExecuTorch — PyTorch's on-device inference runtime — to IBM's AIU, a purpose-built accelerator for enterprise inference, entirely through ExecuTorch's public backend-delegate extension points: a custom partitioner, an AOT-only compilation path with well-defined CPU-fallback boundaries, a mixed-dtype tensor bridge, and static bucketing for dynamic shapes.</p><p
ExecuTorch in PyTorch Core: Running AI Everywhere, from Wearables to Robots
<p>ExecuTorch merged into PyTorch core earlier this year, bringing PyTorch-native on-device inference to production. In this session, the Meta PyTorch team will share how ExecuTorch powers local AI at Meta scale today — from Instagram, Facebook and WhatsApp apps, to Ray-Ban Meta glasses and Quest VR headsets — and how developers can run small open-source models like Muse Glimmer on-device.<br>We'll also cover Physical AI and robotics, including our work with the Hugging Face LeRobot team, w
OneComp: One-Command LLM Quantization
<p>Deploying large language models (LLMs) on limited hardware is challenging. Models may exceed available memory, and selecting quantization methods and bit widths requires specialist expertise. OneComp is an open-source Python library for resource-adaptive post-training quantization. With a single command, it detects available GPU memory, assigns layer-wise bit widths using AutoBit, performs quantization and error correction, evaluates the result, and saves the compressed model. OneComp incorpo
From Learners to Contributors: Growing the PyTorch Ecosystem through PyTorchKR
<p>How can a local PyTorch community turn learners into open-source contributors?</p><p>This talk shares lessons from eight years of PyTorch Korea, focusing on tutorials-kr, a community-maintained Korean translation of the official PyTorch tutorials. We explain how translated learning resources, university lectures, and open-source mentoring create entry points for developers, researchers, and students to contribute.</p><p>Our mentoring efforts supported 26 first-time contributors and 51 merged
Attendee Lunch
Training Robot Policies from Object Storage: LeRobot's Native LanceDB Backend
<p>LeRobot is Hugging Face's PyTorch library for real-world robotics, and its dataset format gave the community a shared standard. But robot learning data is multimodal and growing fast: long multi-camera videos, states, actions and language instructions in every episode. Training from object storage with global shuffling, and finding the right episodes first, still means copying data or stitching systems together.</p><p>LanceDB and the LeRobot team upstreamed LanceDB as the first non-LeRob
From PyTorch to Physical Action: Building Real-Time VLA Inference on Edge Devices
<p>Vision-Language-Action (VLA) models enable AI systems to perceive their environment, understand instructions, and perform physical actions; however, deploying VLA models on edge devices requires more than just getting inference running. Real-world systems have to satisfy strict requirements concerning latency, memory, power, thermal behavior, reliability, and security.</p><p>The talk describes a complete process for moving from a PyTorch VLA model to real-time on-device inference using torch.
Kernels Without Borders: Unifying the Kernel Ecosystem
<p>High-performance PyTorch applications increasingly depend on specialized kernels, but writing a fast kernel is only part of the problem. Projects also need to build across hardware and software configurations, distribute kernels, manage compatibility, version them safely, and integrate them into their own stacks.</p><p>In this talk, Sayak Paul and Aritra Roy Gosthipaty will introduce Hugging Face Kernels and show how the Kernel Hub provides a common layer for building, versioning, distributin
Adapting vLLM and SGLang to Diverse AI Accelerators with FlagOS
<p>Adapting inference frameworks to different AI accelerators requires coordinated work across operators, compilers, runtimes, and communication. This talk uses vLLM and SGLang to show how FlagOS organizes that adaptation while preserving familiar serving interfaces.</p><p>We connect engine plugins to FlagGems operators and explain the roles of FlagTree's supported compiler paths and optional FlagCX communication. A concrete RMSNorm example shows how two engine interfaces reach shared compu
Accelerating PyTorch on Apple silicon
<p>At the start of 2026, running models on PyTorch's MPS backend would mean working around unsupported operations, performance bottlenecks and memory growth even for small models. Since then, MPS has become much more mature platform to run the regular models on locally.</p><p>This talk goes through the progress on the backend such as, fused attention for prefill/decode workloads, Flex Attention, native Metal kernels for linear algebra operations, convolutions, porting MPS Graph to Metal ker
Rethinking Model Loading: safetensors, Portability, and IBM Spyre
<p>safetensors is widely used to distribute model checkpoints, especially through Hugging Face. fastsafetensors is a high-performance loader that reduces loading overhead through optimized file access and device transfers. The speaker is a maintainer of fastsafetensors, which is integrated with PyTorch systems including vLLM and SGLang.</p><p>Portable checkpoints do not always match the tensor formats accelerators need. Loading may require data-type and memory-layout conversions in addition to I
Where Do RL Environments Come From? Turning Real Data into Tasks, Rewards and Training
<p>Frontier labs post-train models across thousands of RL environments. At their core, these environments expose two things: tasks for models to solve and verifiers that determine whether they succeed. The challenge is scaling both: generating diverse, high-quality tasks from real-world data and building robust verifiers that provide reliable rewards. Yet very little of this work is being done openly at scale.<br>We walk through four key decisions in building RL environments: choosing domains wi
When the KV Cache Cannot Roll Back: Prefix Reuse on Android
<p>An agent prompt on a phone can open with about 2,050 tokens of harness, 1,700 of them tool definitions. On a Dimensity 9500s, a 1,036-token prefix takes 9.3 s to prefill on LFM2.5-1.2B and 45 s on SmolLM3-3B; faster decode does nothing to that wait. We solved it with prefix reuse, built twice because our open source app ships two runtimes with different cache constraints.</p><p>ExecuTorch's Android API exposes prefill and full reset, but no partial rollback. We prefill the shared head at
PyTorch Day Japan
Py Torch Day Japan 2026, hosted by the Py Torch Foundation, Hugging Face, IBM, and Mitsubishi Electric, is a pivotal event for community-driven learning, networking, and open source AI innovation.