PyTorch Day China 2025
2025-06-07
Beijing, China
| title | author | length |
|---|---|---|
| Keynote: Welcome & Opening Remarks | Matt White | 15m |
PyTorch Foundation announces it is now an umbrella foundation hosting projects beyond core, with over 30 members and two new hosted projects added recently VLM… | ||
| verl: An Open Source Large Scale LLM RL Framework for Agentic Tasks | Yuxuan Tong | 24m |
verl is an open source RL framework for large-scale LLM agentic tasks that claims to let users implement GRPO/PO-style algorithms in a few lines via a… | ||
| torch.accelerator: A Unified, Device-Agnostic Runtime API for Stream-Based Accelerators | Yu Guangye | 18m |
Proposal introduces a unified, device-agnostic runtime in PyTorch 2.6 that maps one-to-one to existing device APIs so code needs minimal changes, with a C++… | ||
| vLLM: Easy, Fast, and Cheap LLM Serving for Everyone | Kaichao You | 15m |
vLLM claims a 10x speedup over Hugging Face and 2–3x vs TGI by reorganizing KV cache into block-based page attention, enabling much higher GPU memory… | ||
| A torch.fx Based Compression Toolkit Empowered by torch_musa | Fan Mo | 09m |
NeuroTrim is a lightweight FX-based compression toolkit that runs on native PyTorch via torch_musa and the MUSA backend, producing quantized models that can… | ||
| PyTorch in Production: Boosting LLM Training and Inferencing on Ascend NPU | Jiawei Li, Li Jing | 11m |
PyTorch in Production: Boosting LLM Training and Inferencing on Ascend NPU - Jiawei Li & Li Jing, Huawei Technologies Co., Ltd.
Deploying PyTorch in… | ||
| SGLang: An Efficient Open-Source Framework for Large-Scale LLM Serving | Liangsheng Yin | 20m |
SGLang is an open source LLM inference engine that implements PD disaggregation and hierarchical KV cache offloading, adds large-scale expert parallelism and… | ||
| Keynote: Running Large Models on Diverse AI Chips: PyTorch + Open-Source Stack (FlagO... | Yonghua Lin | 25m |
Keynote: Running Large Models on Diverse AI Chips: PyTorch + Open-Source Stack (FlagOS) for Architecture-Free Deployment - Yonghua Lin, BAAI | ||
| Diving in Hugging Face Hub; Share Your Model Weights on the #1 AI Hub, Home of 700k+... | Tiezhen Wang | 14m |
Hugging Face Hub hosts 1.7 million public models and adds 10x growth in three years while offering concrete tools: model hub pages with download counts and… | ||
| PyTorch in China: Community Growth, Localization, and Interaction | Zesheng Zong | 08m |
PyTorch adoption in China is growing with concrete outputs: GitHub shows over 90,000 stars and the ecosystem reports over 3,800 core contributors. A community… | ||
| The Development of AI Open Source and Its Influence on the AI Ecosystem | Jason Li | 31m |
The Development of AI Open Source and Its Influence on the AI Ecosystem - Jason Li, CSDN & Boolan | ||
| Efficient Training of Video Generation Foundation Model at ByteDance | Xiaonan Nie, Heng Zhang | 14m |
Efficient Training of Video Generation Foundation Model at ByteDance - Xiaonan Nie, ByteDance Seed & Heng Zhang, Bytedance
Video generation models have… | ||
| torch.compile Practice and Optimization in Different Scenarios | Yichen Yan | 16m |
Practical fixes and passes extend torch.compile (torch.dynamo + autograd + inductor) to real workloads: DeepSpeed + gradient checkpointing compatibility is… | ||
| Galvatron: An Automatic Distributed Training System for Efficient Large... | Xinyi Liu, Fangcheng Fu | 16m |
Galvatron is a PyTorch-native framework that profiles a Transformer’s execution to build a PyTorch profiler cost model, then prunes a huge hybrid-parallel… | ||
| Intel's PyTorch Journey: Promoting AI Performance and Optimizing Open-Source Software | Mingfei Ma | 18m |
Intel's PyTorch Journey: Promoting AI Performance and Optimizing Open-Source Software - Mingfei Ma, Intel Asia-Pacific Research & Development Ltd.
PyTorch is… | ||
| FlagTree: Unified AI Compiler for Diverse AI Chips | Chunlei Men | 20m |
FlagTree: Unified AI Compiler for Diverse AI Chips - Chunlei Men, Beijing Academy of Artificial Intelligence aka Zhiyuan Institute
In the past two years, the… | ||
| KTransformers: Unleashing the Full Potential of CPU/GPU Hybrid Inference for M... | Dr. Mingxing Zhang | 21m |
KTransformers: Unleashing the Full Potential of CPU/GPU Hybrid Inference for MoE Models - Dr. Mingxing Zhang, Tsinghua University
Due to their sparse nature… | ||
| nothing matches | ||