Tag
PyTorch Conference North America 将于10月20-21日在圣何塞举行，预计有400+演讲者和3,500+参会者，主题涵盖训练、推理、智能体、内核和开源AI。
PyTorch Conference North America 2026 将于 10 月 20-21 日在加州圣何塞举行，由 PyTorch 基金会主办，Google 为钻石赞助商，议题涵盖从裸机基础设施到基于智能体的系统的整个 AI 技术栈。
NVIDIA的Guray Ozen将在PyTorch Conference North America 2026上展示CUTLASS for Python的最新进展，包括CuTe扩展、任务调度器和编译器诊断等面向开发者与AI代理的高效GPU内核编写新功能。
MaRN (Mapping Networks) is a new open-source PyTorch library that trains neural networks by optimizing a compact latent representation instead of all parameters directly, showing up to 57.7× parameter reduction on MNIST with modest accuracy trade-offs. The author presents exploratory benchmarks and invites community feedback on the approach.
AWS 宣布 PyTorch 现已原生支持 AWS Trainium 芯片，并将在 PyTorch Conference North America（10月20-21日，圣何塞）由 Amazon VP Distinguished Engineer Robert Hundt 发表主题演讲、AWS Neuron Principal PM Maen Suleiman 进行现场演示。
NVIDIA recsys-examples now offers an end-to-end HSTU generative recommender inference workflow on Dynamo-Triton, combining PyTorch AOTI compilation with FlexKV-backed KV caching to achieve up to 5.93x lower latency on an RTX PRO 6000 Blackwell GPU.
PyTorch 博客介绍 NVIDIA Dynamo 如何通过统一的 session 级标识符，将请求级推理服务系统升级为程序感知系统，实现会话感知路由、共享 KV 缓存索引与跨 vLLM/SGLang 的程序化缓存迁移，以高效服务具有长上下文和并行子代理的 agentic 工作负载。
AMD promotes its ROCm platform at PyTorch Conference North America 2026 in San Jose (Oct 20-21), featuring hands-on labs, technical sessions, and open-source AI innovation across cloud and client environments.
PyTorch Foundation 博客介绍了 IBM 如何利用 PyTorch 现有的设备、流、内存管理和编译抽象，将 Spyre 数据流 AI 加速器集成成为原生 PyTorch 设备，从而在无需为每种硬件单独开发框架的情况下支持更广泛的 AI 硬件。
PyTorch Conference North America在圣何塞举办，Steve Pousty介绍今年新增的入门分论坛、海报展示、展位及协同活动，并将分享基于PyTorch的卫星影像遥感应用。
PyTorch Conference North America 宣布 AWS 副总裁兼杰出工程师 Robert Hundt 将担任主题演讲嘉宾，分享 Trainium 如何获得原生 PyTorch 支持，涵盖 eager mode、torch.compile、TorchTitan 与 TorchAO。会议将于 10 月 20-21 日在圣何塞举行。
Rgpu 是一个开源工具，可通过 TCP 让 PyTorch 张量和现有 CUDA 程序在远程 NVIDIA GPU 上执行计算，客户端应用保持在本地运行，提供 PyTorch 设备接口和 CUDA shim 两条路径。
Red Hat, NVIDIA AI, and IBM are hosting 'PyTorch, Powering the Enterprise,' a PyTorchCon North America Day 0 event on October 19 exploring how PyTorch and its ecosystem (vLLM, Ray, Helion, llm-d) power enterprise AI at scale, with talks on AI infrastructure, deployment, security, and the economics of serving inference.
Meta has consolidated PyTorch's media processing stack over two years, making TorchCodec the single home for CPU and CUDA decoding/encoding of images, video, and audio, while TorchVision and TorchAudio now focus strictly on high-performance media transforms.
PyTorch Conference North America (Oct 20-21, San Jose) features a keynote by Priya Nagpurkar of IBM Research on IBM and Red Hat's contributions to the PyTorch ecosystem, including out-of-tree accelerator infrastructure, Spyre Accelerator enablement, and llm-d distributed inference work.
Astral's GPU wheel indexes now support PyTorch 2.13 and 2.14 and Python 3.15, with updated builds of FlashAttention, FlashAttention 3, causal-conv1d, DeepSpeed and other GPU packages across CUDA versions.
PyTorch's Accelerator Integration Working Group published a technical recap of H1 2026 progress, including the Cross-Repository CI Relay (CRCR), refactored device-agnostic test suites, maturation of the OpenReg reference backend, distributed and torch.compile support, and the new Additional Compute Platforms portal to standardize vendor-neutral hardware enablement.
PyTorch's Helion kernel DSL is integrated into vLLM's linear backend, where a single autotuned GEMM implementation covers Standard GEMM, Split-K, and Swap-AB variants, outperforming CUTLASS and DeepGEMM backends on NVIDIA Hopper GPUs with over 10% throughput gains on some workloads.
Cohere VP of Engineering Rachad Alao is announced as a keynote speaker at PyTorch Conference North America (Oct 20-21, San Jose), discussing how open science can help AI foundation models advance with transparency, security, speed, and scale.
An announcement for a PyTorch Conference North America talk where speakers from Google and Meta will present TorchTPU, a PyTorch-native TPU backend that lets vLLM and SGLang run natively on TPU while preserving existing schedulers, batching systems, OpenAI-compatible APIs, and torch.compile workflows.