pytorch

PyTorch Con 2026 主题演讲预告：Agentic AI Foundation 执行董事 Mazin Gilbert 探讨 LLM 如何从智能体行为中学习，提出五层智能体栈架构，并将智能体评估反馈为 PyTorch 训练奖励的开放研究方向。

Part 1 of Amazon's tutorial series on profiling TorchNeuron workloads on AWS Trainium, covering how to configure torch.profiler with the Neuron backend, read key_averages() output, and trace operations through Neuron Explorer.

Google and Meta collaborate on TorchTPU, an open-source native PyTorch backend for Google TPUs that enables out-of-the-box execution, eager mode support, and torch.compile integration with competitive performance, replacing the limitations of PyTorch/XLA.

Intel and AMD jointly announced they will collaborate on x86 CPU software optimizations for PyTorch via oneDNN, with AMD contributing its ZenDNN improvements upstream and joining the UXL Foundation as a steering member.

PyTorch's blog details the open-source MLIR-based compiler backend (KTIR to DataflowIR to SpyreCode) that translates PyTorch Inductor's decisions into executable dataflow pipelines for IBM's Spyre AI accelerator.

Google released TorchTPU as an open source project, letting developers run native PyTorch apps on its TPUs without rewriting code, joining Amazon's TorchNeuron in challenging Nvidia's dominance over AI chip software tooling.

Google 与 Meta 联合发布开源项目 TorchTPU，为 Google TPU 提供原生 PyTorch 后端，支持真正的 eager 执行与 torch.compile 集成，并在 Llama 3 405B 和 Qwen 3 MoE 上实现近线性扩展。

Ujval Kapasi of NVIDIA delivers a PyTorchCon 2026 keynote on what agentic workloads demand from the PyTorch stack, using NVIDIA Vera Rubin as a case study for hardware-software co-design across CPUs, GPUs, memory, and networking, covering optimizations in PyTorch, Triton, and vLLM.

Live from PyTorch Con 2026, speakers from IBM, Google Cloud, and Red Hat present "Open Source Inference at Production Scale" with a live LLM demo underway.

Live from #PyTorchCon 2026, Amazon's Robert Hundt introduces TorchNeuron, the native PyTorch backend for AWS Trainium, supporting eager execution and torch.compile for training and serving via vLLM-Neuron, plus Neuron Explorer for profiling down to hardware instructions.

Live from PyTorchCon 2026, Chris Lattner of Qualcomm shows how developers can extend PyTorch with MAX and Mojo, offering a shared foundation for running code across hardware from multiple vendors. Mojo blends Pythonic code with systems-level speed, while MAX provides model authoring and a graph compiler targeting diverse hardware.

Google open-sources TorchTPU, a native PyTorch backend for its TPUs, featuring custom ATen kernels, XLA compilation caching, torch.compile() integration, and distributed training support with examples for models like Llama, Qwen, and ResNet.

PyTorch 2.15 brings major MPS backend improvements for Mac users, including FP8 support, 5-6.5× faster MoE model decoding, 8× faster softmax, and dramatic memory reductions for video VAE decoding, driven by Hugging Face's PyTorch team and community contributors.

Hugging Face 宣布 Transformers、Diffusers 和 Kernels 原生支持 TorchTPU，使 TPU 成为 PyTorch 中的一个普通设备选项，无需再依赖 torch_xla 或 Optimum TPU。

AWS 与 PyTorch 宣布推出 TorchNeuron——一个面向 AWS Trainium 芯片的原生 PyTorch 开源后端，支持 eager 执行、torch.compile 与 NKI 自定义内核，模型代码只需 `.to('neuron')` 即可运行，将于 10 月 20 日以 Apache 2.0 公开 Beta。

Amazon Web Services 发布了 TorchNeuron 开发工具，让 PyTorch 应用仅需一行配置即可迁移到其 Trainium AI 芯片上，直接挑战 Nvidia 在 AI 软件生态上的主导地位，并将相关改动贡献给 PyTorch 上游项目。

PyTorch 官方正在对 PyTorch Conference North America 2026 的同期活动 Jupyter Day 进行现场报道，数据科学家、开发者和研究人员齐聚一堂，探讨交互式计算与 AI 辅助工作流。

Amazon 发布 TorchNeuron，让现有 PyTorch 代码无需重写即可原生运行在 AWS Trainium 芯片上，并将于 10 月 20 日 PyTorch Conference 上开源。该工具旨在降低开发者切换 AI 硬件的成本，已有多家企业在机器人、自动驾驶、大模型等领域采用。

PyTorch Helion now integrates NVIDIA CompileIQ's Advanced Control Files into its autotuner, enabling joint search over kernel configurations and low-level ptxas compiler controls to extract last-mile performance from already autotuned GPU kernels.

PyTorch and AWS announce TorchNeuron, an open-source native PyTorch backend for AWS Trainium chips, letting developers run existing PyTorch code with a simple `.to('neuron')` switch, supporting eager execution, torch.compile, and NKI custom kernels. It will launch as public beta under Apache 2.0 at the PyTorch Conference on Oct. 20.