PyTorch

AMD and Intel have agreed to integrate their PyTorch optimization libraries for x86 CPUs into 'oneDNN'.

• •AMD and Intel announced on October 20 (local time) that they will collaborate to integrate their x86 CPU PyTorch optimizations.
• •The optimization libraries developed by both companies will be unified into the common framework 'oneDNN'.
• •This integration is expected to improve PyTorch execution efficiency in x86 CPU environments.

At PyTorch Conference 2026 in California, AWS and Google Cloud sequentially announced open-source versions of PyTorch supporting their respective AI processors, aiming to facilitate migration from NVIDIA.

• •Two cloud providers announced new versions at PyTorch Conference 2026 in California.
• •The released PyTorch versions are open-source and support AWS Trainium and Google Cloud TPU.
• •These announcements are intended to promote a transition away from NVIDIA.

A video course of over 11 hours of coding videos on building reasoning models, covering evaluation, inference-time scaling, reinforcement learning, and distillation using Python and PyTorch.

• •It consists of over 11 hours of coding videos to learn how to build reasoning models from scratch.
• •It covers key topics such as model evaluation, inference-time scaling, reinforcement learning, and distillation.
• •Python and PyTorch are used for the implementation.

Amazon is launching a new development tool called TorchNeuron to break Nvidia's hold on PyTorch and encourage developers to run AI apps on alternative chips, including its own.

• •Amazon aims to break Nvidia's dominance in the free PyTorch toolkit used for building most AI apps.
• •The goal is to enable the PyTorch environment to support alternative chips.
• •The company seeks to increase the number of developers running AI applications on alternative chips, including Amazon's own.

最新のLLMアーキテクチャをPyTorchで実装したオープンソースプロジェクト「OpenArch」が紹介された。研究者や開発者が最新のモデル構造を学習・実験するための教育用リソースとして機能する。

• •現代の主要なLLMアーキテクチャをPyTorchフレームワークを用いて実装したコードベースを提供している
• •複雑なモデル構造を簡潔に実装することで、AI研究者やエンジニアがアーキテクチャを深く理解できる
• •新しいモデルの試作や既存モデルの比較実験を行う際の実用的な基盤コードとして活用できる