Tag
NVIDIA reports significant speedups with PyTorch AOTI backend in HSTU inference tests, up to 2.38x in ideal scenarios, and highlights tools like recsys-examples and nv-embedding-cache for generative recommender systems at scale.
TorchSpec is a PyTorch-native framework for training speculative decoding draft models, released in collaboration with vllm and demonstrated with Kimi K3 draft models on NVIDIA GB200 hardware.
PyTorch 2.14 release introduces fault tolerance as a first-class c10d concept with in-place process-group reconfiguration, alongside features like NVGEMM kernels, a new nccl2 backend, and native linear algebra for Apple Silicon.
PyTorch Conference North America is an event in San Jose focused on agentic AI and technical challenges, featuring speakers from leading tech organizations.
PyTorch 2.14 introduces updates to Inductor, PyTorch Distributed, dynamic shapes, Apple Silicon support, and more, with a live Q&A and PyTorchCon announcement.
A hands-on guide to learning reinforcement learning from scratch, focusing on the CartPole environment using PyTorch and Stable Baselines3 libraries.
The article discusses using NVIDIA FLARE to build federated multimodal AI workflows, addressing challenges with distributed data through techniques like externalization and tensor streaming to optimize model distribution and reduce memory requirements.
This paper details the architecture of the Kimi K3 model (featuring 2.8 trillion parameters and a 1 million token context) released by Moonshot AI, and demonstrates how to implement its seven core innovations from scratch using PyTorch, including KDA recurrent attention, gated MLA, and stable latent MoE.
AMD releases ROCm 10.0, a major update to its open-source GPU compute platform, marking a decade and introducing native agentic AI developer experience with ROCm.AI.
The PyTorch Conference North America 2026 will feature discussions with experts on PyTorch and related projects, focusing on advancements in AI frameworks, inference engines, and hardware integration.
NVIDIA demonstrates how quantization-aware distillation (QAD) using NVIDIA Model Optimizer improves the Nemotron 3.5 Lightning model, reducing memory usage and increasing throughput while preserving accuracy for agentic benchmarks.
This paper introduces DataKernelBench, a benchmark for evaluating LLMs on optimizing GPU kernels for database queries, achieving speedups over baseline methods like torch.compile.
The inaugural PyTorch Day Japan event is announced, to be hosted by PyTorch Foundation, Hugging Face, IBM, and Mitsubishi Electric in Tokyo on December 10, featuring technical talks on AI training, inference, and responsible AI.
TRANSIT is a runtime that makes unified virtual memory practical for large-scale LLM training, reducing GPU usage by up to 50% without code changes, and will be presented at PyTorch Conference North America.
The keynote lineup for KubeCon, CloudNativeCon, OpenInfraSummit, and PyTorchCon China in Shanghai has been announced, featuring speakers from organizations like The Linux Foundation and NVIDIA.
TimesFM 3.0 is a pretrained time-series foundation model by Google Research, released on Hugging Face with PyTorch weights for time-series forecasting tasks.
The PyTorch Conference North America has announced its keynote lineup, highlighting four main directions for the open-source ecosystem, including strengthening the core and expanding multi-hardware support, with speakers from major tech organizations.
PyTorch provides a native fine-tuning library with Day-0 Hugging Face checkpoint support for Alibaba's open-weight Qwen3.8-2.4T-A95B model, enabling efficient training and deployment on NVIDIA systems with configurable reasoning.
Edward Yang will be a keynote speaker at PyTorchCon North America, discussing the evolution and future of PyTorch. The event is set for October 20-21 in San Jose, CA.
AI coding agents write runtime adapters to bridge gaps between new model architectures and software stacks, enabling thousands of HuggingFace models to run on the IBM Spyre accelerator with day-one support.