# PyTorch

PyTorch is an optimized tensor library for deep learning using GPUs and CPUs.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## AMD ROCm 10.2 Brings hipFile to PyTorch, LVM, and mdraid Volumes, and Adds 18 Ryzen Processors

DevFeed: [AMD ROCm 10.2 Brings hipFile to PyTorch, LVM, and mdraid Volumes, and Adds 18 Ryzen Processors](<https://devfeed.tech/articles/amd-rocm-10-2-brings-hipfile-to-pytorch-lvm-and-mdraid-volumes-and-adds-18-ryzen-processors-146856.md>)

Original publisher: [Read original article](<https://www.storagereview.com/news/amd-rocm-10-2-hipfile-pytorch-lvm-mdraid-aiter-ryzen>)

Author: Harold Fritts

Published: 2026-11-20T16:18:56Z

Content type: release

Language: en

Sources: [StorageReview.com](<https://devfeed.tech/sources/storagereview-com.md>)

Topics: [rocm](<https://devfeed.tech/topics/rocm.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [vllm](<https://devfeed.tech/topics/vllm.md>), [LVM](<https://devfeed.tech/topics/lvm.md>), [Warp](<https://devfeed.tech/topics/warp.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [amd](<https://devfeed.tech/tags/amd.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [lvm](<https://devfeed.tech/tags/lvm.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [rocm](<https://devfeed.tech/tags/rocm.md>), [ryzen](<https://devfeed.tech/tags/ryzen.md>), [volumes](<https://devfeed.tech/tags/volumes.md>)

### AI overview

ROCm 10.2.0 extends AMD Infinity Storage and hipFile into PyTorch, fio, and ior, with support for LVM volumes and Linux mdraid sets. PyTorch's native `load_storage` and `save_storage` APIs move large tensors between GPU memory and storage; hipFile is integrated in fio 3.43 and ior's main branch for benchmarking that path. The stability-focused release also updates vLLM and SGLang, adds AITER support for Radeon and Ryzen AI, and supports 18 additional Ryzen processors. Other changes include expanded GPU libraries and an experimental unified `rocsys` profiler command. AMD's cited MLPerf and scaling figures are vendor-reported and were not independently audited; `rocsys` is experimental and Radeon AI Pro R9700 Hyperloom support is early access.

### Source excerpt

AMD has released ROCm 10.2.0, a stability-focused update to its open GPU software stack that refreshes the serving frameworks, libraries, and tools developers already run on Instinct, Radeon, and Ryzen hardware. For storage teams, hipFile and AMD Infinity Storage now plug into PyTorch, fio, and ior, and they work on volumes built with LVM and The post AMD ROCm 10.2 Brings hipFile to PyTorch, LVM, and mdraid Volumes, and Adds 18 Ryzen Processors appeared first on StorageReview.com.

## KernelOPT: Dispatch-Aware Agentic Search for GPU Kernel Optimization

DevFeed: [KernelOPT: Dispatch-Aware Agentic Search for GPU Kernel Optimization](<https://devfeed.tech/articles/kernelopt-dispatch-aware-agentic-search-for-gpu-kernel-optimization-144934.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2609.30059>)

Author: Aheli Poddar, Sanskar Prasad, Arindam Samanta, Subha Chakraborty, Vishal Goyal, Rohit Singh Rathaur

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Compilers and Linkers](<https://devfeed.tech/topics/compilers-and-linkers.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [arithmetic](<https://devfeed.tech/tags/arithmetic.md>), [artifacts](<https://devfeed.tech/tags/artifacts.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [automatically](<https://devfeed.tech/tags/automatically.md>), [challenge](<https://devfeed.tech/tags/challenge.md>), [changes](<https://devfeed.tech/tags/changes.md>), [code](<https://devfeed.tech/tags/code.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [compiler](<https://devfeed.tech/tags/compiler.md>), [compilers](<https://devfeed.tech/tags/compilers.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dispatch](<https://devfeed.tech/tags/dispatch.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [search](<https://devfeed.tech/tags/search.md>)

### AI overview

KernelOPT uses profiling-guided agents to inspect compiler artifacts, including generated Triton kernels and external library calls, then synthesize and refine kernels and evaluate the assembled model. Its four validation stages combine execution checks, multi-seed numerical comparisons, model-level checks with a float64 fallback, and an end-to-end performance gate of 1.03x. On 250 KernelBench problems run on NVIDIA H200, it reports geometric mean speedups over torch.compile of 1.40x for single operators, 1.15x for composite operators, and 1.07x for model architectures. The authors note that gains include precision changes and algebraic simplifications within evaluated tolerances, and that isolated kernel improvements can differ from end-to-end performance.

### Source excerpt

arXiv:2609.30059v3 Announce Type: replace-cross Abstract: Deep learning inference and training performance depends critically on GPU kernel efficiency. Modern compilers such as PyTorch Inductor automatically generate GPU kernels from high-level model code, but frequently underperform expert-written implementations by wide margins. Recent LLM-assisted optimizers explore these opportunities through kernel generation and refinement; yet translating isolated kernel improvements into model-level gains remains a challenge. We present KernelOPT, a profiling-guided multi-agent system for GPU kernel optimization. It inspects generated Triton kernels and external library calls in compiler artifacts to guide synthesis and refinement, then assembles and evaluates the resulting model. Four stages of validation combine execution checks, multi-seed numerical comparison, model-level checking with a float64 fallback and end-to-end performance gating ($\gamma{=}1.03$). The system supports PyTorch nn.Modules, standalone Triton kernels, and Helion kernels. It is evaluated on all 250 KernelBench problems on NVIDIA H200 which includes 100 single-operator problems (Level1), 100 composite operators (Level2), and 50 model architectures (Level3). KernelOPT achieves geometric mean speedups over torch.compile of 1.40$\times$, 1.15$\times$ and 1.07$\times$ respectively across L1, L2 and L3. Among problems exceeding 1.03$\times$ speedup, the geomeans are 2.54$\times$ (36 L1), 1.84$\times$ (23 L2), and 1.37$\times$ (11 L3). These gains include arithmetic-precision changes and task-specific algebraic simplifications under the evaluated numerical tolerances. The evaluation also reveals discrepancies between isolated kernel improvements and end-to-end performance, motivating model-level measurement alongside kernel-level optimization. Code is available at https://github.com/TorchedHat/KernelOPT.

## torchtest2: One-Call Sanity Checks for PyTorch Models,Evaluated on Real and Unseen Architectures

DevFeed: [torchtest2: One-Call Sanity Checks for PyTorch Models,Evaluated on Real and Unseen Architectures](<https://devfeed.tech/articles/torchtest2-one-call-sanity-checks-for-pytorch-models-evaluated-on-real-and-unseen-architectures-142145.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23273>)

Author: Grzegorz Gwardys

Published: 2026-11-19T05:00:00Z

Content type: news

Language: en

Sources: [arXiv: Software Engineering](<https://devfeed.tech/sources/arxiv-software-engineering.md>)

Topics: [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [Test-driven development](<https://devfeed.tech/topics/tdd.md>), [Transformers](<https://devfeed.tech/topics/transformers.md>), [timm](<https://devfeed.tech/topics/timm.md>)

Tags: [17](<https://devfeed.tech/tags/17.md>), [3](<https://devfeed.tech/tags/3.md>), [78](<https://devfeed.tech/tags/78.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [after](<https://devfeed.tech/tags/after.md>), [all](<https://devfeed.tech/tags/all.md>), [also](<https://devfeed.tech/tags/also.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [approach](<https://devfeed.tech/tags/approach.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [attention](<https://devfeed.tech/tags/attention.md>), [batch](<https://devfeed.tech/tags/batch.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [bootstrap](<https://devfeed.tech/tags/bootstrap.md>), [call](<https://devfeed.tech/tags/call.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [models](<https://devfeed.tech/tags/models.md>), [one](<https://devfeed.tech/tags/one.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>)

### AI overview

torchtest2 runs 18 PyTorch sanity checks on one batch with a single call, restores touched state, and uses plugins to adapt to model families. On 191 models from unseen architecture families, it detected 78% of injected bugs (95% CI 75-80%), compared with 40% for the original torchtest checks and 48% for a training monitor. It also found a batch-coupling bug in timm's GCViT, while real models revealed false alarms missed by the tool's tests and random architectures. Its reach is limited: only 3 of 28 reproduced Stack Overflow bugs and 3 of 18 TrainCheck silent errors were detectable by a pre-training tester. A median check took 1.0 second.

### Source excerpt

arXiv:2611.23273v1 Announce Type: new Abstract: Many bugs in deep learning code are silent: a layer that is never called, an optimizer created before the classification head was replaced, a recurrent layer that treats the batch as time, or an ignored attention mask. The code runs, the loss decreases somewhat, and the defect surfaces, if ever, after days of GPU time. We present torchtest2, an open-source library that runs 18 sanity checks on a single batch with one call, restores the PyTorch state it touches, and adapts to model families through plugins. We evaluate it by mutation analysis with 17 classes of silent bugs on 103 models from torchvision, timm and Hugging Face transformers, on 600 random architectures and on 20 hand-written realistic bugs, against the checks of the original torchtest, overfitting a single batch, a smoke test and a training monitor in the style of AUTOTRAINER and DeepDiagnosis. Every mutant is compared with its healthy counterpart, and intervals come from a bootstrap over models. After freezing the tool, we ran it on 191 models of architecture families it had never seen: counting, for every tool alike, a failing report where the healthy model's report does not fail, it detected 78% of the injected bugs (95% CI 75-80), against 40% for the original torchtest checks and 48% for the training monitor. On real models its batch-independence check found a bug in timm's GCViT that couples the images of a batch. Real models also exposed false alarms that neither the tool's own test suite nor random architectures had revealed. Two independent collections of real silent bugs mark the limits of the approach: only 3 of 28 reproduced PyTorch bugs from Stack Overflow and 3 of the 18 silent training errors released with TrainCheck are within reach of a pre-training model tester. A median check takes 1.0 s. We release the library, the benchmark and all raw results.

## tmol: a GPU-accelerated implementation of the Rosetta energy function for machine learning

DevFeed: [tmol: a GPU-accelerated implementation of the Rosetta energy function for machine learning](<https://devfeed.tech/articles/tmol-a-gpu-accelerated-implementation-of-the-rosetta-energy-function-for-machine-learning-141687.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.21857>)

Author: Andrew Leaver-Fay, Kieran Didi, Jeff Flatten, Alex Ford, Guangfeng Zhou, Nathaniel Corley, Hahnbeom Park, Maxwell W. Kazman, Rafael I. Brent, Joseph Kleinhenz, Henry Solberg, Andrew M. Watkins, Brian

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [AI Development](<https://devfeed.tech/topics/ai-development.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-accelerated](<https://devfeed.tech/tags/gpu-accelerated.md>), [library](<https://devfeed.tech/tags/library.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [machine-learning-models](<https://devfeed.tech/tags/machine-learning-models.md>), [protein](<https://devfeed.tech/tags/protein.md>), [protein-design](<https://devfeed.tech/tags/protein-design.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [reward](<https://devfeed.tech/tags/reward.md>), [rosetta](<https://devfeed.tech/tags/rosetta.md>), [throughput](<https://devfeed.tech/tags/throughput.md>)

### AI overview

tmol brings the RosettaGenFF forcefield into a GPU-native PyTorch library, enabling batched differentiable scoring and whole-structure refinement. The abstract reports up to a 362-fold throughput increase for batched processing. When used as a reward in generative protein design, tmol increased hydrogen bonds and water bridges at designed interfaces. The supplied text does not specify benchmark conditions or broader evaluation details.

### Source excerpt

arXiv:2611.21857v1 Announce Type: new Abstract: Physics-inspired forcefields complement generative machine learning models, showing better generalization beyond the training distribution. However, these forcefields are typically slow to evaluate, and their derivatives are difficult to integrate into ML frameworks. We introduce tmol, a GPU-native PyTorch library implementing the state-of-the-art RosettaGenFF forcefield. tmol supports efficient batched differentiable scoring and whole-structure refinement with up to 362-fold throughput increase (batched). Used as a reward in generative protein design, tmol increases hydrogen bonds and water bridges at designed interfaces.

## MegNIST, A Benchmark for Non-Invasive Inner Speech Decoding

DevFeed: [MegNIST, A Benchmark for Non-Invasive Inner Speech Decoding](<https://devfeed.tech/articles/megnist-a-benchmark-for-non-invasive-inner-speech-decoding-139164.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.21399>)

Author: Teyun Kwon, Pratik Somaiya, Gereon Elvers, Francesco Mantegna, Anna Camera, Sven Braeutigam, Mark Woolrich, Oiwi Parker Jones

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Decoding](<https://devfeed.tech/topics/decoding.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [reproducibility](<https://devfeed.tech/topics/reproducibility.md>)

Tags: [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmark-dataset](<https://devfeed.tech/tags/benchmark-dataset.md>), [classification](<https://devfeed.tech/tags/classification.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [format](<https://devfeed.tech/tags/format.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [python](<https://devfeed.tech/tags/python.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [recordings](<https://devfeed.tech/tags/recordings.md>), [speech](<https://devfeed.tech/tags/speech.md>), [tensors](<https://devfeed.tech/tags/tensors.md>)

### AI overview

MegNIST provides 12,000 MEG trials from one participant imagining the digits zero through nine across four sessions, offering a benchmark for non-invasive inner-speech decoding. Recordings have 306 channels sampled at 250 Hz and are split by experimental run into 10,000 training, 1,000 validation, and 1,000 test trials to preserve temporal independence. The data are available in BIDS and serialized-tensor formats; the description also notes example code and a PyTorch-compatible loader in pnpl. Baseline classifiers decode imagined digits significantly above chance. The evidence describes a single participant and digit task, so it does not establish performance across people or broader inner speech.

### Source excerpt

arXiv:2611.21399v1 Announce Type: new Abstract: MegNIST is a benchmark dataset for developing and evaluating machine learning methods for inner speech decoding from magnetoencephalography (MEG). The dataset comprises 12,000 trials of a single participant imagining speaking the digits zero through nine, recorded across four sessions. Each trial contains 306-channel MEG recordings sampled at 250 Hz, preprocessed using signal space separation, filtering, and downsampling. Data are provided as pre-defined training (10,000 trials), validation (1,000 trials), and test (1,000 trials) splits, with entire experimental runs assigned to each split to preserve temporal independence. The dataset is available in both BIDS format for neuroimaging analysis and as serialised tensors for machine learning workflows. Baseline classification results demonstrate that imagined digits can be decoded significantly above chance. Standardised splits, validated baselines, example code, and a PyTorch-compatible loader in the pnpl Python library enable straightforward integration with machine learning frameworks, lowering barriers for researchers to rapidly develop and benchmark inner speech decoding methods for non-invasive brain-computer interfaces.

## CrypTorch: Unified PyTorch-based Auto-tuning Compiler for Private Machine Learning Inference

DevFeed: [CrypTorch: Unified PyTorch-based Auto-tuning Compiler for Private Machine Learning Inference](<https://devfeed.tech/articles/cryptorch-unified-pytorch-based-auto-tuning-compiler-for-private-machine-learning-inference-139225.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2511.19711>)

Author: Jinyu Liu, Gang Tan, Kiwan Maeng

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Cryptography](<https://devfeed.tech/topics/cryptography.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Inference](<https://devfeed.tech/topics/inference.md>), [FHE](<https://devfeed.tech/topics/fhe.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [Code generation](<https://devfeed.tech/topics/code-generation.md>)

Tags: [compiler](<https://devfeed.tech/tags/compiler.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-pl](<https://devfeed.tech/tags/cs-pl.md>), [fhe](<https://devfeed.tech/tags/fhe.md>), [inference](<https://devfeed.tech/tags/inference.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [mpc](<https://devfeed.tech/tags/mpc.md>), [performance](<https://devfeed.tech/tags/performance.md>), [private](<https://devfeed.tech/tags/private.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [share](<https://devfeed.tech/tags/share.md>)

### AI overview

CrypTorch addresses the cost and accuracy limits of globally fixed operator approximations in MPC-based machine learning. Its compiler estimates costs from the target MPC backend and quality through proxy PyTorch execution, then auto-selects approximations to maximize performance under a quality constraint. The abstract reports 1.14-1.60x performance improvement across multiple MPC backends with no accuracy loss, or 1.17-1.92x with less than 3% accuracy degradation. Its shared compiler design supports multiple backends, and its auto-tuner also reduces bootstrapping operations by 1.31-1.65x on a related FHE backend.

### Source excerpt

arXiv:2511.19711v3 Announce Type: replace Abstract: MPC-based ML uses secure multi-party computation (MPC) to execute machine learning (ML) workloads across multiple parties without requiring any party to reveal its private data or model parameters. Our characterization study finds that the performance of state-of-the-art MPC frameworks is often dominated by operator-approximation strategies, where ML operators not natively supported by MPC are replaced with iterative methods or polynomials. These approximations are typically chosen heuristically by each framework and applied globally. As a result, they can be unnecessarily expensive in some cases, yet insufficiently accurate in others. We find that the best approximation choice depends on the context, target framework, and application, making a single globally fixed approximation fundamentally suboptimal. To address this problem, we design CrypTorch, a multi-stage compiler for MPC-based ML that supports multiple MPC frameworks as backends. At the core of CrypTorch is its auto-tuning stage, which extracts cost estimates from the target MPC backend and quality estimates from a proxy PyTorch execution, and uses them to automatically select a set of approximations that maximizes performance while satisfying the given quality constraint. We show that CrypTorch's auto-tuner improves performance across multiple MPC backends by 1.14--1.60x without degrading accuracy, and by 1.17--1.92x with less than 3% accuracy degradation. CrypTorch's unified design also improves programmability and usability by allowing multiple backends to share other common compiler functionality and optimizations. We further show that the auto-tuner extends to a related fully homomorphic encryption (FHE) backend, reducing the number of bootstrapping operations by 1.31--1.65x.

## Celebrating the 2026 PyTorch Foundation Contributor Award Winners and Nominees

DevFeed: [Celebrating the 2026 PyTorch Foundation Contributor Award Winners and Nominees](<https://devfeed.tech/articles/celebrating-the-2026-pytorch-foundation-contributor-award-winners-and-nominees-136553.md>)

Original publisher: [Read original article](<https://pytorch.org/blog/2026-pytorch-foundation-contributor-awards/>)

Author: PyTorch Foundation

Published: 2026-11-16T20:16:55Z

Content type: news

Language: en

Sources: [PyTorch](<https://devfeed.tech/sources/blog-pytorch.md>)

Topics: [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [open source maintenance](<https://devfeed.tech/topics/open-source-maintenance.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [accessible](<https://devfeed.tech/tags/accessible.md>), [ai](<https://devfeed.tech/tags/ai.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [automation](<https://devfeed.tech/tags/automation.md>), [award](<https://devfeed.tech/tags/award.md>), [awards](<https://devfeed.tech/tags/awards.md>), [blog](<https://devfeed.tech/tags/blog.md>), [bug](<https://devfeed.tech/tags/bug.md>), [bugs](<https://devfeed.tech/tags/bugs.md>), [buster](<https://devfeed.tech/tags/buster.md>), [catalyst](<https://devfeed.tech/tags/catalyst.md>), [categories](<https://devfeed.tech/tags/categories.md>), [communities](<https://devfeed.tech/tags/communities.md>), [community](<https://devfeed.tech/tags/community.md>), [conference](<https://devfeed.tech/tags/conference.md>), [contributor](<https://devfeed.tech/tags/contributor.md>), [foundation](<https://devfeed.tech/tags/foundation.md>), [keynote](<https://devfeed.tech/tags/keynote.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [pytorch-conference](<https://devfeed.tech/tags/pytorch-conference.md>), [stage](<https://devfeed.tech/tags/stage.md>)

### AI overview

The 2026 PyTorch Foundation Contributor Awards recognized 15 winners across nine categories, up from 11 winners in 2025, for work on PyTorch, vLLM, Ray, and DeepSpeed. Three new categories broadened recognition to documentation and education (Knowledge Enhancer), bug reporting and fixes (Bug Buster), and developer workflows, automation, infrastructure, and AI-supported development (Catalyst). The excerpt names winners by category and project and lists 11 additional nominees. The awards were presented at PyTorch Conference 2026 in San Jose; Red Hat sponsored the program.

### Source excerpt

The PyTorch Foundation is proud to celebrate the winners and nominees of the fourth annual PyTorch Foundation Contributor Awards! Presented on the keynote stage at PyTorch Conference 2026 in San... The post Celebrating the 2026 PyTorch Foundation Contributor Award Winners and Nominees appeared first on PyTorch.

## PyTorch Foundation Ambassador Community Highlights November 2026

DevFeed: [PyTorch Foundation Ambassador Community Highlights November 2026](<https://devfeed.tech/articles/pytorch-foundation-ambassador-community-highlights-november-2026-136554.md>)

Original publisher: [Read original article](<https://pytorch.org/blog/pytorch-ambassador-community-highlights-november-2026/>)

Author: PyTorch Foundation

Published: 2026-11-16T16:20:31Z

Content type: news

Language: en

Sources: [PyTorch](<https://devfeed.tech/sources/blog-pytorch.md>)

Topics: [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [coding-community](<https://devfeed.tech/topics/coding-community.md>), [Developer Events](<https://devfeed.tech/topics/developer-events.md>), [Learning](<https://devfeed.tech/topics/learning.md>), [developer-relations](<https://devfeed.tech/topics/developer-relations.md>), [open source maintenance](<https://devfeed.tech/topics/open-source-maintenance.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [academic](<https://devfeed.tech/tags/academic.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ambassador](<https://devfeed.tech/tags/ambassador.md>), [ambassador-post](<https://devfeed.tech/tags/ambassador-post.md>), [ambassadors](<https://devfeed.tech/tags/ambassadors.md>), [blog](<https://devfeed.tech/tags/blog.md>), [community](<https://devfeed.tech/tags/community.md>), [contribute](<https://devfeed.tech/tags/contribute.md>), [ecosystem](<https://devfeed.tech/tags/ecosystem.md>), [foundation](<https://devfeed.tech/tags/foundation.md>), [highlights](<https://devfeed.tech/tags/highlights.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [learn](<https://devfeed.tech/tags/learn.md>), [november-2026](<https://devfeed.tech/tags/november-2026.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [open-source-ai](<https://devfeed.tech/tags/open-source-ai.md>), [open-source-ai-ecosystem](<https://devfeed.tech/tags/open-source-ai-ecosystem.md>), [projects](<https://devfeed.tech/tags/projects.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [share](<https://devfeed.tech/tags/share.md>), [technical](<https://devfeed.tech/tags/technical.md>)

### AI overview

PyTorch Foundation Ambassadors organized meetups, workshops, lectures, and technical sessions across several countries, pairing practical PyTorch learning with open-source community building. Activities included a Kampala workshop on embeddings and semantic search (32 attendees), a Dhaka demonstration of native PyTorch on Google TPUs using free Colab notebooks (50 approved seats filled), and a Nepal session on speculative decoding for edge-server LLM inference (20 attendees). Other sessions introduced PyTorch for academic AI research, LLM application development, and open-source contributions. Ambassadors also shared work on training-inference consistency in LLM post-training, a Simplified Chinese PyTorch 2.14 article, and transfer-learning code using Torchvision's pretrained ResNet-18. The text reports individual event outcomes and resources, not an overall measure of community impact.

### Source excerpt

PyTorch Foundation Ambassadors bring people together to learn, share technical knowledge and contribute to PyTorch Foundation projects and the wider open-source AI ecosystem. In this November roundup, we highlight some... The post PyTorch Foundation Ambassador Community Highlights November 2026 appeared first on PyTorch.

## Energy-Based Models for Enterprise Decision-Making: Architecture and PyTorch Implementation

DevFeed: [Energy-Based Models for Enterprise Decision-Making: Architecture and PyTorch Implementation](<https://devfeed.tech/articles/energy-based-models-for-enterprise-decision-making-architecture-and-pytorch-implementation-136173.md>)

Original publisher: [Read original article](<https://hackernoon.com/energy-based-models-for-enterprise-decision-making-architecture-and-pytorch-implementation?source=rss>)

Author: M Sheik Uduman Ali

Published: 2026-11-16T13:16:17Z

Content type: article

Language: en

Sources: [HackerNoon](<https://devfeed.tech/sources/hackernoon.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [decision-making](<https://devfeed.tech/topics/decision-making.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [ai-decision-making](<https://devfeed.tech/tags/ai-decision-making.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [candidate](<https://devfeed.tech/tags/candidate.md>), [contrastive-learning](<https://devfeed.tech/tags/contrastive-learning.md>), [decision](<https://devfeed.tech/tags/decision.md>), [decision-intelligence](<https://devfeed.tech/tags/decision-intelligence.md>), [decision-making](<https://devfeed.tech/tags/decision-making.md>), [energy](<https://devfeed.tech/tags/energy.md>), [energy-based-ml-models](<https://devfeed.tech/tags/energy-based-ml-models.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [enterprise-ai](<https://devfeed.tech/tags/enterprise-ai.md>), [enterprise-decision](<https://devfeed.tech/tags/enterprise-decision.md>), [examples](<https://devfeed.tech/tags/examples.md>), [implementation](<https://devfeed.tech/tags/implementation.md>), [implicit-behavioral-cloning](<https://devfeed.tech/tags/implicit-behavioral-cloning.md>), [learn](<https://devfeed.tech/tags/learn.md>), [llms](<https://devfeed.tech/tags/llms.md>), [ml](<https://devfeed.tech/tags/ml.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [regression-models](<https://devfeed.tech/tags/regression-models.md>), [score](<https://devfeed.tech/tags/score.md>), [system-one-ai](<https://devfeed.tech/tags/system-one-ai.md>)

### AI overview

Energy-based models (EBMs) score candidate actions or plans for a given context; lower energy means a better fit, and the lowest-energy feasible candidate can be selected. In the delivery example, an ETA model predicts duration, while a separate cost rule chooses a customer promise: with lateness four times as costly as early arrival, the excerpt's discrete candidates favor 55 minutes. Quantile or probabilistic regression can handle uncertainty for a single ETA, so the article presents whole-plan scoring--such as driver, batching, route and promise--as the more distinctive EBM use case. Its PyTorch example concatenates context and candidate features, masks infeasible actions, and trains with cross-entropy over negative energies using preferred-action labels. Those labels must come from reliable outcomes, optimization, simulation, demonstrations or a utility function; historical choices alone may encode mistakes. Softmax scores over candidates are not necessarily calibrated success probabilities. The proposed architecture combines predictive ML, System One models for structured judgments, EBMs for candidate scoring, optimization and deterministic constraints, and LLMs for interpreting

### Source excerpt

Learn how energy-based models score candidate actions, support enterprise decisions, and work alongside LLMs and traditional ML, with PyTorch examples.

## mlr3torch: A Deep Learning Framework in R based on mlr3 and torch

DevFeed: [mlr3torch: A Deep Learning Framework in R based on mlr3 and torch](<https://devfeed.tech/articles/mlr3torch-a-deep-learning-framework-in-r-based-on-mlr3-and-torch-135680.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2604.18152>)

Author: Sebastian Fischer, Lukas Burk, Carson Zhang, Bernd Bischl, Martin Binder

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Deep learning](<https://devfeed.tech/topics/deep-learning.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Hyperparameter optimization](<https://devfeed.tech/topics/hyperparameter-optimization.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [data augmentation](<https://devfeed.tech/topics/data-augmentation.md>)

Tags: [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [classification](<https://devfeed.tech/tags/classification.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data-augmentation](<https://devfeed.tech/tags/data-augmentation.md>), [deep-learning](<https://devfeed.tech/tags/deep-learning.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [framework](<https://devfeed.tech/tags/framework.md>), [graph](<https://devfeed.tech/tags/graph.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [neural-networks](<https://devfeed.tech/tags/neural-networks.md>), [r](<https://devfeed.tech/tags/r.md>), [regression](<https://devfeed.tech/tags/regression.md>), [runtime](<https://devfeed.tech/tags/runtime.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>), [tabular-data](<https://devfeed.tech/tags/tabular-data.md>), [tensors](<https://devfeed.tech/tags/tensors.md>), [torch](<https://devfeed.tech/tags/torch.md>), [use-cases](<https://devfeed.tech/tags/use-cases.md>), [workflow](<https://devfeed.tech/tags/workflow.md>)

### AI overview

mlr3torch brings deep learning into R's mlr3 ecosystem, using torch to define, train, and evaluate neural networks for classification and regression on tabular data or tensors such as images. It provides predefined architectures, converts torch models into mlr3 learners, and lets users represent preprocessing, data augmentation, and network architecture together as graphs based on mlr3pipelines. Its integration supports resampling and benchmarking. The abstract also reports examples of hyperparameter tuning, fine-tuning, multimodal architectures, and runtime benchmarks, but gives no benchmark results.

### Source excerpt

arXiv:2604.18152v2 Announce Type: replace-cross Abstract: Deep learning (DL) has become a cornerstone of modern machine learning (ML) praxis. We introduce the R package mlr3torch, which is an extensible DL framework for the mlr3 ecosystem. It is built upon the torch package, and simplifies the definition, training, and evaluation of neural networks for both tabular data and generic tensors (e.g., images) for classification and regression. The package implements predefined architectures, and torch models can easily be converted to mlr3 learners. It also allows users to define neural networks as graphs. This representation is based on the graph language defined in mlr3pipelines and allows users to define the entire modeling workflow, including preprocessing, data augmentation, and network architecture, in a single graph. Through its integration into the mlr3 ecosystem, the package allows for convenient resampling, benchmarking, preprocessing, and more. We explain the package's design and features and show how to customize and extend it to new problems. Furthermore, we demonstrate the package's capabilities using three use cases, namely hyperparameter tuning, fine-tuning, and defining architectures for multimodal data. Finally, we present some runtime benchmarks.

## DRTriton: Large-Scale Synthetic Data Driven Reinforcement Learning for Triton Kernel Generation

DevFeed: [DRTriton: Large-Scale Synthetic Data Driven Reinforcement Learning for Triton Kernel Generation](<https://devfeed.tech/articles/drtriton-large-scale-synthetic-data-driven-reinforcement-learning-for-triton-kernel-generation-135675.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2603.21465>)

Author: Siqi Guo, Ming Lin, Tianbao Yang

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>)

Tags: [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [generation](<https://devfeed.tech/tags/generation.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [learning](<https://devfeed.tech/tags/learning.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>), [synthetic-data](<https://devfeed.tech/tags/synthetic-data.md>), [triton](<https://devfeed.tech/tags/triton.md>)

### AI overview

DRTriton trains LLMs to translate PyTorch programs into Triton kernels, which compile to CUDA at runtime. It combines CSP-DAG synthetic data generation with controlled difficulty, curriculum reinforcement learning using separate rewards for successful conversion and speed, and test-time search to improve execution speed. After supervised fine-tuning on a limited set of PyTorch-Triton pairs, DRTriton-7B reportedly achieved speedups over PyTorch on 92% of KernelBench Level 2 tasks, versus 23% for GPT-5.2 and 19% for Claude-Sonnet-4.5. These figures describe the reported benchmark results; the supplied excerpt does not establish performance beyond that evaluation.

### Source excerpt

arXiv:2603.21465v3 Announce Type: replace-cross Abstract: Developing efficient CUDA kernels is a fundamental yet challenging task in the generative AI industry. Recent research leverages Large Language Models (LLMs) to automatically convert PyTorch reference implementations to CUDA kernels, significantly reducing engineering effort. State-of-the-art LLMs, such as GPT-5.2 and Claude-Sonnet-4.5, still struggle with this task. To address this challenge, we propose DRTriton, a scalable learning framework for training LLMs to convert PyTorch programs into highly optimized Triton kernels, which are then compiled to CUDA kernels at runtime. DRTriton consists of three key components: (i) a data synthetic algorithm CSP-DAG that guarantees full coverage and unbiased uniform sampling over the operator space with controlled difficulty; (ii) a curriculum RL framework with decoupled rewards that jointly optimizes conversion success rate and execution speed; and (iii) a test-time search algorithm that further improves the execution speed of the generated Triton kernels. With a warmup stage of SFT on limited PyTorch-Triton pairs curated using existing LLMs, DRTriton trained by RL on synthesized PyTorch programs generalizes effectively to real-world CUDA kernels that are challenging even for human experts. Experimental results show that DRTriton-7B achieves speedup over PyTorch on 92% of KernelBench Level 2 tasks, compared to 23% for GPT-5.2 and 19% for Claude-Sonnet-4.5.

## Computing Effective Dimension for Quantum ML Models in PyTorch

DevFeed: [Computing Effective Dimension for Quantum ML Models in PyTorch](<https://devfeed.tech/articles/computing-effective-dimension-for-quantum-ml-models-in-pytorch-134418.md>)

Original publisher: [Read original article](<https://towardsdatascience.com/computing-effective-dimension-for-quantum-ml-models-in-pytorch/>)

Author: Adrian Panasiewicz

Published: 2026-11-15T13:00:00Z

Content type: tutorial

Language: en

Sources: [Towards Data Science](<https://devfeed.tech/sources/towards-data-science.md>)

Topics: [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [Quantum Computing](<https://devfeed.tech/topics/quantum-computing.md>), [AI Models](<https://devfeed.tech/topics/ai-models.md>)

Tags: [code](<https://devfeed.tech/tags/code.md>), [computing](<https://devfeed.tech/tags/computing.md>), [data](<https://devfeed.tech/tags/data.md>), [deep-learning](<https://devfeed.tech/tags/deep-learning.md>), [dimension](<https://devfeed.tech/tags/dimension.md>), [editor-s-picks](<https://devfeed.tech/tags/editor-s-picks.md>), [effective](<https://devfeed.tech/tags/effective.md>), [experiment](<https://devfeed.tech/tags/experiment.md>), [first](<https://devfeed.tech/tags/first.md>), [global](<https://devfeed.tech/tags/global.md>), [local](<https://devfeed.tech/tags/local.md>), [log](<https://devfeed.tech/tags/log.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [ml](<https://devfeed.tech/tags/ml.md>), [ml-models](<https://devfeed.tech/tags/ml-models.md>), [models](<https://devfeed.tech/tags/models.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [quantum](<https://devfeed.tech/tags/quantum.md>), [quantum-computing](<https://devfeed.tech/tags/quantum-computing.md>), [quantum-machine-learning](<https://devfeed.tech/tags/quantum-machine-learning.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [science](<https://devfeed.tech/tags/science.md>), [series](<https://devfeed.tech/tags/series.md>), [space](<https://devfeed.tech/tags/space.md>), [statistics](<https://devfeed.tech/tags/statistics.md>), [time](<https://devfeed.tech/tags/time.md>)

### AI overview

The article shows how to estimate effective dimension for quantum ML models in PyTorch using an empirical Fisher matrix and Monte Carlo parameter sampling. It distinguishes global sampling across quantum rotation angles in [0, 2π] from local sampling inside an ε-ball around trained weights; for uniform ball coverage, it samples a normalized Gaussian direction and scales the radius by εq^(1/k). It recommends skipping outcomes below a probability threshold and computing log-determinants with `torch.linalg.slogdet` and sample averages with `torch.logsumexp` to avoid numerical instability. In a PennyLane/PyTorch Mackey-Glass experiment, added circuit depth initially improved test loss and effective dimension, while deeper models showed diminishing gains and redundant parameter directions. The method is computationally expensive: runtime scales with parameter samples times data samples, and the article estimates several days for training and evaluation on high-end computers.

### Source excerpt

Local and global variants, Fisher estimation, epsilon-ball sampling, and log-space stability, with PennyLane code and a time-series experiment. The post Computing Effective Dimension for Quantum ML Models in PyTorch appeared first on Towards Data Science.

## MegaOps: A Kernel Overlapping Framework with PyTorch-like Modularity and Megakernel Performance

DevFeed: [MegaOps: A Kernel Overlapping Framework with PyTorch-like Modularity and Megakernel Performance](<https://devfeed.tech/articles/megaops-a-kernel-overlapping-framework-with-pytorch-like-modularity-and-megakernel-performance-131544.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.16248>)

Author: Jinkun Lin, Byungsoo Oh, Rachee Singh

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>)

Topics: [Code Modularity](<https://devfeed.tech/topics/code-modularity.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>)

Tags: [a100](<https://devfeed.tech/tags/a100.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [best](<https://devfeed.tech/tags/best.md>), [blocks](<https://devfeed.tech/tags/blocks.md>), [but](<https://devfeed.tech/tags/but.md>), [can](<https://devfeed.tech/tags/can.md>), [cause](<https://devfeed.tech/tags/cause.md>), [communication](<https://devfeed.tech/tags/communication.md>), [compose](<https://devfeed.tech/tags/compose.md>), [consumer](<https://devfeed.tech/tags/consumer.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cublas](<https://devfeed.tech/tags/cublas.md>), [cuda](<https://devfeed.tech/tags/cuda.md>), [framework](<https://devfeed.tech/tags/framework.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [like](<https://devfeed.tech/tags/like.md>), [modularity](<https://devfeed.tech/tags/modularity.md>), [performance](<https://devfeed.tech/tags/performance.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

MegaOps aims to retain kernel-per-operator modularity while removing whole-kernel dependency barriers. It launches dependent kernels on separate CUDA streams and synchronizes on individual intermediate-tensor tiles, allowing kernels with matching data tilings to be composed without modifying them. On dense Llama3 and Qwen3 across single H100, A100, and L40 GPUs and multi-GPU servers, the authors report throughput matching hand-tuned ThunderKittens megakernels; against the best-performing baseline, decode throughput is 1.09-1.10x and prefill throughput 0.98-1.10x. The framework reuses existing Triton, cuBLAS, and cuDNN kernels.

### Source excerpt

arXiv:2611.16248v1 Announce Type: new Abstract: Modern ML frameworks execute models using the kernel-per-operator paradigm, which runs each operator as a separate GPU kernel. This paradigm is modular, since developers can compose various kernel implementations and profile each kernel in isolation. However, this paradigm enforces dependencies at the granularity of entire kernels, which places a barrier between dependent kernels. These barriers cause streaming multiprocessors to idle, serialize computation and communication, and prevent a kernel from prefetching its weights while the preceding kernel runs. Megakernels remove kernel barriers by fusing the operators of an entire layer or model into a single kernel, but they give up the modularity of the kernel-per-operator paradigm. In this work, we ask whether a kernel-per-operator ML framework can perform on par with megakernels. Our key insight is that removing kernel barriers does not require fusing kernels. We develop MegaOps, a framework that launches dependent kernels on separate CUDA streams and enforces each dependency on individual tiles of the intermediate tensor that the producer kernel writes and the consumer kernel reads. Since kernels synchronize on data tiles instead of on each other's thread blocks, any two kernels that adopt the same data tiling can be composed without modifying either kernel. We evaluate MegaOps on dense Llama3 and Qwen3 models on single H100, A100, and L40 GPUs and on multi-GPU servers with 4$\times$A100 and 8$\times$H100. Our MegaOps framework matches the throughput of hand-tuned ThunderKittens megakernels, and achieves 1.09$\times$-1.10$\times$ the decode throughput and 0.98$\times$-1.10$\times$ the prefill throughput of the best-performing baseline, while reusing existing Triton, cuBLAS, and cuDNN kernels.

## GPUOS: A GPU Operating System Primitive for Transparent Operation Fusion

DevFeed: [GPUOS: A GPU Operating System Primitive for Transparent Operation Fusion](<https://devfeed.tech/articles/gpuos-a-gpu-operating-system-primitive-for-transparent-operation-fusion-131511.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2604.17861>)

Author: Yiwei Yang, Xiangyu Gao, Yuan Zhou, Yuhang Gan, Yusheng Zheng, Andi Quinn

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>)

Topics: [Warp](<https://devfeed.tech/topics/warp.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [Kernel](<https://devfeed.tech/topics/kernel.md>)

Tags: [architecture](<https://devfeed.tech/tags/architecture.md>), [attention](<https://devfeed.tech/tags/attention.md>), [broadcasting](<https://devfeed.tech/tags/broadcasting.md>), [code](<https://devfeed.tech/tags/code.md>), [concurrent](<https://devfeed.tech/tags/concurrent.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-os](<https://devfeed.tech/tags/cs-os.md>), [data-types](<https://devfeed.tech/tags/data-types.md>), [deep-learning](<https://devfeed.tech/tags/deep-learning.md>), [device-code](<https://devfeed.tech/tags/device-code.md>), [ecosystem](<https://devfeed.tech/tags/ecosystem.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [fusion](<https://devfeed.tech/tags/fusion.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [jit](<https://devfeed.tech/tags/jit.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [launch](<https://devfeed.tech/tags/launch.md>), [operating-system](<https://devfeed.tech/tags/operating-system.md>), [overhead](<https://devfeed.tech/tags/overhead.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [system](<https://devfeed.tech/tags/system.md>), [transparent](<https://devfeed.tech/tags/transparent.md>)

### AI overview

GPUOS targets workloads where repeated GPU kernel launches can cost more than the small tensor operations themselves. It keeps one persistent worker kernel running, processes tasks from a host-managed atomic queue, and uses NVIDIA NVRTC to compile operators at runtime and inject them through device function pointer tables. A dual-slot aliasing scheme supports concurrent operator updates without restarting the kernel. Through PyTorch's TorchDispatch, GPUOS batches micro-operations into unified submissions; its generic tensor abstraction supports varied shapes, strides, data types, and broadcasting. The abstract reports up to 15.3x speedup over standard PyTorch on small-operation-heavy workloads such as micro-batched inference and attention. These are abstract-level claims; no benchmark details are provided in the supplied text.

### Source excerpt

arXiv:2604.17861v2 Announce Type: replace Abstract: Modern deep learning workloads often consist of many small tensor operations, especially in inference, attention, and micro-batched training. In these settings, kernel launch overhead can become a major bottleneck, sometimes exceeding the actual computation time. We present GPUOS, a GPU runtime JIT system that reduces launch overhead using a persistent kernel architecture with runtime operator injection. GPUOS runs a single long-lived GPU kernel that continuously processes tasks from a host-managed work queue, eliminating repeated kernel launches. To support diverse operations, GPUOS uses NVIDIA NVRTC to just-in-time compile operators at runtime and inject them into the running kernel through device function pointer tables. This design enables operator updates without restarting the kernel or recompiling the system. GPUOS introduces four key ideas: (1) a persistent worker kernel with atomic task queues, (2) a runtime operator injection mechanism based on NVRTC and relocatable device code, (3) a dual-slot aliasing scheme for safe concurrent operator updates, and (4) transparent PyTorch integration through TorchDispatch that batches micro-operations into unified submissions. The system supports arbitrary tensor shapes, strides, data types, and broadcasting through a generic tensor abstraction. Experiments show that GPUOS achieves up to 15.3x speedup over standard PyTorch on workloads dominated by small operations, including micro-batched inference and attention patterns. GPUOS improves utilization while remaining compatible with the PyTorch ecosystem.

## Deploy frontier AI models on AWS Trainium with SGLang-Neuron

DevFeed: [Deploy frontier AI models on AWS Trainium with SGLang-Neuron](<https://devfeed.tech/articles/deploy-frontier-ai-models-on-aws-trainium-with-sglang-neuron-130933.md>)

Original publisher: [Read original article](<https://aws.amazon.com/blogs/machine-learning/deploy-frontier-ai-models-on-aws-trainium-with-sglang-neuron/>)

Author: Alex Nails

Published: 2026-11-12T22:03:34Z

Content type: article

Language: en

Sources: [Artificial Intelligence](<https://devfeed.tech/sources/artificial-intelligence.md>)

Topics: [AWS Trainium](<https://devfeed.tech/topics/aws-trainium.md>), [neuron](<https://devfeed.tech/topics/neuron.md>), [AWS AI chips](<https://devfeed.tech/topics/aws-ai-chips.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>)

Tags: [advanced-300](<https://devfeed.tech/tags/advanced-300.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-chips](<https://devfeed.tech/tags/ai-chips.md>), [ai-models](<https://devfeed.tech/tags/ai-models.md>), [ai-training](<https://devfeed.tech/tags/ai-training.md>), [ai-training-and-inference](<https://devfeed.tech/tags/ai-training-and-inference.md>), [amazon-ec2](<https://devfeed.tech/tags/amazon-ec2.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [aws](<https://devfeed.tech/tags/aws.md>), [aws-trainium](<https://devfeed.tech/tags/aws-trainium.md>), [container](<https://devfeed.tech/tags/container.md>), [deep-learning](<https://devfeed.tech/tags/deep-learning.md>), [deploy](<https://devfeed.tech/tags/deploy.md>), [frontier](<https://devfeed.tech/tags/frontier.md>), [frontier-ai](<https://devfeed.tech/tags/frontier-ai.md>), [frontier-ai-models](<https://devfeed.tech/tags/frontier-ai-models.md>), [models](<https://devfeed.tech/tags/models.md>), [neuron](<https://devfeed.tech/tags/neuron.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [serve](<https://devfeed.tech/tags/serve.md>), [sglang](<https://devfeed.tech/tags/sglang.md>), [trainium](<https://devfeed.tech/tags/trainium.md>)

### AI overview

SGLang-Neuron brings SGLang serving and reinforcement-learning rollouts to AWS Trainium through an AWS Deep Learning Container, while retaining SGLang's server interface and generation APIs. The out-of-tree plugin selects the Neuron platform with `SGLANG_PLATFORM=neuron`; SGLang core handles the API server, scheduler, request lifecycle, RadixAttention, and prefix-cache bookkeeping, while the plugin supplies Trainium-specific execution. It uses native PyTorch operations and Neuron Kernel Interface kernels compiled with `torch.compile`. For PD disaggregation, a prefill server transfers prompt KV cache to a decode server using NIXL over EFA. The walkthrough uses a Trainium3 EC2 instance and the `pytorch-inference-sglang-neuronx` container; supported models, features, and validated configurations are documented in the project resources.

### Source excerpt

SGLang-Neuron is now publicly available through an AWS Deep Learning Container. With SGLang-Neuron, you can serve models and generate reinforcement learning rollouts with SGLang on AWS Trainium using the same SGLang interface you already know, all running on native PyTorch.

## Training a 1T-Parameter MoE on EU public infrastructure with PyTorch and TorchTitan

DevFeed: [Training a 1T-Parameter MoE on EU public infrastructure with PyTorch and TorchTitan](<https://devfeed.tech/articles/training-a-1t-parameter-moe-on-eu-public-infrastructure-with-pytorch-and-torchtitan-131026.md>)

Original publisher: [Read original article](<https://pytorch.org/blog/training-a-1t-parameter-moe-on-eu-public-infrastructure-with-pytorch-and-torchtitan/>)

Author: The EUROPA consortium (Domyn, Fraunhofer IAIS and IIS)

Published: 2026-11-12T12:00:19Z

Content type: article

Language: en

Sources: [PyTorch](<https://devfeed.tech/sources/blog-pytorch.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [accelerators](<https://devfeed.tech/tags/accelerators.md>), [activation](<https://devfeed.tech/tags/activation.md>), [ai](<https://devfeed.tech/tags/ai.md>), [api](<https://devfeed.tech/tags/api.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [arm](<https://devfeed.tech/tags/arm.md>), [attention](<https://devfeed.tech/tags/attention.md>), [award](<https://devfeed.tech/tags/award.md>), [blog](<https://devfeed.tech/tags/blog.md>), [bottlenecks](<https://devfeed.tech/tags/bottlenecks.md>), [eu](<https://devfeed.tech/tags/eu.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [interconnect](<https://devfeed.tech/tags/interconnect.md>), [mixture-of-experts-moe](<https://devfeed.tech/tags/mixture-of-experts-moe.md>), [moe](<https://devfeed.tech/tags/moe.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [public](<https://devfeed.tech/tags/public.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

Using PyTorch and TorchTitan, the EUROPA consortium raised training throughput for a 1T-parameter MoE on JUPITER from 90 to about 1,300 tokens/s per GPU. The setup uses 384 GH200 GPUs per data-parallel replica, with four-way expert parallelism within each node and no pipeline parallelism, to keep MoE token all-to-all traffic on NVLink while managing limited inter-node bandwidth. Key changes included FlexShard placement so Muon can update whole experts without communication, activation-checkpoint offloading to Grace CPU memory to fit 65,536-token sequences, and running FSDP's all-gather setup on the compute stream to reduce allocator retries and free about 14 GiB of peak reserved HBM per GPU. The excerpt reports these optimizations and several PyTorch and TorchTitan contributions; it notes that a companion post covers additional techniques.

### Source excerpt

TL;DR Scaling pretraining for a ~1-trillion-parameter Mixture-of-Experts (MoE) model on shared, public infrastructure was previously considered unfeasible due to unique architectural and interconnect bottlenecks. However, by leveraging the fine-grained flexibility... The post Training a 1T-Parameter MoE on EU public infrastructure with PyTorch and TorchTitan appeared first on PyTorch.

## TorchCurves: Differentiable Parametric Curves

DevFeed: [TorchCurves: Differentiable Parametric Curves](<https://devfeed.tech/articles/torchcurves-differentiable-parametric-curves-128854.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.14516>)

Author: Alex Shtoff

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [procedural curves](<https://devfeed.tech/topics/procedural-curves.md>), [Library](<https://devfeed.tech/topics/library.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [vectorization techniques](<https://devfeed.tech/topics/vectorization-techniques.md>), [Maintainability](<https://devfeed.tech/topics/maintainability.md>), [SciPy](<https://devfeed.tech/topics/scipy.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [algorithms](<https://devfeed.tech/tags/algorithms.md>), [applications](<https://devfeed.tech/tags/applications.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [batches](<https://devfeed.tech/tags/batches.md>), [can](<https://devfeed.tech/tags/can.md>), [central](<https://devfeed.tech/tags/central.md>), [combines](<https://devfeed.tech/tags/combines.md>), [compare](<https://devfeed.tech/tags/compare.md>), [continuous](<https://devfeed.tech/tags/continuous.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [custom](<https://devfeed.tech/tags/custom.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [feature](<https://devfeed.tech/tags/feature.md>), [functional](<https://devfeed.tech/tags/functional.md>), [github](<https://devfeed.tech/tags/github.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [https](<https://devfeed.tech/tags/https.md>), [implementation](<https://devfeed.tech/tags/implementation.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>)

### AI overview

TorchCurves makes vector-valued parametric curves learnable in PyTorch: it maps batches of scalar inputs to curve points using B-spline or Legendre bases and differentiates through both inputs and coefficients. Coefficients can be optimized with a model or generated by another network. The library offers trainable and reusable basis modules plus a functional interface, with applications including continuous feature embeddings, Kolmogorov-Arnold networks, and functional regression. Its vectorized implementation uses classical evaluation algorithms without custom hardware kernels. Timing experiments compare evaluation with NumPy and SciPy, and KAN training with EfficientKAN; the abstract says these show both practical cases and substantial overhead, but gives no specific results.

### Source excerpt

arXiv:2611.14516v1 Announce Type: new Abstract: We present TorchCurves, a PyTorch library that treats vector-valued parametric curves as learnable primitives. Its central operation maps batches of scalar inputs to points on curves represented in B-spline or Legendre polynomial bases, with differentiation through both inputs and coefficients. Curve coefficients can be optimized jointly with the surrounding model or produced by another network. The library provides trainable modules, reusable basis modules, and a functional interface, supporting applications ranging from continuous feature embeddings to Kolmogorov-Arnold networks and functional regression. Its implementation combines classical evaluation algorithms with vectorized PyTorch operations, emphasizing readability and maintainability without custom hardware kernels. Timing experiments compare curve evaluation against NumPy and SciPy and KAN training against EfficientKAN across a range of problem sizes. The results illustrate where this generality remains computationally practical and where overhead is substantial. Library is available on GitHub at https://github.com/alexshtf/torchcurves.

## CP-MLPs: A Tensor-Rank Theory of Tied and Untied MLP Blocks

DevFeed: [CP-MLPs: A Tensor-Rank Theory of Tied and Untied MLP Blocks](<https://devfeed.tech/articles/cp-mlps-a-tensor-rank-theory-of-tied-and-untied-mlp-blocks-128745.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.13647>)

Author: Md Rifat Arefin, Farzaneh Heidari, Irina Rish, Guillaume Rabusseau

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [mlp](<https://devfeed.tech/topics/mlp.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>)

Tags: [approximation](<https://devfeed.tech/tags/approximation.md>), [blocks](<https://devfeed.tech/tags/blocks.md>), [border](<https://devfeed.tech/tags/border.md>), [budget](<https://devfeed.tech/tags/budget.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [mlp](<https://devfeed.tech/tags/mlp.md>), [rank](<https://devfeed.tech/tags/rank.md>), [sharing](<https://devfeed.tech/tags/sharing.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>), [theory](<https://devfeed.tech/tags/theory.md>)

### AI overview

CP-MLPs models each multiplicative unit as a rank-1 interaction that detects an input feature, reads a value, and writes to the output. In tied gated blocks such as ReLU², detector and value directions must match; untied blocks such as ReGLU and SwiGLU can separate them. Exact tensor-rank results show untied units use less width for indefinite quadratic interactions and role-asymmetric monomials, while tied units suit symmetric pure-power targets, so the advantage depends on the target. Experimental loss plateaus below exact-representation width do not by themselves prove a positive approximation gap, which is governed by border rank. In selected trained-model comparisons with matched input-projection budget, untied blocks retain an advantage.

### Source excerpt

arXiv:2611.13647v1 Announce Type: new Abstract: Transformer feed-forward blocks combine several design choices: squared activations, gating, and tied versus untied parameter sharing, whose separate contributions to expressivity remain poorly understood. We introduce CP-MLPs, a tensor framework in which each multiplicative unit contributes a rank-1 interaction with three roles: detecting an input feature, reading a value, and writing to the output. Tied gated blocks such as ReLU$^2$ force the detector and value roles to share the same direction; untied gated blocks such as ReGLU and SwiGLU let them differ. We show that this distinction yields exact, degree-wise tensor-rank separations: untied detector-value units are strictly more width-efficient for indefinite quadratic interactions and role-asymmetric monomials, while tied units are structurally matched to symmetric pure-power targets. The width advantage of untying therefore depends on the target structure. Experiments show loss plateaus below the exact-representation width, but these plateaus alone do not establish a positive approximation gap: arbitrarily accurate approximation is governed by border rank. In trained gated MLPs, untied blocks retain an advantage in selected comparisons at matched input-projection budget. Code is available at https://github.com/rarefin/CP-MLPs.

## ExecuTorch WebGPU: PyTorch Models on the GPU, in the Browser

DevFeed: [ExecuTorch WebGPU: PyTorch Models on the GPU, in the Browser](<https://devfeed.tech/articles/executorch-webgpu-pytorch-models-on-the-gpu-in-the-browser-129598.md>)

Original publisher: [Read original article](<https://pytorch.org/blog/executorch-webgpu-pytorch-models-on-the-gpu-in-the-browser/>)

Author: Julian Ng-Thow-Hing, Siddartha Pothapragada, Stephen Jia, Digant Desai, Bilgin Cagatay

Published: 2026-11-10T22:15:07Z

Content type: article

Language: en

Sources: [PyTorch](<https://devfeed.tech/sources/blog-pytorch.md>)

Topics: [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [lora](<https://devfeed.tech/topics/lora.md>)

Tags: [accelerated](<https://devfeed.tech/tags/accelerated.md>), [android](<https://devfeed.tech/tags/android.md>), [api](<https://devfeed.tech/tags/api.md>), [apple](<https://devfeed.tech/tags/apple.md>), [apple-m1](<https://devfeed.tech/tags/apple-m1.md>), [blog](<https://devfeed.tech/tags/blog.md>), [browser](<https://devfeed.tech/tags/browser.md>), [conversion](<https://devfeed.tech/tags/conversion.md>), [executorch](<https://devfeed.tech/tags/executorch.md>), [export](<https://devfeed.tech/tags/export.md>), [file](<https://devfeed.tech/tags/file.md>), [framework](<https://devfeed.tech/tags/framework.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [models](<https://devfeed.tech/tags/models.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [run](<https://devfeed.tech/tags/run.md>), [the-web](<https://devfeed.tech/tags/the-web.md>), [torch](<https://devfeed.tech/tags/torch.md>), [web](<https://devfeed.tech/tags/web.md>), [web-browser](<https://devfeed.tech/tags/web-browser.md>), [webgpu](<https://devfeed.tech/tags/webgpu.md>)

### AI overview

ExecuTorch's WebGPU backend runs PyTorch models in WebGPU-enabled browsers from a single `.pte` artifact exported with `torch.export`, without converting to another framework's format. A WebAssembly runtime dispatches WGSL compute shaders; the excerpt says the same artifact can also run natively through Metal, Vulkan, or D3D12. In Chrome on an Apple M1 Pro, tests on quantized Llama 3.2 1B and Qwen3-0.6B found WebGPU 4-7x faster than ExecuTorch's multithreaded XNNPACK WebAssembly CPU path on prefill and 2-10x faster on decode. The backend has been validated on language, vision, speech, and on-device fine-tuning workloads. Inference runs locally in the browser tab, keeping inputs and outputs on the device. The backend is still early in development, with operator, data type, and model coverage and stability identified as ongoing work.

### Source excerpt

TL;DR PyTorch models now run in the web browser on the GPU through ExecuTorch, with no conversion to another framework's format. You export with torch.export to an ExecuTorch .pte file,... The post ExecuTorch WebGPU: PyTorch Models on the GPU, in the Browser appeared first on PyTorch.

## AWS Launches TorchNeuron to Run PyTorch Models on AWS Trainium Without Modifications

DevFeed: [AWS Launches TorchNeuron to Run PyTorch Models on AWS Trainium Without Modifications](<https://devfeed.tech/articles/aws-launches-torchneuron-to-run-pytorch-models-on-aws-trainium-without-modifications-123022.md>)

Original publisher: [Read original article](<https://www.infoq.com/news/2026/11/aws-torchneuron-trainium-pytorch/>)

Author: Sergio De Simone

Published: 2026-11-09T11:00:00Z

Content type: release

Language: en

Sources: [InfoQ](<https://devfeed.tech/sources/infoq.md>)

Topics: [AWS Trainium](<https://devfeed.tech/topics/aws-trainium.md>), [neuron](<https://devfeed.tech/topics/neuron.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [AWS AI chips](<https://devfeed.tech/topics/aws-ai-chips.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-chips](<https://devfeed.tech/tags/ai-chips.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-ml-data-engineering](<https://devfeed.tech/tags/ai-ml-data-engineering.md>), [amazon](<https://devfeed.tech/tags/amazon.md>), [annapurna-labs](<https://devfeed.tech/tags/annapurna-labs.md>), [apis](<https://devfeed.tech/tags/apis.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [aws](<https://devfeed.tech/tags/aws.md>), [aws-torchneuron-trainium-pytorch](<https://devfeed.tech/tags/aws-torchneuron-trainium-pytorch.md>), [aws-trainium](<https://devfeed.tech/tags/aws-trainium.md>), [backend](<https://devfeed.tech/tags/backend.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [conference](<https://devfeed.tech/tags/conference.md>), [developers](<https://devfeed.tech/tags/developers.md>), [development](<https://devfeed.tech/tags/development.md>), [distributed-training](<https://devfeed.tech/tags/distributed-training.md>), [drug-discovery](<https://devfeed.tech/tags/drug-discovery.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [ml-data-engineering](<https://devfeed.tech/tags/ml-data-engineering.md>), [native](<https://devfeed.tech/tags/native.md>), [news](<https://devfeed.tech/tags/news.md>), [no-code](<https://devfeed.tech/tags/no-code.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [pytorch-conference](<https://devfeed.tech/tags/pytorch-conference.md>), [torchneuron](<https://devfeed.tech/tags/torchneuron.md>)

### AI overview

Amazon introduced TorchNeuron, a native PyTorch backend for AWS Trainium that aims to reduce migration work: with PyTorch automatic device detection, the text says switching from CUDA requires replacing `.to('cuda')` with `.to('neuron')`. It supports distributed training APIs including DDP, FSDP2 and DTensor, Torch streams and events, and custom performance-critical operators through the Neuron Kernel Interface. Amazon says it validated 100,000 popular models and more than 800 architectures. The backend is also presented as enabling shared hardware capacity for training and inference. The excerpt does not establish production performance or cost; cited observers say workload-specific performance, cost per token, reproducibility, and total cost of ownership remain to be tested.

### Source excerpt

At the recent PyTorch Conference North America 2026, Amazon introduced TorchNeuron, a native PyTorch backend that enables developers to run existing PyTorch code on AWS Trainium AI chips with virtually no code modifications. By Sergio De Simone

## AutoParallel: Composable, PyTorch-Native Automatic Parallelism for Production Training

DevFeed: [AutoParallel: Composable, PyTorch-Native Automatic Parallelism for Production Training](<https://devfeed.tech/articles/autoparallel-composable-pytorch-native-automatic-parallelism-for-production-training-123673.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.08660>)

Author: Kaijian Wang, Sanket Purandare, Aditya Venkataraman, Will Constable, Ivan Kobzarev, Elias Ellison, Simon Fan, Edward Z. Yang, Davide Italiano, Patrick Labatut, Yuke Wang, Francisco Massa

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>)

Topics: [distributed-training](<https://devfeed.tech/topics/distributed-training.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [Parallelism](<https://devfeed.tech/topics/parallelism.md>), [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>)

Tags: [automatic](<https://devfeed.tech/tags/automatic.md>), [compilation](<https://devfeed.tech/tags/compilation.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [native](<https://devfeed.tech/tags/native.md>), [parallelism](<https://devfeed.tech/tags/parallelism.md>), [planner](<https://devfeed.tech/tags/planner.md>), [production](<https://devfeed.tech/tags/production.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [throughput](<https://devfeed.tech/tags/throughput.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

AutoParallel jointly searches parallelization choices across a model's training computation and emits ordinary PyTorch programs using native abstractions, so it composes with PyTorch's compilation and training stack without a separate runtime. It can combine automatic decisions with model-specific distributed code, leaving dynamic or specialized regions under user control; plans are inspectable and replayable. Across dense, hybrid, and mixture-of-experts models on up to four-dimensional meshes of 128 NVIDIA H100 GPUs, it matched or exceeded expert-written TorchTitan recipes on most tested configurations, with throughput gains up to 18.3%, and found plans those uniform recipes did not express.

### Source excerpt

arXiv:2611.08660v1 Announce Type: new Abstract: Training large models requires multiple interacting forms of parallelism. As models and clusters scale, finding and maintaining efficient parallelization plans becomes essential for training performance. Diverse model architectures, workloads, and hardware topologies demand flexible strategies, yet production PyTorch systems typically rely on hand-written recipes that are costly to adapt and tend to reuse familiar patterns, which may be suboptimal for a given workload. We present AutoParallel, a PyTorch-native planner for distributed model training. AutoParallel searches parallelization choices jointly across the training computation, allowing established techniques to be combined and unconventional strategies to emerge. It expresses its decisions using native PyTorch abstractions and produces ordinary PyTorch programs, avoiding a separate execution runtime while composing directly with the PyTorch compilation and training stack. AutoParallel also composes automatic decisions with model-specific distributed code, allowing dynamic or specialized regions to remain under user control. Plans are inspectable and replayable artifacts. Across dense, hybrid, and mixture-of-experts models on meshes of up to four dimensions and 128 NVIDIA H100 GPUs, AutoParallel matches or exceeds expert-written TorchTitan recipes on most configurations, with throughput gains of up to 18.3%, and discovers plans that uniform expert recipes do not express.

## PyTorch 2.15 tackles the 'one bad GPU' problem

DevFeed: [PyTorch 2.15 tackles the 'one bad GPU' problem](<https://devfeed.tech/articles/pytorch-2-15-tackles-the-one-bad-gpu-problem-122495.md>)

Original publisher: [Read original article](<https://www.devclass.com/devops/2026/11/07/pytorch-215-tackles-the-one-bad-gpu-problem/5308234>)

Author: Joab Jackson

Published: 2026-11-07T18:38:00Z

Content type: news

Language: en

Sources: [DevClass.com](<https://devfeed.tech/sources/devclass-com.md>)

Topics: [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [NCCL](<https://devfeed.tech/topics/nccl.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [api](<https://devfeed.tech/tags/api.md>), [apple](<https://devfeed.tech/tags/apple.md>), [backend](<https://devfeed.tech/tags/backend.md>), [blackwell](<https://devfeed.tech/tags/blackwell.md>), [bottlenecks](<https://devfeed.tech/tags/bottlenecks.md>), [cluster](<https://devfeed.tech/tags/cluster.md>), [clusters](<https://devfeed.tech/tags/clusters.md>), [communication](<https://devfeed.tech/tags/communication.md>), [communications](<https://devfeed.tech/tags/communications.md>), [computing](<https://devfeed.tech/tags/computing.md>), [data-movement](<https://devfeed.tech/tags/data-movement.md>), [deadlocks](<https://devfeed.tech/tags/deadlocks.md>), [devops](<https://devfeed.tech/tags/devops.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [distributed-computing](<https://devfeed.tech/tags/distributed-computing.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [job](<https://devfeed.tech/tags/job.md>), [ml](<https://devfeed.tech/tags/ml.md>), [nccl](<https://devfeed.tech/tags/nccl.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

PyTorch 2.15 rewrites its NCCL backend to improve diagnosis and recovery in multi-GPU jobs. The former ProcessGroupNCCL could create many communicators per process group, contributing to intermittent deadlocks and unpredictable performance; the new backend limits each group to one communicator at initialization, so programs needing parallelism must dispatch more groups. It also supports easier reconfiguration when a GPU is slow or fails: a third-party orchestrator can exclude it and assign a replacement worker. A new flight recorder stores operation snapshots for inspection after failures. The legacy backend remains available as a rollback in 2.15 and is slated for removal in 2.16 if the replacement works as expected. Other reported additions include an asynchronous one-way Python transport API and parallel-processing improvements for Apple hardware and Nvidia Blackwell GPUs.

### Source excerpt

ML admins will be happy to learn that one bad GPU may no longer mean starting an entire training job ...

## NUMA-Torch: Compiler-Driven NUMA-Aware LLM Prefill on Multi-Socket CPUs

DevFeed: [NUMA-Torch: Compiler-Driven NUMA-Aware LLM Prefill on Multi-Socket CPUs](<https://devfeed.tech/articles/numa-torch-compiler-driven-numa-aware-llm-prefill-on-multi-socket-cpus-121261.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.07105>)

Author: Thisara Rathnayaka Weerakoon (Department of Computer Science & Engineering, University of Moratuwa, Sri Lanka), Ishan Dewpura (Department of Computer Science & Engineering, University of Moratuwa, Sri

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>)

Topics: [NUMA](<https://devfeed.tech/topics/numa.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>)

Tags: [affinity](<https://devfeed.tech/tags/affinity.md>), [attention](<https://devfeed.tech/tags/attention.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [banks](<https://devfeed.tech/tags/banks.md>), [bert](<https://devfeed.tech/tags/bert.md>), [compilation](<https://devfeed.tech/tags/compilation.md>), [compiler](<https://devfeed.tech/tags/compiler.md>), [compute](<https://devfeed.tech/tags/compute.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cpu-affinity](<https://devfeed.tech/tags/cpu-affinity.md>), [cpus](<https://devfeed.tech/tags/cpus.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-pf](<https://devfeed.tech/tags/cs-pf.md>), [cs-pl](<https://devfeed.tech/tags/cs-pl.md>), [data](<https://devfeed.tech/tags/data.md>), [data-locality](<https://devfeed.tech/tags/data-locality.md>), [dram](<https://devfeed.tech/tags/dram.md>), [floating-point](<https://devfeed.tech/tags/floating-point.md>), [llm](<https://devfeed.tech/tags/llm.md>), [numa](<https://devfeed.tech/tags/numa.md>), [prefill](<https://devfeed.tech/tags/prefill.md>), [socket](<https://devfeed.tech/tags/socket.md>), [torch](<https://devfeed.tech/tags/torch.md>)

### AI overview

NUMA-Torch targets multi-socket CPU LLM prefill bottlenecks caused by weights being accessed across NUMA sockets, which reduces effective DRAM bandwidth. Its NUMA-aware Tensor Parallelism treats each socket as a partition, shards weights across sockets, and uses CPU affinity to keep thread pools and memory allocations local. The abstract reports evaluations on GPT-2, BERT, and Gemma 3 270M, with speedups of up to 2.63x on dual-socket Haswell, 1.87x on dual-socket Cascade Lake, and 2.27x on four-socket Haswell-EX. These are reported maxima; the excerpt does not provide per-model results or broader evaluation details.

### Source excerpt

arXiv:2611.07105v1 Announce Type: new Abstract: For sufficiently large transformer-based Large Language Models (LLMs), whose parameters cannot be retained in on-chip caches, prefill throughput on multi-socket CPUs is often limited by sustainable DRAM bandwidth rather than peak floating-point throughput. In the standard PyTorch compilation (i.e., TorchInductor) path, model weights are allocated without considering the Non-Uniform Memory Access (NUMA) topology: specifically, the affinity between memory banks and compute cores is disregarded. This causes a substantial fraction of weight accesses to traverse the inter-socket interconnect, severely degrading the effective memory bandwidth available to each socket. We address this bottleneck through a compiler-level solution that makes TorchInductor NUMA-aware. Treating each NUMA socket as an independent tensor-parallel partition, we introduce NUMA-aware Tensor Parallelism (NUMA-TP). This approach transparently shards weights across sockets and schedules compute kernels such that thread pools and memory allocations remain strictly local to their respective sockets using CPU affinity. We evaluate the approach on GPT-2 (117M, 345M, 762M) and BERT (Small, Large), which exercise standard multi-head attention and MLP, as well as Gemma 3 270M, which combines Multi-Query Attention with a GeGLU-gated MLP. By strictly enforcing data locality, our approach successfully mitigates cross-socket traffic, achieving prefill speedups of up to 2.63x on a dual-socket Haswell platform, up to 1.87x on a dual-socket Cascade Lake platform, and up to 2.27x on a four-socket Haswell-EX platform.

## XStack: A Benchmark for ML Repo Migration

DevFeed: [XStack: A Benchmark for ML Repo Migration](<https://devfeed.tech/articles/xstack-a-benchmark-for-ml-repo-migration-120329.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.08472>)

Author: Linrong Cai, Taiming Lu, Haozhe Chen, Zhuang Liu

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [Benchmark](<https://devfeed.tech/topics/benchmark.md>)

Tags: [benchmark](<https://devfeed.tech/tags/benchmark.md>), [claude-opus-5](<https://devfeed.tech/tags/claude-opus-5.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [frameworks](<https://devfeed.tech/tags/frameworks.md>), [gpt-5-6-sol](<https://devfeed.tech/tags/gpt-5-6-sol.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [jax](<https://devfeed.tech/tags/jax.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [migration](<https://devfeed.tech/tags/migration.md>), [ml](<https://devfeed.tech/tags/ml.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [repo](<https://devfeed.tech/tags/repo.md>), [tpu](<https://devfeed.tech/tags/tpu.md>)

### AI overview

XStack benchmarks agents on whole-repository migration from PyTorch to JAX: each receives a PyTorch repository and a clean TPU machine, then must produce a JAX repository that meets grading criteria tied to the source project's published result, which the authors say they authenticated and calibrated. The benchmark targets migration across frameworks and hardware and is intended to make progress measurable. Reported pass rates show migration remains unreliable: Claude Opus 5 passed 75%, GPT-5.6-sol 45%, and all other evaluated agents at most 40%.

### Source excerpt

arXiv:2611.08472v1 Announce Type: cross Abstract: ML code should run where researchers choose, not where it was first written. Today, machine learning code mainly lives in two ecosystems, PyTorch on NVIDIA GPUs and JAX on Google TPUs. Migrating repositories between them means building and testing across frameworks and hardware. This migration is (1) a practical need, since it unlocks different hardware for researchers, and (2) a natural benchmark: it involves different frameworks and hardware, covers whole repositories, and is long-horizon and challenging. We introduce XStack, a benchmark that evaluates agents on migrating ML repositories between these two ecosystems. Each agent receives a PyTorch repository and a clean TPU machine and must deliver a JAX repository. Its output passes if it meets our grading criteria, based on the source repository's published result, authenticated and calibrated by us. Migration by agents remains unreliable: Claude Opus 5 passes 75%, GPT-5.6-sol 45%, and every other agent at most 40%. XStack is a step toward freeing ML code from its ecosystem. We release XStack to make progress toward reliable migration across frameworks and hardware measurable.

[Next page](<https://devfeed.tech/topics/pytorch.md?cursor=WyIyMDI2LTExLTA2VDA1OjAwOjAwKzAwOjAwIiwgIjA2YjE0NGY4LTQ2NmUtNDFlMy04OWVkLTJmNzEzNmJhODRlYyJd>)