Mastodon

HGPU groupPyGraph: Robust Compiler Support for CUDA Graphs in PyTorch<a href="https://mast.hpc.social/tags/CUDA" class="mention hashtag" rel="nofollow noopener" target="_blank">#CUDA</a> <a href="https://mast.hpc.social/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a><a href="https://hgpu.org/?p=29838" rel="nofollow noopener" translate="no" target="_blank">https://hgpu.org/?p=29838</a>

sc0v0neMCP, Agentic Knowledge Graphs & AI Models: Solving Conversational Analytics<a href="https://www.eventbrite.com/e/mcp-agentic-knowledge-graphs-ai-models-solving-conversational-analytics-tickets-1304648411519?aff=ebemoffollowpublishemail&ref=eemail&utm_campaign=following_published_event&utm_content=follow_notification&utm_medium=email&utm_source=eventbrite" rel="nofollow noopener" translate="no" target="_blank">https://www.eventbrite.com/e/mcp-agentic-knowledge-graphs-ai-models-solving-conversational-analytics-tickets-1304648411519?aff=ebemoffollowpublishemail&ref=eemail&utm_campaign=following_published_event&utm_content=follow_notification&utm_medium=email&utm_source=eventbrite</a>In this free webinar led by ex-Snowflake, Cloudera, and Amazon leaders, we'll unveil how cutting-edge LLMs (GPT 4.5, Sonnet 3.7, Deepseek V3/R1, Gemini 2.5, etc.) are revolutionizing data products. <a href="https://mastodon.social/tags/python" class="mention hashtag" rel="nofollow noopener" target="_blank">#python</a> <a href="https://mastodon.social/tags/machinelearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#machinelearning</a> <a href="https://mastodon.social/tags/deeplearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#deeplearning</a> <a href="https://mastodon.social/tags/ai" class="mention hashtag" rel="nofollow noopener" target="_blank">#ai</a> <a href="https://mastodon.social/tags/developer" class="mention hashtag" rel="nofollow noopener" target="_blank">#developer</a> <a href="https://mastodon.social/tags/dev" class="mention hashtag" rel="nofollow noopener" target="_blank">#dev</a> <a href="https://mastodon.social/tags/devsecops" class="mention hashtag" rel="nofollow noopener" target="_blank">#devsecops</a> <a href="https://mastodon.social/tags/devops" class="mention hashtag" rel="nofollow noopener" target="_blank">#devops</a> <a href="https://mastodon.social/tags/mlops" class="mention hashtag" rel="nofollow noopener" target="_blank">#mlops</a> <a href="https://mastodon.social/tags/learn" class="mention hashtag" rel="nofollow noopener" target="_blank">#learn</a> <a href="https://mastodon.social/tags/learning" class="mention hashtag" rel="nofollow noopener" target="_blank">#learning</a> <a href="https://mastodon.social/tags/study" class="mention hashtag" rel="nofollow noopener" target="_blank">#study</a> <a href="https://mastodon.social/tags/git" class="mention hashtag" rel="nofollow noopener" target="_blank">#git</a> <a href="https://mastodon.social/tags/github" class="mention hashtag" rel="nofollow noopener" target="_blank">#github</a> <a href="https://mastodon.social/tags/codeberg" class="mention hashtag" rel="nofollow noopener" target="_blank">#codeberg</a> <a href="https://mastodon.social/tags/tensorflow" class="mention hashtag" rel="nofollow noopener" target="_blank">#tensorflow</a> <a href="https://mastodon.social/tags/pytorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#pytorch</a> <a href="https://mastodon.social/tags/jax" class="mention hashtag" rel="nofollow noopener" target="_blank">#jax</a> <a href="https://mastodon.social/tags/huggingface" class="mention hashtag" rel="nofollow noopener" target="_blank">#huggingface</a> <a href="https://mastodon.social/tags/linux" class="mention hashtag" rel="nofollow noopener" target="_blank">#linux</a> <a href="https://mastodon.social/tags/ubuntu" class="mention hashtag" rel="nofollow noopener" target="_blank">#ubuntu</a> <a href="https://mastodon.social/tags/popos" class="mention hashtag" rel="nofollow noopener" target="_blank">#popos</a> <a href="https://mastodon.social/tags/llm" class="mention hashtag" rel="nofollow noopener" target="_blank">#llm</a>

fsoltTorch Lens Maker<a href="https://victorpoughon.github.io/torchlensmaker" rel="nofollow noopener" target="_blank">https://victorpoughon.github.io/torchlensmaker</a> – Librería en Python para el diseño de elementos ópticos utilizando modelos de optimización con PyTorch

LavX NewsUnlocking Deep Learning Performance: Understanding Compute, Memory Bandwidth, and OverheadIn the quest for optimizing deep learning models, developers often resort to trial-and-error methods that can lead to suboptimal performance. This article explores the critical factors affecting deep ...<a href="https://news.lavx.hu/article/unlocking-deep-learning-performance-understanding-compute-memory-bandwidth-and-overhead" rel="nofollow noopener" target="_blank">https://news.lavx.hu/article/unlocking-deep-learning-performance-understanding-compute-memory-bandwidth-and-overhead</a><a href="https://mastodon.cloud/tags/news" class="mention hashtag" rel="nofollow noopener" target="_blank">#news</a> <a href="https://mastodon.cloud/tags/tech" class="mention hashtag" rel="nofollow noopener" target="_blank">#tech</a> <a href="https://mastodon.cloud/tags/DeepLearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#DeepLearning</a> <a href="https://mastodon.cloud/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a> <a href="https://mastodon.cloud/tags/OperatorFusion" class="mention hashtag" rel="nofollow noopener" target="_blank">#OperatorFusion</a>

Kyle TaylorMaking the rounds. Worth a reshare on the fedi... This post is a long form essay version of a talk about PyTorch internals given at the PyTorch NYC meetup on May 14, 2019....[1] <a href="https://blog.ezyang.com/2019/05/pytorch-internals/" rel="nofollow noopener" translate="no" target="_blank">https://blog.ezyang.com/2019/05/pytorch-internals/</a><a href="https://hostux.social/tags/pytorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#pytorch</a> <a href="https://hostux.social/tags/machinelearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#machinelearning</a>

st1nger :unverified: 🏴‍☠️ :linux: :freebsd:<a href="https://infosec.exchange/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a> internals (2019) <a href="https://infosec.exchange/tags/python" class="mention hashtag" rel="nofollow noopener" target="_blank">#python</a> <a href="https://infosec.exchange/tags/tensor" class="mention hashtag" rel="nofollow noopener" target="_blank">#tensor</a> <a href="https://infosec.exchange/tags/cuda" class="mention hashtag" rel="nofollow noopener" target="_blank">#cuda</a> <a href="https://infosec.exchange/tags/ML" class="mention hashtag" rel="nofollow noopener" target="_blank">#ML</a> <a href="https://infosec.exchange/tags/NLP" class="mention hashtag" rel="nofollow noopener" target="_blank">#NLP</a> <a href="https://blog.ezyang.com/2019/05/pytorch-internals/" rel="nofollow noopener" translate="no" target="_blank">https://blog.ezyang.com/2019/05/pytorch-internals/</a>

N-gated Hacker NewsAh, nothing screams "exciting" like a deep dive into <a href="https://mastodon.social/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a> internals! 🎉 Let's unravel the mysteries of <a href="https://mastodon.social/tags/tensors" class="mention hashtag" rel="nofollow noopener" target="_blank">#tensors</a>, because who doesn't love a bedtime story about C codebases? 💤 Spoiler: it's as thrilling as watching paint dry, but with extra parentheses. 🤓 <a href="https://blog.ezyang.com/2019/05/pytorch-internals/" rel="nofollow noopener" translate="no" target="_blank">https://blog.ezyang.com/2019/05/pytorch-internals/</a> <a href="https://mastodon.social/tags/CCodebase" class="mention hashtag" rel="nofollow noopener" target="_blank">#CCodebase</a> <a href="https://mastodon.social/tags/DeepDive" class="mention hashtag" rel="nofollow noopener" target="_blank">#DeepDive</a> <a href="https://mastodon.social/tags/TechHumor" class="mention hashtag" rel="nofollow noopener" target="_blank">#TechHumor</a> <a href="https://mastodon.social/tags/ProgrammingInsights" class="mention hashtag" rel="nofollow noopener" target="_blank">#ProgrammingInsights</a> <a href="https://mastodon.social/tags/HackerNews" class="mention hashtag" rel="nofollow noopener" target="_blank">#HackerNews</a> <a href="https://mastodon.social/tags/ngated" class="mention hashtag" rel="nofollow noopener" target="_blank">#ngated</a>

Hacker NewsPyTorch Internals: Ezyang's Blog<a href="https://blog.ezyang.com/2019/05/pytorch-internals/" rel="nofollow noopener" translate="no" target="_blank">https://blog.ezyang.com/2019/05/pytorch-internals/</a><a href="https://mastodon.social/tags/HackerNews" class="mention hashtag" rel="nofollow noopener" target="_blank">#HackerNews</a> <a href="https://mastodon.social/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a> <a href="https://mastodon.social/tags/Internals" class="mention hashtag" rel="nofollow noopener" target="_blank">#Internals</a> <a href="https://mastodon.social/tags/Ezyang" class="mention hashtag" rel="nofollow noopener" target="_blank">#Ezyang</a> <a href="https://mastodon.social/tags/Blog" class="mention hashtag" rel="nofollow noopener" target="_blank">#Blog</a> <a href="https://mastodon.social/tags/MachineLearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#MachineLearning</a> <a href="https://mastodon.social/tags/DeepLearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#DeepLearning</a> <a href="https://mastodon.social/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#AI</a>

LavX NewsUnlocking AI's Potential with Scallop: The Neurosymbolic Programming RevolutionScallop, a groundbreaking declarative language, merges symbolic reasoning with AI applications, offering developers a powerful tool to enhance their machine learning models. By integrating seamlessly ...<a href="https://news.lavx.hu/article/unlocking-ai-s-potential-with-scallop-the-neurosymbolic-programming-revolution" rel="nofollow noopener" target="_blank">https://news.lavx.hu/article/unlocking-ai-s-potential-with-scallop-the-neurosymbolic-programming-revolution</a><a href="https://mastodon.cloud/tags/news" class="mention hashtag" rel="nofollow noopener" target="_blank">#news</a> <a href="https://mastodon.cloud/tags/tech" class="mention hashtag" rel="nofollow noopener" target="_blank">#tech</a> <a href="https://mastodon.cloud/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a> <a href="https://mastodon.cloud/tags/NeurosymbolicAI" class="mention hashtag" rel="nofollow noopener" target="_blank">#NeurosymbolicAI</a> <a href="https://mastodon.cloud/tags/Scallop" class="mention hashtag" rel="nofollow noopener" target="_blank">#Scallop</a>

nearshorecyberAI/ML Engineers - Check out this job in Mexico City!👉 Sr. ML Engineer (Prioritization Engine) - Hybrid with salary of CDMX $140,000 to 150,000 MEX pesos per month <a href="https://www.careers-page.com/nearshore-cyber/job/QWWY6846" rel="nofollow noopener" translate="no" target="_blank">https://www.careers-page.com/nearshore-cyber/job/QWWY6846</a> <a href="https://mastodon.social/tags/Sagemaker" class="mention hashtag" rel="nofollow noopener" target="_blank">#Sagemaker</a> <a href="https://mastodon.social/tags/ClearML" class="mention hashtag" rel="nofollow noopener" target="_blank">#ClearML</a> <a href="https://mastodon.social/tags/DeepLearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#DeepLearning</a> <a href="https://mastodon.social/tags/Python" class="mention hashtag" rel="nofollow noopener" target="_blank">#Python</a> <a href="https://mastodon.social/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a>

N-gated Hacker News🤔 Ah, the riveting world of "differentiable geometric optics in PyTorch"—because nothing screams excitement like a virtual optometrist in your <a href="https://mastodon.social/tags/GPU" class="mention hashtag" rel="nofollow noopener" target="_blank">#GPU</a>. 📈🔍 Here's a wild idea: instead of pondering the geometric optics through tiny lenses, maybe take a look through your real-world windows and donate to the author while you're at it. 🤑👓 <a href="https://victorpoughon.github.io/torchlensmaker/" rel="nofollow noopener" translate="no" target="_blank">https://victorpoughon.github.io/torchlensmaker/</a> <a href="https://mastodon.social/tags/differentiableOptics" class="mention hashtag" rel="nofollow noopener" target="_blank">#differentiableOptics</a> <a href="https://mastodon.social/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a> <a href="https://mastodon.social/tags/virtualOptometrist" class="mention hashtag" rel="nofollow noopener" target="_blank">#virtualOptometrist</a> <a href="https://mastodon.social/tags/geometricOptics" class="mention hashtag" rel="nofollow noopener" target="_blank">#geometricOptics</a> <a href="https://mastodon.social/tags/donations" class="mention hashtag" rel="nofollow noopener" target="_blank">#donations</a> <a href="https://mastodon.social/tags/HackerNews" class="mention hashtag" rel="nofollow noopener" target="_blank">#HackerNews</a> <a href="https://mastodon.social/tags/ngated" class="mention hashtag" rel="nofollow noopener" target="_blank">#ngated</a>

Hacker NewsTorch Lens Maker – Differentiable Geometric Optics in PyTorch<a href="https://victorpoughon.github.io/torchlensmaker/" rel="nofollow noopener" translate="no" target="_blank">https://victorpoughon.github.io/torchlensmaker/</a><a href="https://mastodon.social/tags/HackerNews" class="mention hashtag" rel="nofollow noopener" target="_blank">#HackerNews</a> <a href="https://mastodon.social/tags/TorchLensMaker" class="mention hashtag" rel="nofollow noopener" target="_blank">#TorchLensMaker</a> <a href="https://mastodon.social/tags/DifferentiableOptics" class="mention hashtag" rel="nofollow noopener" target="_blank">#DifferentiableOptics</a> <a href="https://mastodon.social/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a> <a href="https://mastodon.social/tags/ComputerVision" class="mention hashtag" rel="nofollow noopener" target="_blank">#ComputerVision</a> <a href="https://mastodon.social/tags/MachineLearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#MachineLearning</a>

nearshorecyberAI/ML Engineers - Check out this job in Mexico City!👉 Sr. ML Engineer (Prioritization Engine) - Hybrid with salary of CDMX $140,000 to 150,000 MEX pesos per month <a href="https://www.careers-page.com/nearshore-cyber/job/QWWY6846" rel="nofollow noopener" translate="no" target="_blank">https://www.careers-page.com/nearshore-cyber/job/QWWY6846</a> <a href="https://infosec.exchange/tags/Sagemaker" class="mention hashtag" rel="nofollow noopener" target="_blank">#Sagemaker</a> <a href="https://infosec.exchange/tags/ClearML" class="mention hashtag" rel="nofollow noopener" target="_blank">#ClearML</a> <a href="https://infosec.exchange/tags/DeepLearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#DeepLearning</a> <a href="https://infosec.exchange/tags/Python" class="mention hashtag" rel="nofollow noopener" target="_blank">#Python</a> <a href="https://infosec.exchange/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a>

Titus von der Malsburg 📖👀💭I'd like to buy <a href="https://scholar.social/tags/GPU" class="mention hashtag" rel="nofollow noopener" target="_blank">#GPU</a> cloud compute, with SSH access for <a href="https://scholar.social/tags/LLM" class="mention hashtag" rel="nofollow noopener" target="_blank">#LLM</a> inference (<a href="https://scholar.social/tags/pytorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#pytorch</a>). 32GB GPU memory is enough because I'm working with smaller models. Any recommendations?

HabrПишем свой TransformerЧтобы поупражняться я решила более детально разобраться и попробовать самостоятельно написать Transformer на PyTorch. Результатом захотелось поделиться здесь. Надеюсь, так же как и мне, это поможет доразобраться в данной архитектуре и ответить на какие-то вопросы.<a href="https://habr.com/ru/articles/891972/" rel="nofollow noopener" translate="no" target="_blank">https://habr.com/ru/articles/891972/</a><a href="https://zhub.link/tags/transformer" class="mention hashtag" rel="nofollow noopener" target="_blank">#transformer</a> <a href="https://zhub.link/tags/attention" class="mention hashtag" rel="nofollow noopener" target="_blank">#attention</a> <a href="https://zhub.link/tags/pytorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#pytorch</a>

HabrДесять уроков развития аппаратных ускорителей для ИИ: как эволюция TPU привела к созданию TPUv4iВ последние годы стало очевидно, что классические центральные процессоры (CPU) и видеокарты (GPU) уже не всегда поспевают за непрерывным ростом и усложнением нейронных сетей. Вместо бесконечного наращивания «универсального» железа, компании начали разрабатывать и внедрять в своих дата-центрах Domain-Specific Architecture (DSA) — аппаратные ускорители, заточенные под конкретные задачи. Google TPU (Tensor Processing Unit) — одно из первых крупных решений такого рода. Начиная с 2015 года (поколение TPUv1), Google успела вывести на рынок несколько поколений TPU для внутренних нужд: TPUv1 и TPUv2/v3, а в 2020 году — новое решение TPUv4i . Если первые версии TPU были ориентированы исключительно на ускорение инференса (выполнение уже обученных моделей), то TPUv2 и TPUv3 смогли взять на себя ещё и тренировку крупных нейросетей. Но в дальнейшем выяснилось, что для оптимальной работы дата-центров в масштабах Google рациональнее разделить решения для тренировки и инференса. TPUv4i — это результат учёта многих уроков и ограничений, проявившихся в предыдущих чипах. В этом материале разберём, какие «десять уроков» сформировали подход Google к созданию TPUv4i , что это за архитектура и какие проблемы дата-центров она решает.<a href="https://habr.com/ru/articles/892102/" rel="nofollow noopener" translate="no" target="_blank">https://habr.com/ru/articles/892102/</a><a href="https://zhub.link/tags/ml" class="mention hashtag" rel="nofollow noopener" target="_blank">#ml</a> <a href="https://zhub.link/tags/pytorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#pytorch</a> <a href="https://zhub.link/tags/proceesors" class="mention hashtag" rel="nofollow noopener" target="_blank">#proceesors</a> <a href="https://zhub.link/tags/deep_learning" class="mention hashtag" rel="nofollow noopener" target="_blank">#deep_learning</a> <a href="https://zhub.link/tags/inference" class="mention hashtag" rel="nofollow noopener" target="_blank">#inference</a>

LavX NewsTorchServe: The Future of PyTorch Model Deployment Faces Limited MaintenanceTorchServe, a pivotal tool for serving PyTorch models in production, has announced that it is no longer actively maintained. This development raises concerns about the future of model serving in AI ap...<a href="https://news.lavx.hu/article/torchserve-the-future-of-pytorch-model-deployment-faces-limited-maintenance" rel="nofollow noopener" target="_blank">https://news.lavx.hu/article/torchserve-the-future-of-pytorch-model-deployment-faces-limited-maintenance</a><a href="https://mastodon.cloud/tags/news" class="mention hashtag" rel="nofollow noopener" target="_blank">#news</a> <a href="https://mastodon.cloud/tags/tech" class="mention hashtag" rel="nofollow noopener" target="_blank">#tech</a> <a href="https://mastodon.cloud/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a> <a href="https://mastodon.cloud/tags/TorchServe" class="mention hashtag" rel="nofollow noopener" target="_blank">#TorchServe</a> <a href="https://mastodon.cloud/tags/ModelServing" class="mention hashtag" rel="nofollow noopener" target="_blank">#ModelServing</a>

Towards Data ScienceIs your metric collection slowing down your training? Chaim Rand explores how inefficient metric computation can impact performance and provides optimization techniques using TorchMetrics and <a href="https://hachyderm.io/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a> Profiler. <a href="https://towardsdatascience.com/efficient-metric-collection-in-pytorch-avoiding-the-performance-pitfalls-of-torchmetrics/" rel="nofollow noopener" translate="no" target="_blank">https://towardsdatascience.com/efficient-metric-collection-in-pytorch-avoiding-the-performance-pitfalls-of-torchmetrics/</a>

Dantali0n :arch: :i3:I wish I had a GPU where <a href="https://fosstodon.org/tags/PyTorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyTorch</a> would actually work on.But after an extensive battle with <a href="https://fosstodon.org/tags/ROCm" class="mention hashtag" rel="nofollow noopener" target="_blank">#ROCm</a> and HSA_OVERRIDE_GFX_VERSION. I have given up. Seems PyTorch on RDNA1 is out of the question.

Dmitry TantsurAha, roc* stuff is coming from <a href="https://rocm.docs.amd.com/en/latest" target="_blank" rel="nofollow noopener" translate="no">https://rocm.docs.amd.com/en/latest</a> which is something AMD-specific. Same for miopen.I'm pretty annoyed that they are hard dependencies even when I don't have the hardware. Like, it's fine for several megabytes, but we're talking about ~20 GiB of dead weight (that makes <a href="https://mastodon.online/tags/PyTorch" class="mention hashtag" rel="tag">#PyTorch</a> not installable on my humble root partition).It does not look like PyTorch actually requires these unconditionally. Another bug to file?<a href="https://mastodon.online/tags/Fedora" class="mention hashtag" rel="tag">#Fedora</a>

Recent searches

Search options

Administered by:

Server stats:

#pytorch