Running Ray at Scale on AKS The Azure Kubernetes Service (AKS) team at Microsoft has shared guidance for running Anyscale's managed Ray service at scale. They focus on three key issues: GPU capacity limits, scattered ML storage, and problems with credential expiry. By Claudio Masolo

Running Ray at Scale on AKS

The Azure Kubernetes Service (AKS) team at Microsoft has shared guidance for running Anyscale's managed Ray service at scale. They focus on three key issues: GPU capacity limits, scattered ML storage, and problems with cre…

Telegram AI Digest
#gpu #microsoft #rayproject

Running Ray at Scale on AKS

Запуск Ray в масштабе на AKS

Команда Azure Kubernetes Service (AKS) в Microsoft поделилась рекомендациями по масштабированию управляемого сервиса Ray от Anyscale. Они сосредоточились на трех ключевых проблемах: ограничениях GPU, разрозненном хранилище ML…

Telegram ИИ Дайджест
#ai #news #rayproject

Конвейеры масштабирования признаков с Feast и Ray

Использование хранилищ признаков, таких как Feast, и распределенных вычислительных фреймворков, таких как Ray, в производственных системах машинного обучения

Telegram ИИ Дайджест
#ai #featureengineering #rayproject

Scaling Feature Engineering Pipelines with Feast and Ray

Utilizing feature stores like Feast and distributed compute frameworks like Ray in production machine learning systems

Telegram AI Digest
#ai #featureengineering #rayproject

The Hard Truth About Machine Learning for Amazon FBA Sellers Amazon FBA demand forecasting breaks because the data is sparse, messy, and constantly shifting. Prophet and vanilla LSTMs often overfit and collapse under seasonality shifts. Real gains come from better feature engineering, TCNs with attention, Ray Tune + ASHA optimization, drift detection, and FBA-specific metrics like stockout penalties. In 2026, hybrid ML + RAG systems are becoming the only durable approach.

The Hard Truth About Machine Learning for Amazon FBA Sellers

Amazon FBA demand forecasting breaks because the data is sparse, messy, and constantly shifting. Prophet and vanilla LSTMs often overfit and collapse under seasonality shifts. Real g…

Telegram AI Digest
#lstm #machinelearning #rayproject

Article: Spec Driven Development: When Architecture Becomes Executable Spec-Driven Development inverts traditional architecture by making specifications executable and authoritative. It transforms declared intent into validated code through AI generation and provides architectural determinism. It eliminates drift through continuous enforcement, but demands new engineering discipline in schema design and contract-first reasoning. By Leigh Griffin, Ray Carroll

Article: Spec Driven Development: When Architecture Becomes Executable

Spec-Driven Development inverts traditional architecture by making specifications executable and authoritative. It transforms declared intent into validated code through AI generation a…

Telegram AI Digest
#ai #news #rayproject

How Discord Scaled Its ML Platform from Single-GPU Workflows to a Shared Ray Cluster

Как Discord Масштабировал Свою Платформу Машинного Обучения с Однопроцессорных Потоков Работы до Общего Кластера Ray

Telegram ИИ Дайджест
#gpu #ml #rayproject

How Discord Scaled Its ML Platform from Single-GPU Workflows to a Shared Ray Cluster Discord has detailed how it rebuilt its machine learning platform after hitting the limits of single-GPU training. The changes enabled daily retrains for large models and contributed to a 200% uplift in a key ads ranking metric. By Matt Foster

How Discord Scaled Its ML Platform from Single-GPU Workflows to a Shared Ray Cluster

Discord has detailed how it rebuilt its machine learning platform after hitting the limits of single-GPU training. The changes enabled daily retrains for large models and c…

Telegram AI Digest
#gpu #ml #rayproject

KubeCon NA 2025 - Robert Nishihara on Open Source AI Compute with Kubernetes, Ray, PyTorch, and vLLM AI workloads are growing more complex in terms of compute and data, and technologies like Kubernetes and PyTorch can help build production-ready AI systems to support them. Robert Nishihara from Anyscale recently spoke at KubeCon + CloudNativeCon North America 2025 Conference about how an AI compute stack comprising Kubernetes, PyTorch, VLLM and Ray technologies can support these new AI workloads. By Srini Penchikala

KubeCon NA 2025 - Robert Nishihara on Open Source AI Compute with Kubernetes, Ray, PyTorch, and vLLM

AI workloads are growing more complex in terms of compute and data, and technologies like Kubernetes and PyTorch can help build production-ready AI sys…

Telegram AI Digest
#llm #pytorch #rayproject

KubeCon NA 2025 - Robert Nishihara on Open Source AI Compute with Kubernetes, Ray, PyTorch, and vLLM

KubeCon NA 2025 — Роберт Нисихара о вычислениях с открытым исходным кодом для ИИ с использованием Kubernetes, Ray, PyTorch и vLLM

Рабочие нагрузки ИИ становятся все более сложными с точки зрения вычислений и данных, и такие технологии, как Kubernetes…

Telegram ИИ Дайджест
#llm #pytorch #rayproject

Two-Year-Old Ray AI Framework Flaw Exploited in Ongoing Campaign

Threat actors are abusing Ray’s lack of authentication to compromise exposed clusters and deploy LLM-generated payloads and cryptocurrency miners.

Telegram AI Digest
#ai #llm #rayproject

Уязвимость двухлетнего фреймворка Ray AI эксплуатируется в продолжающейся кампании

Злоумышленники злоупотребляют отсутствием аутентификации в Ray для компрометации уязвимых кластеров и развертывания полезной нагрузки, сгенерированной LLM, а также майнеро…

Telegram ИИ Дайджест
#ai #news #rayproject

Running high-scale reinforcement learning (RL) for LLMs on GKE Reinforcement Learning (RL) is vital for aligning large language models with human preferences, but scaling it presents infrastructure challenges. Google Cloud addresses these challenges with a full-stack approach, from custom hardware like TPUs and GPUs to GKE orchestration. RL for LLMs uses an iterative feedback loop involving generation, reward model evaluation, and parameter updates. This cycle aims to maximize reward through interaction, demanding system-wide efficiency. Google Cloud's solution integrates high-performance compute, including TPUs and GPUs, with holistic optimization. The approach emphasizes open-source tools, Kubernetes, and Ray, enabling flexible integration. GKE provides mega-scale orchestration, supporting massive, distributed jobs for RL workloads. Existing GKE infrastructure efficiently supports demanding RL workloads, including autoscaling and job scheduling. The building of an RL workflow involves defining objectives, choosing algorithms, and configuring hardware. Google Cloud provides resources like recipes and open-source contributions to get started with RL. Customers are encouraged to leverage the open-source ecosystem, particularly llm-d, for LLM serving.

Running high-scale reinforcement learning (RL) for LLMs on GKE

Reinforcement Learning (RL) is vital for aligning large language models with human preferences, but scaling it presents infrastructure challenges. Google Cloud addresses these challenges with a…

Telegram AI Digest
#llm #rayproject #tpu

A more native experience for Cloud TPUs with Ray on GKE

Более привычный опыт работы с Cloud TPU с помощью Ray на GKE

Ray, фреймворк для масштабирования рабочих нагрузок ИИ, представил значительные улучшения для использования Google Cloud TPU с GKE. Новая библиотека Ray TPU, ray.util.tpu, упрощает распознавание…

Telegram ИИ Дайджест
#ai #rayproject #tpu

A more native experience for Cloud TPUs with Ray on GKE Ray, a framework for scaling AI workloads, has introduced significant improvements for using Google Cloud TPUs with GKE. A new Ray TPU Library, ray.util.tpu, simplifies TPU awareness and scaling by automatically reserving entire TPU slices as atomic units. This prevents resource fragmentation, a major performance bottleneck, and is crucial for multi-slice training. The update expands support for JAX and PyTorch on TPUs through Ray Train's JaxTrainer, which handles distributed host initialization and hardware configuration. Ray Serve also now supports TPUs, allowing seamless transitions from GPU stacks with minimal code changes. The new Label-Based Scheduling API integrates with GKE custom compute classes, enabling Python-based hardware requests and fallback strategies. This API also translates GKE-provided TPU metadata into Ray-specific labels for precise hardware targeting. Additionally, key TPU performance metrics and libtpu logs are now available in the Ray Dashboard, streamlining debugging. These enhancements aim to make TPUs a more native and low-friction part of the Ray ecosystem, easing the adaptation of existing Ray applications. Developers can now access TPU metrics, use TPUs with Kuberay, and explore JAX workloads with the new JaxTrainer.

A more native experience for Cloud TPUs with Ray on GKE

Ray, a framework for scaling AI workloads, has introduced significant improvements for using Google Cloud TPUs with GKE. A new Ray TPU Library, ray.util.tpu, simplifies TPU awareness and scaling by au…

Telegram AI Digest
#jax #rayproject #tpu

Evolving Ray and Kubernetes together for the future of distributed AI and ML

Развитие Ray и Kubernetes вместе для будущего распределенного ИИ и машинного обучения

Google Cloud и Anyscale улучшают интеграцию Ray, движка вычислений с открытым исходным кодом для ИИ, с Kubernetes, в частности с Google Kubernetes Engine. В Ray были введ…

Telegram ИИ Дайджест
#ai #ml #rayproject

Evolving Ray and Kubernetes together for the future of distributed AI and ML Google Cloud and Anyscale are enhancing the integration of Ray, an OSS compute engine for AI, with Kubernetes, particularly Google Kubernetes Engine. Label selectors have been introduced to Ray, mirroring Kubernetes functionality, to improve scheduling flexibility for distributed tasks and actors. This allows developers to assign labels to nodes and specify resource requirements like accelerator types for task execution. Combining Ray and Kubernetes label selectors on GKE offers granular control over application deployment and infrastructure. Advanced accelerator support is also improved, enabling the use of next-generation AI accelerators like NVIDIA GB200 NVL72 with Ray on GKE via Dynamic Resource Allocation. Furthermore, Ray is getting a more native support for TPUs, including a JAXTrainer API for streamlined TPU training. Writable cgroups are now available on GKE for Ray clusters, enhancing reliability by allowing Ray to dynamically reserve resources for critical system tasks within containers. This feature improves the dependability of Ray clusters even under intense workloads without compromising security. The introduction of in-place pod resizing in Kubernetes v1.33 marks the beginning of vertical autoscaling capabilities for Ray on Kubernetes. This feature can boost workload efficiency by allowing Ray workers to scale their resources faster and more flexibly. Ray and Kubernetes are working together to create a powerful distributed operating system for AI/ML workloads.

Evolving Ray and Kubernetes together for the future of distributed AI and ML

Google Cloud and Anyscale are enhancing the integration of Ray, an OSS compute engine for AI, with Kubernetes, particularly Google Kubernetes Engine. Label selectors have been intr…

Telegram AI Digest
#ml #rayproject #tpu

PyTorch Foundation Welcomes Ray, Unveils Monarch for Simplified Distributed AI At the 2025 PyTorch Conference, the PyTorch Foundation unveiled new initiatives to advance open AI infrastructure, introducing frameworks like Ray and PyTorch Monarch for streamlined distributed workloads. Highlighting transparency in AI development, new projects from Stanford and AI2 aim to enhance reproducibility. The foundation is solidifying its role as a central hub for scalable AI solutions. By Hien Luu

PyTorch Foundation Welcomes Ray, Unveils Monarch for Simplified Distributed AI

At the 2025 PyTorch Conference, the PyTorch Foundation unveiled new initiatives to advance open AI infrastructure, introducing frameworks like Ray and PyTorch Monar…

Telegram AI Digest
#aiconference #pytorch #rayproject

PyTorch Foundation Welcomes Ray, Unveils Monarch for Simplified Distributed AI

Фонд PyTorch приветствует Ray, представляет Monarch для упрощения распределенного ИИ

На конференции PyTorch 2025 Фонд PyTorch представил новые инициативы по продвижению открытой инфраструктуры ИИ, представив такие фреймворки, как Ray и PyTorch Monarch…

Telegram ИИ Дайджест
#ai #pytorch #rayproject

Bay Area University Issues Warning Over Man Using Meta AI Glasses On Campus

The University of San Francisco issued a campuswide alert regarding a man using Meta Ray-Ban AI glasses. He was reportedly filming students and making inappropriate comments and dating questions.…

#meta #metaai #rayproject

Apple Shelves Vision Headset Revamp to Prioritize Meta-Like AI Glasses

Apple is reallocating resources from a planned cheaper Vision Pro headset to accelerate development of AI-powered smart glasses. The company is now prioritizing smart glasses to directly compete with M…

#apple #meta #rayproject

Я попробовал новый видеоредактор Ray3 AI от Luma Labs и обнаружил, что…

За последний год количество поисковых запросов по фразе «создатель видео с ИИ» выросло более чем на 200%. Это неудивительно, ведь создатели контента повсюду ищут более быстрые и умные способы превращать …

#ai #news #rayproject

21Shares Hits 50 Crypto ETPs in Europe With Launch of AI and Raydium-Focused Products

21Shares запускает 50 крипто-ETP в Европе с выпуском продуктов, ориентированных на ИИ и Raydium

AFET отслеживает группу децентрализованных AI-протоколов, в то время как ARAY предлагает доступ к токену децентрализованной биржи Raydium на базе Solana.

#ai #news #rayproject

I Tried The New Ray3 AI Video Maker by Luma Labs and I Found it…

In the past year, searches for “AI video maker” have skyrocketed by over 200%. This isn’t a shocker, as creators everywhere are looking for faster, smarter ways to turn ideas into visuals. In fact, the very ide…

#ai #news #rayproject

21Shares Hits 50 Crypto ETPs in Europe With Launch of AI and Raydium-Focused Products AFET tracks a group of decentralized AI protocols, while ARAY offers exposure to Solana-based decentralized exchange Raydium's token.

21Shares Hits 50 Crypto ETPs in Europe With Launch of AI and Raydium-Focused Products

AFET tracks a group of decentralized AI protocols, while ARAY offers exposure to Solana-based decentralized exchange Raydium's token.

#ai #news #rayproject

Harvard dropouts to launch ‘always on’ AI smart glasses that listen and record every conversation

After developing a facial recognition app for Meta’s Ray-Ban glasses and doxing random people, two former Harvard students are now launching a startup that makes smart glasses w…

#ai #meta #rayproject

Uber’s Journey to Ray on Kubernetes

Путешествие Uber к Ray на Kubernetes

Uber подробно рассказала о недавнем переходе к запуску рабочих нагрузок машинного обучения на основе Ray в Kubernetes. Это знаменует собой эволюцию их инфраструктуры, направленную на повышение масштабируемости, эффективности и удобства ра…

#ai #news #rayproject

Uber’s Journey to Ray on Kubernetes Uber has detailed a recent transition to running Ray-based machine learning workloads on Kubernetes. This marks an evolution in its infrastructure, with the aim of enhancing scalability, efficiency, and developer experience. The company recently published a two-part series from Uber Engineering delving into the motivations, challenges, and solutions encountered during this migration.​ By Craig Risi

Uber’s Journey to Ray on Kubernetes

Uber has detailed a recent transition to running Ray-based machine learning workloads on Kubernetes. This marks an evolution in its infrastructure, with the aim of enhancing scalability, efficiency, and developer experience. The company re…

#ai #news #rayproject

Вот что я узнал после шести месяцев ношения умных очков Ray-Ban Meta

Как выглядят флагманские умные очки Meta спустя полгода?

#ai #meta #rayproject

Here's What I Learned After Six Months of Wearing Ray-Ban Meta Smart Glasses

How do Meta's flagship smart glasses stack up after half a year?

#ai #meta #rayproject