Join us

ContentUpdates and recent posts about Pelagia..
Link
@faun shared a link, 1 month, 4 weeks ago

OpenAI to launch its first AI chip in 2026 with Broadcom, FT reports

OpenAI’s firstin-house AI chipis nearly out of the oven. It’s headed for fabrication atTSMCand built to handle OpenAI’s own workloads—no outside sales, according to theFinancial Times. Why it matters:Big AI shops are going vertical. Custom silicon means tighter control over runtime, reliability, an..

OpenAI to launch its first AI chip in 2026 with Broadcom, FT reports
Link
@faun shared a link, 1 month, 4 weeks ago

From Zero to GPU: A Guide to Building and Scaling Production-Ready CUDA Kernels

Hugging Face just dropped Kernel Builder—a full-stack toolchain for building, versioning, and shippingcustom CUDA kernels as native PyTorch ops. Kernels arearchitecture-aware,semantically versioned, andpullable straight from the Hub. It tracks changes with lockfiles and bakes inDocker deploysout of..

Link
@faun shared a link, 1 month, 4 weeks ago

GPT-5 Thinking in ChatGPT (aka Research Goblin) is shockingly good at search

GPT-5's“thinking” modeljust leveled up. It's not just answering queries—it’s doing full-on research. Picture deep, multi-step Bing searches mixed with tool use and reasoning chains. It reads PDFs. Analyzes them. Suggests what to do next. Then actually does it. All from your phone. What’s changing:L..

GPT-5 Thinking in ChatGPT (aka Research Goblin) is shockingly good at search
Link
@faun shared a link, 1 month, 4 weeks ago

The Big LLM Architecture Comparison

Architectures since GPT-2 still ride transformers. They crank memory and performance withRoPE, swapGQAforMLA, sprinkle in sparseMoE, and roll sliding-window attention. Teams shiftRMSNorm. They tweak layer norms withQK-Norm, locking in training stability across modern models. Trend to watch:In 2025,..

The Big LLM Architecture Comparison
Link
@faun shared a link, 1 month, 4 weeks ago

Hermes V3: Building Swiggy’s Conversational AI Analyst

Swiggy just gave its GenAI tool, Hermes, a serious glow-up. What started as a simple text-to-SQL bot is now acontext-aware AI analystthat lives inside Slack. The upgrade? Not just tweaks—an overhaul. Think: vector-based prompt retrieval, session-level memory, an Agent orchestration layer, and a SQL..

Hermes V3: Building Swiggy’s Conversational AI Analyst
Link
@faun shared a link, 1 month, 4 weeks ago

Simplifying Large-Scale LLM Processing across Instacart with Maple

Instacart builtMaple, a backend brain for handling millions of LLM prompts—fast, cheap, and shared across teams. It’s not just another service. Maple runs onTemporal,PyArrow, andS3, strip-mines away provider-specific boilerplate, auto-batches prompts, retries failures, and slashes LLM costs by up t..

Simplifying Large-Scale LLM Processing across Instacart with Maple
Link
@faun shared a link, 1 month, 4 weeks ago

Why language models hallucinate

OpenAI sheds light on the persistence ofhallucinationsin language models due to evaluation methods favoring guessing over honesty, requiring a shift towards rewarding uncertainty acknowledgment. High model accuracy does not equate to the eradication of hallucinations, as some questions are inherentl..

Why language models hallucinate
Link
@faun shared a link, 1 month, 4 weeks ago

Best Practices for High Availability of LLM Based on AI Gateway

Alibaba Cloud’s AI Gateway just got sharper. It now handlesreal-time overload protectionandLLM fallback routingusing passive health checks, first packet timeouts, and traffic shaping. It proxies both BYO and cloud LLMs—think PAI-EAS, Tongyi Qianwen—and redirects load spikes or failures on the fly. F..

Best Practices for High Availability of LLM Based on AI Gateway
Story
@laura_garcia shared a post, 1 month, 4 weeks ago
Software Developer, RELIANOID

RELIANOID Load Balancer Community Edition v7 on AWS using Terraform

🚀 New Guide Available! Learn how to quickly deploy RELIANOID Load Balancer Community Edition v7 on AWS using Terraform. Our step-by-step article shows you how to provision everything automatically — from VPCs and subnets to EC2 and key pairs — in just minutes. 👉 https://www.relianoid.com/resources/k..

Knowledge base Deploy RELIANOID Load Balancer Community Edition v7 with Terraform on AWS
Link
@faun shared a link, 1 month, 4 weeks ago

Deploy a containerized application with Kamal and Terraform

A Docker-first workflow combinesTerraformandKamalinto a lean, Elastic Beanstalk-ish alternative—without the bloat. Terraform spins up a three-tier VPC and wires it toECR. Kamal takes it from there, booting containers on a raw EC2 box: app, proxy, monitor. One script. Done...

Deploy a containerized application with Kamal and Terraform
Pelagia is a Kubernetes controller that provides all-in-one management for Ceph clusters installed by Rook. It delivers two main features:

Aggregates all Rook Custom Resources (CRs) into a single CephDeployment resource, simplifying the management of Ceph clusters.
Provides automated lifecycle management (LCM) of Rook Ceph OSD nodes for bare-metal clusters. Automated LCM is managed by the special CephOsdRemoveTask resource.

It is designed to simplify the management of Ceph clusters in Kubernetes installed by Rook.

Being solid Rook users, we had dozens of Rook CRs to manage. Thus, one day we decided to create a single resource that would aggregate all Rook CRs and deliver a smoother LCM experience. This is how Pelagia was born.

It supports almost all Rook CRs API, including CephCluster, CephBlockPool, CephFilesystem, CephObjectStore, and others, aggregating them into a single specification. We continuously work on improving Pelagia's API, adding new features, and enhancing existing ones.

Pelagia collects Ceph cluster state and all Rook CRs statuses into single CephDeploymentHealth CR. This resource highlights of Ceph cluster and Rook APIs issues, if any.

Another important thing we implemented in Pelagia is the automated lifecycle management of Rook Ceph OSD nodes for bare-metal clusters. This feature is delivered by the CephOsdRemoveTask resource, which automates the process of removing OSD disks and nodes from the cluster. We are using this feature in our everyday day-2 operations routine.