Join us

ContentUpdates and recent posts about Slurm..
Link
@faun shared a link, 6 months ago
FAUN.dev()

Perplexity offers training wheels for building AI agents

Perplexity Labsis your quick-draw tool for crafting apps and digital delights, powered by LLMs likeGPT-4 Omni. It’s a star where others stumble: fast, project-driven tasks. Expect example-heavy insights and real-world project demos. While competitors dawdle, it delivers. Need deep web browsing, code.. read more  

Link
@faun shared a link, 6 months ago
FAUN.dev()

Why GCP Load Balancers Struggle with Stateful LLM Traffic — and How to Fix It

Deploying LLMs onGCP Load Balancersis like fitting a square peg in a round hole. These models aren't stateless, so skip HTTP, go straight forTCP Load Balancing. Toss in Redis to keep those sessions on a leash. Tweak load balancer settings to dodge mid-stream socket calamities. Embrace the power ofGK.. read more  

Link
@faun shared a link, 6 months ago
FAUN.dev()

Want a humanoid, open source robot for just $3,000? Hugging Face is on it.

Hugging Facejust pulled the curtain back onHopeJR, a humanoid robot that swings 66 degrees of freedom—at just$3,000. This price tag shames the $16,000 slapped on Unitree's G1. Together with The Robot Studio, they've created this robot with a dash of Bender's charisma. The kicker? It's fully open-sou.. read more  

Want a humanoid, open source robot for just $3,000? Hugging Face is on it.
Link
@faun shared a link, 6 months ago
FAUN.dev()

A visual introduction to vector embeddings

OpenAI's text-embedding-ada-002often gets a peculiar itch at dimension 196—vectors peaking awkwardly there. Entertext-embedding-3-small, swooping in to smooth out the distribution. Now, ontosimilarity metrics. For unit vectors, the dot product is your fast friend. It's interchangeable with cosine si.. read more  

A visual introduction to vector embeddings
Link
@faun shared a link, 6 months ago
FAUN.dev()

Using AI to outsmart AI-driven phishing scams

Phishing scamsare growing craftier, employing AI likeFraudGPTto weave through filters and masquerade as real emails, boosting scam rates by70%. AI can unveil sneaky phishing patterns humans miss, but it loves a good panic. It often cries wolf with false alarms and needs a babysitter to adjust to eve.. read more  

Using AI to outsmart AI-driven phishing scams
Link
@faun shared a link, 6 months ago
FAUN.dev()

AI agents have access to key data across the enterprise

82% of organizations have AI agents on deck; a mere 44% bother with security policies.That leaves a lot of open doors. A staggering 96% of tech pros are side-eyeing these agents as ticking time bombs, yet 98% plan to unleash more. It's like setting out catnip for hackers. These agents wield power wi.. read more  

AI agents have access to key data across the enterprise
Link
@faun shared a link, 6 months ago
FAUN.dev()

AI didn’t kill Stack Overflow

Stack Overflow once buzzed with collective brainpower. But then, it got too wrapped up in reputation points, a full-on leaderboard obsession. This detour dimmed its shine. It turns out, platforms flourish on real teamwork, not just gamified dick measuring contests. As AI sweeps through the coding wo.. read more  

AI didn’t kill Stack Overflow
Link
@faun shared a link, 6 months ago
FAUN.dev()

From Zero to Hero: Build your first voice agent with Voice Live API

TheVoice Live APIditches the clutter of juggling models. One API call, and voilà—real-time,natural-sounding bots. It’s harnessed over WebSocket, keeping everything sharp and efficient... read more  

From Zero to Hero: Build your first voice agent with Voice Live API
Link
@faun shared a link, 6 months ago
FAUN.dev()

We rewrote large parts of our API in Go using AI: we are now ready to handle one billion databases

Tursooverhauled its API withGoand AI, gunning for 1 billion databases. Think big, act smart. They squeezed every byte by adopting string interning. No more in-memory maps—they swapped them for aSQLite-backedLRU cache. The result? Leaner memory usage and hassle-free proxy bootstrapping... read more  

We rewrote large parts of our API in Go using AI: we are now ready to handle one billion databases
Link
@faun shared a link, 6 months ago
FAUN.dev()

Linear Programming for Fun and Profit

Modal’s "resource solver" hacks cloud volatility. It taps into thesimplex algorithmto snag cheap GPUs. Scale-ups? Lightning-fast. Savings? In the millions... read more  

Linear Programming for Fun and Profit
Slurm Workload Manager is an open-source, fault-tolerant, and highly scalable cluster management and scheduling system widely used in high-performance computing (HPC). Designed to operate without kernel modifications, Slurm coordinates thousands of compute nodes by allocating resources, launching and monitoring jobs, and managing contention through its flexible scheduling queue.

At its core, Slurm uses a centralized controller (slurmctld) to track cluster state and assign work, while lightweight daemons (slurmd) on each node execute tasks and communicate hierarchically for fault tolerance. Optional components like slurmdbd and slurmrestd extend Slurm with accounting and REST APIs. A rich set of commands—such as srun, squeue, scancel, and sinfo—gives users and administrators full visibility and control.

Slurm’s modular plugin architecture supports nearly every aspect of cluster operation, including authentication, MPI integration, container runtimes, resource limits, energy accounting, topology-aware scheduling, preemption, and GPU management via Generic Resources (GRES). Nodes are organized into partitions, enabling sophisticated policies for job size, priority, fairness, oversubscription, reservation, and resource exclusivity.

Widely adopted across academia, research labs, and enterprise HPC environments, Slurm serves as the backbone for many of the world’s top supercomputers, offering a battle-tested, flexible, and highly configurable framework for large-scale distributed computing.