slurm runs a copy of your script. edit it after sbatch and nothing changes.

backfill kicked in ~30s after the priority pass. it plans with declared limits, not actual runtimes. a bad time limit is a bad plan.

a 1-cpu job took a whole 4-cpu node. AllocTRES: cpu=1,mem=7851M CR_CORE_MEMORY + DefMemPerNode=UNLIMITED meant every job asked for all the memory. my array ran 2 at a time. with --mem=500M, 8 at once.

starting something new, in public. for the past year i’ve been building infra that schedules compute: distributed systems, resource allocation, keeping long-running jobs fed. quantum computers are now getting plugged into the same kind of infra. supercomputing centers are adding QPUs to slurm, the scheduler most of them run. the catch: the scheduler treats a QPU like a static slot. it doesn’t know the QPU’s calibration drifts over hours, or that it’s about to go offline to recalibrate. so for the next 6 weeks i’m building a scheduler that does know. simulated QPU, real slurm cluster, real numbers. what you’ll see here: - weekly build logs: what i built, what broke - numbers, not vibes - a full write-up at week 6 week 0 findings tomorrow. repo + devlog below. github.com/ArshMalik02/qpu-a…

first @a16z hackathon was a blast until next time!! #speedhacks26

Closing out SF Tech Week at Speed Hacks 🏁 Let's ship. @speedrun @a16z @hackersquad
#SpeedHacks #SFTechWeek #SFHackWeek #speedrun #buildinpublic

i hope claude opus 5.5 can do better..

convince me that vibe coding works i tried to build a form for an event which collects data in google sheets ui: claude design code: claude code expectation: it will only take me 30 min to get this to work result: broken form that doesn’t work

Flying back from Yellowstone to SF and realizing that no one cared about agentic enterprise AI here. Life is much more