Slurm ya AI Amahugurwa
Slurm nisoko rifunguye-ryakazi rishinzwe gucunga gahunda no gukora imirimo kumasoko yo kubara cyane, kandi byahindutse amahitamo asanzwe mumahugurwa manini ya AI.
Incamake
It matters because it reliably distributes massive training runs across thousands of GPUs.
Kwibira cyane
Slurm (Byoroheje Linux Utility for Resource Management) yatangiriye muri supercomputing none iha imbaraga amatsinda menshi yo guhugura AI. Abakoresha batanga ibyiciro byanditse hamwe na sbatch, basaba ibikoresho nka node na GPUs hamwe nubuyobozi nka --gres = gpu: 8, na Slurm umurongo, bashyira imbere, bagatangira akazi. Srun itangiza ibyara inzira ihuza inzira, ikomatanya muburyo busanzwe hamwe na PyTorch DDP na NCCL. Slurm ikurikirana ibaruramari ryumutungo, ikurikiza neza-kugabana no kugabanya imipaka, kandi ikemura gahunda yo gusubira inyuma kugirango uhindure imirimo mito mu cyuho. Ku myitozo yicyitegererezo cyumupaka, amakipe yishingikiriza kuri Slurm gucunga ibihumbi bya GPU, gutangira kuri bariyeri nyuma yo kunanirwa, no kubika ubushobozi bwabigenewe kumara ibyumweru byinshi.
Ubushishozi
Umugenzuzi wa Slurm daemon (slurmctld) afata ibyemezo byigihe mugihe umukozi wa slurmd kuri buri node atangiza imirimo na raporo imiterere. Amacomeka rusange (GRES) akurikirana GPUs kuburyo imirimo ibasaba neza. srun ishyiraho ibidukikije bihinduka (urwego, ingano yisi, master adresse) yagabanije amasomero yamahugurwa yasomwe kuri bootstrap itumanaho rya NCCL. Gahunda yo gusubira inyuma ireka imirimo migufi ikora hakiri kare mugihe idatinze cyane-kubika ibyingenzi, kugumya gukoresha hejuru.
Ingaruka z'Ingamba
Igiciro na bije
Ibyemezo byubwubatsi bitwara imikorere nigiciro cyimikorere kumyaka.
Ibyemezo bisobanutse
Ubuhanga bwa tekinike bufasha amakipe guhitamo umurongo ukwiye, ntabwo ari shyashya gusa.
Kugenzura ubuziranenge
Guhitamo neza bya injeniyeri bigabanya ibintu byizewe mubikorwa.
Ejo hazaza ha Slurm kumasomo ya AI
Slurm ikomeje kongeramo ibicu biturika, inkunga ya kontineri ikoresheje Pyxis na Enroot, hamwe nibiranga GPU-izi neza. Nkuko ihuriro rya AI rigera ku 100.000-wongeyeho GPUs, tegereza kwihanganira amakosa akomeye, kugenzura byikora-gutangiza, hamwe nakazi ka elastike gahinduka nyuma yo gutsindwa. Amashyirahamwe menshi ubu akoresha Slurm kuruhande cyangwa munsi ya Kubernetes, kandi gahunda ya Hybrid igamije guhuza imikorere ya HPC hamwe nigicu-kavukire cyoroshye kugirango imyitozo ihore nini.
Gushyira mu bikorwa Isi
Laboratoire yumupaka itangiza ibyumweru byinshi byamahugurwa ikora ibihumbi n'ibihumbi GPU hamwe ninyandiko imwe ya sbatch isaba amajana.
Umushakashatsi atanga 'srun --gres = gpu: 8' gufata GPU umunani kumurongo umwe kugirango igeragezwa rya PyTorch DDP.
Guteganya gusubira inyuma umwanya muto wo gusuzuma akazi muri GPU idafite akazi mugihe amahugurwa manini yabitswe ategereje gutangira.
Nyuma yuko node yananiwe gukora hagati, Slurm isaba akazi kandi igakomeza kuva kuri bariyeri iheruka aho gutangira.
Ingaruka & Kurinda
Gutezimbere igipimo kimwe gishobora guhisha intege nke za sisitemu.
Ibikorwa Remezo no kubungabunga akenshi usanga bidahabwa agaciro.
Icyuho cyumutekano no kwitegereza birashobora kwiyongera uko sisitemu igenda igorana.
Igishushanyo mbonera
Sobanura ubukererwe, ubuziranenge, nigiciro cyibiciro mbere yo kubishyira mubikorwa.
Ibipimo byerekana umutwaro ufatika hamwe namakuru yimiterere.
Gukurikirana ibikoresho kubikosa, drift, ningaruka zabakoresha.
Tegura inzira yo gusubiza ibyabaye mbere yo gupima.
Komeza Ubushakashatsi
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Slurm for AI Training Clusters quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Ubuyobozi bukurikira
Imyitozo yihuta na Megatron
Ibibazo bikunze kubazwa
What is Slurm for AI Training Clusters?
Slurm nisoko rifunguye-ryakazi rishinzwe gucunga gahunda no gukora imirimo kumasoko yo kubara cyane, kandi byahindutse amahitamo asanzwe mumahugurwa manini ya AI. Nibyingenzi kuko ikwirakwiza rwose imyitozo nini ikora ibihumbi bya GPU.
Ni irihe tegeko abakoresha basanzwe bakoresha kugirango batange akazi kamwe muri Slurm?
sbatch itanga inyandiko yinyandiko isobanura ibikoresho namabwiriza kumurimo kumurongo wa Slurm.
Nigute akazi ka Slurm gasaba GPUs?
Sisitemu Rusange (GRES) sisitemu ireka imirimo isaba GPU mu buryo bweruye, urugero --gres = gpu: 8.
Nibihe Slurm ifata ibyemezo byingengabihe?
slurmctld ni mugenzuzi daemon uteganya imirimo, mugihe slurmd ikora kuri buri node kugirango itangire imirimo.
Kuki srun ihuza neza hamwe nogukwirakwiza amahugurwa nka PyTorch DDP?
srun itangiza imirimo ikomatanyirijwe hamwe kandi itanga urwego na master-adresse yamakuru yatanzwe amasomero akoresha kuri bootstrap.
Niyihe ntego yo gusubira inyuma muri Slurm?
Backfill itezimbere imikoreshereze ihuza imirimo mito muguteganya icyuho mugihe cyose badasubiza inyuma imirimo yabitswe.