Ntụziaka nka

Slurm maka ụyọkọ Ọzụzụ AI

Slurm bụ onye njikwa ibu ọrụ mepere emepe nke na-ahazi ma na-arụ ọrụ na ụyọkọ kọmpụta dị elu, ọ ghọwokwa nhọrọ ndabara maka nnukwu ọzụzụ AI.

2 nkeji na-agụEmelitere ikpeazụ

Nchịkọta

It matters because it reliably distributes massive training runs across thousands of GPUs.

Ime miri emi

Slurm (Mfe Linux Utility for Resource Management) sitere na supercomputing ma na-enye ike ugbu a ọtụtụ ụyọkọ ọzụzụ AI kachasị ukwuu n'ụwa. Ndị ọrụ na-edobe scripts batch na sbatch, rịọ akụrụngwa dị ka ọnụ ọnụ na GPU nwere ntuziaka dị ka --gres=gpu: 8, na Slurm kwụ n'ahịrị, na-ebute ụzọ, wee malite ọrụ ahụ. Onye na-ebupụta srun ya na-akwalite usoro achikọtara n'ofe ọnụ, nke jikọtara ya na usoro ekesara dị ka PyTorch DDP na NCCL. Slurm na-agbaso ndekọ ndekọ akụrụngwa, na-amanye oke oke na oke nkewa, ma na-ejikwa nhazi oge njupụta iji tinye obere ọrụ n'ime oghere. Maka ọzụzụ ihe atụ n'ihu, ndị otu na-adabere na Slurm iji jikwaa puku kwuru puku GPU, malitegharịa site na ebe nlele mgbe ọdịda ọnụ gasịrị, ma debe ikike raara onwe ya nye maka ọsọ ọtụtụ izu.

Nghọta nka nka

Onye na-ahụ maka Slurm daemon (slurmctld) na-eme mkpebi nhazi oge ebe onye nnọchi anya n'ọnụ ọnụ ọ bụla na-ebupụta ọrụ yana ọkwa ọkwa. Ihe mgbakwunye Generic Resource (GRES) na-akwado GPU ka ọrụ wee rịọ ha n'ụzọ doro anya. srun na-esetịpụ mgbanwe gburugburu ebe obibi (ọkwa, nha ụwa, adreesị nna ukwu) nke na-ekesa ọba akwụkwọ ọzụzụ na-agụ na nzikọrịta ozi NCCL bootstrap. Ịhazi nhazi azụ azụ na-eme ka ọrụ dị mkpụmkpụ na-agba ọsọ n'oge ma ọ bụrụhaala na ha anaghị egbu oge ndoputa dị elu, na-eme ka ojiji dị elu.

Mmetụta atụmatụ

Ọnụ ego na mmefu ego

Mkpebi ihe owuwu ụlọ na-akwalite arụmọrụ yana ọnụ ahịa ọrụ ruo ọtụtụ afọ.

Mkpebi doro anya

Nkà mmụta nka na-enyere ndị otu egwuregwu aka ịhọrọ nchịkọta ziri ezi, ọ bụghị naanị nke kachasị ọhụrụ.

Quality akara

Nhọrọ injinia ka mma na-ebelata ihe omume ntụkwasị obi na mmepụta.

Ọdịnihu nke Slurm maka ụyọkọ Ọzụzụ AI

Slurm na-aga n'ihu na-agbakwunye mgbawa igwe ojii, nkwado akpa site na Pyxis na Enroot, yana njiri mara GPU siri ike. Dị ka ụyọkọ AI na-erute 100,000-gbakwunyere GPUs, na-atụ anya nnabata mmejọ siri ike, ntinye nyocha-akpaghị aka, yana ọrụ na-agbanwe agbanwe na-agbanwe nha mgbe ọdịda gasịrị. Ọtụtụ ụlọ ọrụ ugbu a na-agba ọsọ Slurm n'akụkụ ma ọ bụ n'okpuru Kubernetes, na ndị na-ahazi ngwakọ na-achọ ijikọ arụmọrụ ụdị HPC na mgbanwe igwe ojii-nwa afọ maka ọzụzụ na-ebuwanye ibu.

Mmejuputa n'ezie n'ụwa

Ụlọ nyocha ihu na-ewepụta ọzụzụ ọzụzụ ọtụtụ izu gafere puku kwuru puku GPU nwere otu edemede sbatch na-arịọ ọtụtụ narị ọnụ.

Onye nyocha nyefere 'srun --gres=gpu:8' iji jide GPU asatọ n'otu ọnụ maka nnwale PyTorch DDP.

Ịhazi oge nlegharị anya na-ewepụta obere ọrụ nleba anya n'ime GPU ndị na-adịghị arụ ọrụ ebe nnukwu ọzụzụ ọzụzụ echekwara na-echere ịmalite.

Mgbe ọnụ na-ada n'etiti oge, Slurm na-arịọ ọrụ ahụ ma maliteghachi na ebe nlele ọhụrụ kama ịmalite ọzọ.

Ihe ize ndụ & okporo ụzọ nche

Ịkwalite otu akara ngosi nwere ike zoo adịghị ike sistemụ sara mbara.

A na-eledakarị ihe akụrụngwa na ụgwọ ọrụ anya.

Ọdịiche nchekwa na nleba anya nwere ike itolite ka sistemu na-adịwanye mgbagwoju anya.

Map mmejuputa

1

Kọwaa latency, ịdịmma na ebumnuche ọnụ ahịa tupu mmejuputa ya.

2

Benchmark n'okpuru ibu dị adị na ọnọdụ data.

3

Nleba anya akụrụngwa maka mperi, ịkpafu na mmetụta onye ọrụ.

4

Kwadebe ụzọ nzaghachi azụghachi azụ na ihe omume tupu ịchachaa.

Nọgide na-eme nchọpụta

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.

Malite ajụjụ

Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation

Ntuziaka na-esote

DeepSpeed ​​​​na Megatron Ọzụzụ Stacks

Ajụjụ a na-ajụkarị

What is Slurm for AI Training Clusters?

Slurm bụ onye njikwa ibu ọrụ mepere emepe nke na-ahazi ma na-arụ ọrụ na ụyọkọ kọmpụta dị elu, ọ ghọwokwa nhọrọ ndabara maka nnukwu ọzụzụ AI. Ọ dị mkpa n'ihi na ọ na-ekesa nnukwu ọzụzụ na-aga n'ofe puku GPU.

Kedu iwu ndị ọrụ na-ejikarị nyefee ọrụ batch na Slurm?

sbatch na-enyefe script ogbe nke na-akọwa akụrụngwa na iwu maka ọrụ na Slurm kwụ n'ahịrị.

Kedu ka ọrụ Slurm si arịọ GPU?

Usoro Generic Resource (GRES) na-ahapụ ọrụ ka ọ rịọ GPU n'ụzọ doro anya, dịka ọmụmaatụ --gres=gpu:8.

Kedu akụrụngwa Slurm na-eme mkpebi nhazi oge etiti?

slurmctld bụ daemon njikwa na-ahazi ọrụ, ebe slurmd na-agba ọsọ na ọnụ nke ọ bụla iji malite ọrụ.

Kedu ihe kpatara na-ejikọta nke ọma na usoro ọzụzụ ekesa dị ka PyTorch DDP?

srun na-ewepụta ọrụ emekọrịtara ọnụ n'ofe ọnụ ma na-enye ozi ọkwa na nna-ukwu nke ụlọ akwụkwọ na-ekesa na-eji bootstrap.

Gịnị bụ ebumnobi nke nhazi oge njupụta na Slurm?

Backfill na-eme ka ojiji ya dịkwuo mma site n'itinye obere ọrụ n'ime nhazi oge ọ bụrụhaala na ha anaghị azụghachi ọrụ echekwabara.