Ray fun Pinpin AI
Ray jẹ ilana orisun-ìmọ ti o jẹ ki o rọrun lati ṣe iwọn Python ati awọn iṣẹ ṣiṣe AI lati kọǹpútà alágbèéká kan si iṣupọ ẹgbẹẹgbẹrun awọn ẹrọ.
Akopọ
It matters because it gives a simple, unified way to distribute training, tuning, data processing, and serving without rewriting your code for each.
Jin Dive
Ero mojuto Ray ni titan awọn iṣẹ Python lasan ati awọn kilasi sinu awọn ipin pinpin pẹlu iyipada kekere. Iṣẹ kan ti a samisi bi 'iṣẹ' latọna jijin nṣiṣẹ ni asynchronously lori eyikeyi oṣiṣẹ ninu iṣupọ; kilasi ti a samisi bi 'oṣere' latọna jijin di iṣẹ ti ipinlẹ ti o ngbe lori oṣiṣẹ kan. Ray da awọn ọjọ iwaju iwuwo fẹẹrẹ pada (awọn itọkasi nkan) ati mimu ṣiṣe ṣiṣe ṣiṣe, gbigbe data nipasẹ ile itaja ohun kan ti o pin, ati ifarada ẹbi. Lori oke ti awọn ile-ikawe ti a ṣe idi idi pataki yii: Ray Train fun ikẹkọ awoṣe pinpin, Ray Tune fun wiwa hyperparameter, Ray Data fun ṣiṣan data pipelines, RLlib fun ẹkọ imuduro, ati Ray Sin fun iṣẹ awoṣe iwọn. Eyi jẹ ki iṣupọ kan mu gbogbo ṣiṣiṣẹsẹhin ML kan lati opin si ipari.
Imọ-imọ-ẹrọ
Awọn alakoko akọkọ jẹ awọn iṣẹ-ṣiṣe (aini ipinlẹ, awọn ipe iṣẹ ti o jọra) ati awọn oṣere (awọn oṣiṣẹ ipinlẹ ti o mu awọn nkan bii awoṣe ti kojọpọ tabi counter). Nigbati o ba pe iṣẹ-ṣiṣe latọna jijin, Ray lẹsẹkẹsẹ pada si iwaju ati ṣeto iṣẹ naa kọja awọn CPUs/GPU ti o wa; o pe ray.get () lati mu awọn esi. Itaja ohun-ini iranti ti a pin kaakiri pẹlu iranti idaako pinpin odo n gbe awọn nkan nla bii awọn akojọpọ laarin awọn oṣiṣẹ ni imunadoko, yago fun isọdọtun leralera ati ṣiṣe data-eru data AI pipelines.
Ipa Ilana
Iye owo ati isuna
Awọn ipinnu faaji ṣe awakọ iṣẹ ati idiyele iṣẹ fun awọn ọdun.
Awọn ipinnu diẹ sii
Ẹkọ imọ-ẹrọ ṣe iranlọwọ fun awọn ẹgbẹ lati yan akopọ to tọ, kii ṣe ọkan tuntun nikan.
Iṣakoso didara
Awọn yiyan imọ-ẹrọ to dara julọ dinku awọn iṣẹlẹ igbẹkẹle ni iṣelọpọ.
Ojo iwaju ti Ray fun AI pinpin
Ray ti di eegun ẹhin fun AI iwọn-nla, ni pataki ti a lo ninu ikẹkọ ati ṣiṣe awọn awoṣe ede nla. Reti idagbasoke ni iṣẹ-iṣẹ LLM-pato (Ray Serve pẹlu vLLM), ṣiṣe eto GPU oriṣiriṣi, isọpọ titọ pẹlu awọn adagun data ati Kubernetes nipasẹ KubeRay, ati adaṣe adaṣe ti o dara julọ fun awọn iṣẹ ṣiṣe ipilẹṣẹ spiky. Bi awọn awoṣe ti ndagba, ipa Ray ni ṣiṣe adaṣe ikẹkọ ọpọ-opopona, awọn opo gigun ti epo RLHF, ati itọkasi ipele kọja ẹgbẹẹgbẹrun awọn accelerators ṣee ṣe lati faagun.
Real-World imuse
Ṣiṣe Ray Tune lati wa awọn ọgọọgọrun awọn akojọpọ hyperparameter ni afiwe kọja iṣupọ GPU kan lati wa iṣeto awoṣe to dara julọ
Lilo Ray Train lati kaakiri ikẹkọ ti awoṣe ẹkọ ti o jinlẹ kọja ọpọlọpọ awọn GPUs ati awọn apa pẹlu awọn iyipada koodu iwonba
Ṣiṣe opo gigun ti epo-ipin pẹlu Ray Data lati ṣe Dimegilio awọn miliọnu awọn igbasilẹ nipa ṣiṣanwọle wọn nipasẹ awoṣe kan kọja iṣupọ kan
Gbigbe awọn awoṣe lọpọlọpọ lẹhin aaye ipari autoscaling kan pẹlu Ray Serve lati mu ijabọ iṣelọpọ oniyipada
Awọn ewu & Awọn ọna iṣọ
Ṣiṣepe ala-ilẹ kan le tọju awọn ailagbara eto ti o gbooro.
Awọn ohun elo amayederun ati awọn idiyele itọju nigbagbogbo ni aibikita.
Aabo ati awọn ela akiyesi le dagba bi awọn eto ṣe di eka sii.
Ilana Ilana imuse
Ṣetumo lairi, didara, ati awọn ibi-afẹde idiyele ṣaaju imuse.
Aṣepari labẹ ẹru ojulowo ati awọn ipo data.
Abojuto ohun elo fun awọn aṣiṣe, fiseete, ati ipa olumulo.
Mura ipadasẹhin pada ati awọn ipa ọna esi iṣẹlẹ ṣaaju iwọn.
Tesiwaju Ṣiṣawari
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 Ray for Distributed AI 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
Itọsọna atẹle
Ṣiṣayẹwo Gradient
Awọn ibeere ti a beere nigbagbogbo
What is Ray for Distributed AI?
Ray jẹ ilana orisun-ìmọ ti o jẹ ki o rọrun lati ṣe iwọn Python ati awọn iṣẹ ṣiṣe AI lati kọǹpútà alágbèéká kan si iṣupọ ẹgbẹẹgbẹrun awọn ẹrọ. O ṣe pataki nitori pe o funni ni ọna ti o rọrun, ti iṣọkan lati kaakiri ikẹkọ, iṣatunṣe, sisẹ data, ati ṣiṣe laisi atunkọ koodu rẹ fun ọkọọkan.
Isoro wo ni Ray nipataki yanju?
Ray n pese ọna iṣọkan, ọna ti o rọrun lati pin kaakiri iṣiro kọja ọpọlọpọ awọn ero laisi atunkọ koodu rẹ fun iru fifuye iṣẹ kọọkan.
Ni Ray, kini iyatọ laarin 'iṣẹ-ṣiṣe' ati 'oṣere' kan?
Awọn iṣẹ-ṣiṣe jẹ awọn iṣẹ isakoṣo latọna jijin ti orilẹ-ede nṣiṣẹ ni afiwe, lakoko ti awọn oṣere jẹ awọn ohun elo pipẹ ti o tọju ipo, bii awoṣe ti kojọpọ.
Kini Ray pada lẹsẹkẹsẹ nigbati o ṣe ifilọlẹ iṣẹ-ṣiṣe latọna jijin?
Ray pada a lightweight ojo iwaju lẹsẹkẹsẹ ki iṣẹ nṣiṣẹ asynchronously; o pe ray.get () lati gba esi gangan nigba ti o nilo.
Iru ile-ikawe Ray wo ni a ṣe fun wiwa hyperparameter pinpin?
Ray Tune ṣe amọja ni ṣiṣiṣẹ ọpọlọpọ awọn idanwo hyperparameter ni afiwe kọja iṣupọ kan.
Bawo ni ile-itaja ohun Ray ṣe jẹ ki awọn opo gigun ti AI wuwo data daradara?
Ile-itaja ohun iranti ti o pin ni lilo ẹda-odo kika kika ti o tobi le jẹ wọle nipasẹ awọn oṣiṣẹ laisi isọdọtun iye owo.