Kubernetes ya ML Imirimo Yakazi
Kubernetes ni sisitemu ifunguye-sisitemu ihita itegura gahunda, umunzani, kandi igatangira porogaramu yabitswe muri cluster yimashini.
Incamake
For machine learning, it lets teams pack GPU-hungry training jobs and latency-sensitive model servers onto shared hardware without babysitting individual servers.
Kwibira cyane
Ubusanzwe yubatswe kuri Google kugirango ukore serivise zurubuga, Kubernetes ifata cluster yawe nkikidendezi kinini cya CPU, kwibuka, na GPUs, hanyuma igahitamo imashini ikoresha buri kintu. Amakipe ya ML ayishingikirije kuberako akazi kenshi karaturika kandi karahenze: imyitozo irashobora gukenera GPU umunani mumasaha atandatu, hanyuma ntakintu. Kubernetes ingengabihe yerekana kuri node hamwe na GPU yubuntu, kandi iyo akazi karangiye birekura ibyuma. Bituma kandi seriveri zifatika zizima, zigatangira ibintu byangiritse no gukwirakwiza kopi kumashini kugirango zihangane. Ibikoresho byubatswe hejuru, nka Kubeflow, Ray, na KServe, ongeraho ibice byihariye bya ML nko kugabura-guhugura abakoresha, guhuza hyperparameter, hamwe na autoscaling yerekana impera, bityo abahanga mu bumenyi bakorana na abstraction yo mu rwego rwo hejuru aho kuba YAML mbisi.
Ubushishozi
Kubernetes igenera GPU ibinyujije mumacomeka yamamaza yamamaza ibikoresho nka nvidia.com/gpu, iyo gahunda ihuye nibyifuzo bya pod. Irangi no kwihanganira bituma imirimo ya CPU ihendutse itahenze kuri GPU ihenze, mugihe abatoranya node hamwe n amategeko agenga imiyoborere pin imyitozo kubikoresho byihariye. Kubwamahugurwa menshi-GPU, abashoramari barema itsinda ryibiti bivumburana kandi bigakoresha urwego nka PyTorch DDP cyangwa Horovod, guhanahana gradients hejuru ya cluster ikoresheje NCCL.
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.
Kazoza ka Kubernetes kubikorwa bya ML
Tegereza gukomera kwa ML: gahunda y'agatsiko itangiza ibyiciro byose byagabanijwe-icyarimwe icyarimwe cyangwa ntanakimwe na kimwe, kugabana GPU kugabanywa no kugabanya igihe kuburyo imirimo myinshi yoroheje isangira ikarita imwe, hamwe na topologiya-yamenyesheje ahantu hubahirizwa imiyoboro ya NVLink yihuse. Seriveri idafite umwanzuro kuri Kubernetes, gupima amaherezo kuri zeru hagati y'ibisabwa, birakuze. Nka ballon yicyitegererezo, abategura gahunda barushaho guhuza ibice byinshi hamwe nibicu, hamwe na sisitemu ishingiye kumurongo-gusaranganya imurikagurisha nka Kueue na Volcano bigenda biba bisanzwe mugucunga ubushobozi buke bwa GPU.
Gushyira mu bikorwa Isi
Laboratoire yubushakashatsi ikoresha Kubeflow Training Operator kugirango itangire 32-GPU PyTorch yagabanijwe-amahugurwa akazi kumpande enye, hanyuma ihita irekura GPU iyo ihuye.
Isosiyete ikora e-ubucuruzi ikora icyitegererezo cyayo hamwe na KServe, iyo autoscales ikoporora mugihe cyo kugurisha flash hanyuma ikamanuka ijoro ryose.
Banki ikora imirimo yo gutanga amanota nijoro nka Kubernetes CronJobs, ibatonda umurongo kuri node ya CPU isanzwe kugirango badahangana na traffic traffic kumanywa.
Gutangira ukoresha Ray kuri Kubernetes kugirango ukore parallel hyperparameter yohanagura, uzunguruka ibice byinshi byigihe gito cyo kugerageza kumwanya wabigenewe kugirango ugabanye igiciro.
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 Kubernetes for ML Workloads 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
A / B Ikizamini cya ML Model
Ibibazo bikunze kubazwa
What is Kubernetes for ML Workloads?
Kubernetes ni sisitemu ifunguye-sisitemu ihita itegura gahunda, umunzani, kandi igatangira porogaramu yabitswe muri cluster yimashini. Kwiga imashini, ireka amakipe apakira akazi ka GPU ashonje hamwe nakazi ka moderi yerekana ubukererwe kubikoresho bisangiwe nta kurera seriveri kugiti cye.
Nuwuhe murimo wibanze wa gahunda ya Kubernetes kumurimo wa ML?
Gahunda ihuye na pod yumutungo usaba (CPU, kwibuka, GPUs) kurwanya node iboneka hanyuma igashyira pod aho ihuye. Ntabwo ikora kode yicyitegererezo cyangwa amakuru.
Nigute Kubernetes ituma GPU iboneka kuri pod?
Amacomeka yibikoresho yerekana GPU nkibikoresho byateganijwe (urugero, nvidia.com/gpu), kureka pods ibasaba hamwe na gahunda yo kuboneka iboneka.
Ni ubuhe butumwa no kwihanganira bikoreshwa muri cluster ya ML?
Umwanda wirukana inkono kuva kuri node; ibishishwa gusa hamwe no kwihanganira bihuye birashobora kugwa hariya. Ibi bibitse GPU idafite akazi kubikorwa bikeneye GPUs.
Niki gikoresho cyongeramo ubushobozi bwa ML nkukwirakwiza-imyitozo hejuru ya Kubernetes?
Kubeflow ibice ML ikora kuri Kubernetes, harimo abakora imyitozo, imiyoboro, hamwe no guhuza, bityo amakipe yirinda kwandika intoki-urwego rwo hasi rwa cluster config.
Kuki Kubernetes ikwiranye no guturika ML akazi?
Amahugurwa ni spiky: GPU nyinshi muri make, hanyuma ntayo. Kubernetes ishyira akazi mugihe umutungo ari ubuntu ukayirekura kurangiza, kunoza imikoreshereze.