Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Kubernetes fun ML Workloads adanwo

Apply Kubernetes ML pod placement, GPU device resources, and taint tolerations, including why distributed training needs appropriate per-node capacity.

Jẹmọ itọsọna awọn ọnaKubernetes Fun ML Workloads
Ibeere 1 ti 8

Kini iṣẹ akọkọ ti oluṣeto Kubernetes fun awọn iṣẹ ṣiṣe ML?

Awotẹlẹ gbogbo ibeere8 ibeere

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

  1. Kini iṣẹ akọkọ ti oluṣeto Kubernetes fun awọn iṣẹ ṣiṣe ML?
    • Kikọ koodu ikẹkọ fun awọn onimọ-jinlẹ data
    • Ti pinnu iru ipade ti o wa ninu iṣupọ n ṣiṣẹ adarọ-ese kọọkan ti a fi sinu apoti
    • Iforukọsilẹ data ikẹkọ laifọwọyi
    • Yiyan faaji nẹtiwọki nkankikan
  2. Bawo ni Kubernetes ṣe deede jẹ ki awọn GPU wa si adarọ-ese kan?
    • Nipasẹ awọn afikun ẹrọ ti o polowo awọn orisun bii nvidia.com/gpu
    • Nipa imeeli si ataja GPU fun iwe-aṣẹ kan
    • GPU ko ṣee lo ni Kubernetes
    • Nipa yiyipada awọn ohun kohun Sipiyu sinu awọn GPU ni akoko asiko
  3. With a NoSchedule taint on a GPU node, what does a matching pod toleration do?
    • Forces the pod onto that node regardless of capacity
    • Changes the model’s optimizer
    • Allows the pod to be considered for that node without guaranteeing placement
    • Encrypts every GPU tensor
  4. Ọpa wo ni o ṣafikun awọn agbara pato-ML bi awọn oniṣẹ ikẹkọ pinpin lori oke Kubernetes?
    • Microsoft Ọrọ
    • Apache fun ṣiṣe awọn oju-iwe wẹẹbu aimi
    • Makiro iwe kaunti kan
    • Kubeflow
  5. Kini idi ti Kubernetes ṣe deede si awọn ẹru iṣẹ ML ti nwaye?
    • O ṣeto awọn iṣẹ sori awọn orisun ọfẹ ati sọ di ohun elo yẹn nigbati awọn iṣẹ ba pari
    • O ni ipamọ gbogbo awọn GPUs patapata fun olumulo kan
    • O ṣe idiwọ iṣẹ eyikeyi lati dinku nigbagbogbo
    • O kan gba eiyan kan fun iṣupọ
  6. Kini 'iwọn aaye ipari ipari si odo' tumọ si lori Kubernetes?
    • Piparẹ awoṣe patapata
    • Ṣiṣeto deede si ogorun odo
    • Ṣiṣe awọn adakọ odo nigbati ko si ijabọ, lẹhinna yiyi soke lori ibeere atẹle
    • Atehinwa awọn awoṣe ká sile si odo
  7. One ML pod requests four GPUs, while two nodes each have only two available. Can ordinary scheduling combine those nodes for that one pod?
    • Yes; every pod automatically spans all cluster nodes
    • Yes; the scheduler converts CPU cores into the missing GPUs
    • Only if the pod name contains the word distributed
    • No; the pod must fit on one eligible node
  8. An inference service scales to zero while idle. Which user-facing delay should its team test when traffic resumes?
    • The training-set labeling speed
    • Cold-start delay while a serving replica becomes ready
    • The file size of an unrelated notebook
    • The number of discarded vocabulary tokens