Skip to content

Latest commit

 

History

History
19 lines (15 loc) · 1.63 KB

automated.md

File metadata and controls

19 lines (15 loc) · 1.63 KB

Inference Project Automation

  1. We have used chameleon cloud

    1. Steps to leasing a server from Chameleon cloud chameleon-lease and make instances according to the paper machine specifications. We recommend using Chameleon UC datacenter and compute_cascadelake_r_ib machines with with 96 Intel(R) Xeon(R) Gold 6240R CPU @2.40GHz cores and 188 Gb of RAM. Use ubuntu20.04 as the base images.

  2. On the Main node (Could be any of the cluster nodes):

    1. run source hack/zsh.sh in the root directory; reply 'Y' to setting zsh as default
    2. run source build.sh PUBLIC_IP in the project root directory. This will install all the necessary components, including kuberentes, supporting libraries (e.g. MLServer and load tester) and dependancies. It will also download data from the object storage. Logs of the previous experiments and modes to be stored in Minio Object storage are saved in the two google cloud bucket ipa-results-1 and ipa-models

  3. To make sure IPA routers are always scheduled on the manager node:

kubectl label node <manager node name> router-node=true

You can get the manager node name with:

hostname
  1. For attaching minions to the Main node and the cluster use this instructions, video and audio-qa/sent pipeline fit into one node so we recommend to test them, however in case you are interested to generate full results attach the extra nodes as per instructed in the paper