scramjetorg / transform-hub

Scramjet Transform Hub (STH) is a runtime supervisor that can run data processing programs called Sequences and manage local resources on any Linux server, Docker on small edge servers, and even large-scale Kubernetes clusters in the cloud or datacenters. It connects to Scramjet Spaces in Scramjet Cloud Platform.
GNU Affero General Public License v3.0
67 stars 8 forks source link

GPU in Python Runner (docker) #928

Closed patuwwy closed 1 year ago

patuwwy commented 1 year ago

What? Start Runner container with GPU access enabled.

Why? ML

Usage:

cd packages/python-runner
yarn build:docker-tf
sth --runner-py-image scramjetorg/runner-py-tf:5daaf8ea450d90e7b969be42e2d67c06c253b093
si seq send <SEQ_DIR>
si seq start - --limits "{\"gpu\": true}"
si seq deploy <SEQ_DIR>--limits "{\"gpu\": true}"

Clickup Task: https://app.clickup.com/t/24308805/VDM-1364 https://app.clickup.com/t/862k8pzhh

How it works:

https://docs.docker.com/compose/gpu-support/#enabling-gpu-access-to-service-containers

Review checks:

These aspects need to be checked by the reviewer: