[GSoC] Project 7: GPU Testing for LLM Blueprints

[jaiakash]

This is the tracking issue for the GSoC Project 7: GPU Testing for LLM Blueprints. The initial issue is #2432.

GSoC Proposal Doc - Link

The new feature aims at setup self runner ( using GitHub ARC ) where we can execute GPU intensive tasks on dedicated GPU infra.

Tasks

• Convert GSoC proposal to KEP. feat: KEP-2432: GPU Testing for LLM Blueprints #2689
• Have sample LLM bluprint for testing.
• Setup CI Action to direct LLM task to GPU infra.
• Setup OKE Infra with GPU nodes
• Setup ARC on OKE Cluster
• Setup metrics and analytics

cc @andreyvelich @varodrig @Electronic-Waste

[jaiakash]

/assign
/area gsoc
/kind feature

[Electronic-Waste]

Great! Wish to see more works in the future.

[jaiakash]

• feat: refactor and merge setup-gpu-cluster into single script based on flag. #2812

• Update OCI GPU VM tf and bootstrap file #2809

• fix: gpu e2e test to run on pull_request_target and use kubeflow/trainer secret #2814

• feat: add HF token and allow gpu workflow to run from pull request target #2818

• feat: KEP-2432: GPU Testing for LLM Blueprints #2689

• feat: support for managing gpu enabled self runner infra #2762

• fix: teraform for oci gpu based vm #2810

[jaiakash]

• [GH ARC]: Docs for setup of ARC, yaml manifests, config and gpu based cloudrunner code. #2849
• [GH ARC]: Setup Ubuntu 24.04, GPU A10 custom image on cncfkubeflow OCI tenancy #2848