Skip to content

NCCL benchmark error #352

@ultrons

Description

@ultrons

The command snippet for GKE cluster creation as specified in a3/README.md is missing user's own credential mount (which is correctly done in the main readme for MIG example).
It's also missing working GKE version specification.
NCCL benchmark example as well as LIT GPT fail with cryptic errors until I mentioned the GKE version = "1.27.8-gke.1067000"
(Thanks to the recommendation from @samuelkarp.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions