test: fix large job

[cdoern]

Fixes: #494

[github-actions]

E2E (NVIDIA L40S x4) workflow launched on this PR: View run

[ktdreyer]

I edited the PR description to link this with #496 , hope you don't mind.

From discussion with @cdoern today: A while ago the .github/workflows/e2e-nvidia-l40s-x4.yml file in this repository was copied from https://github.com/instructlab/instructlab/blob/main/.github/workflows/e2e-nvidia-l40s-x4.yml , and nothing/no one keeps it up-to-date with the instructlab copy. This training copy can fall behind the instructlab canonical version. @cdoern has copied the latest changes from instructlab into this repo in this PR.

[github-actions]

e2e workflow failed on this PR: View run, please investigate.

[booxter]

#496 was merged and linked issue closed. Do we still need this PR?

[ktdreyer]

Yeah, we solved #494 with #496, rather than solving it in this PR.

Charlie's point from last week remains true: The .github/workflows/e2e-nvidia-l40s-x4.yml file in this tree was a copy of the one in instructlab's tree, and when we edit instructlab's copy without making those changes here as well, then this file falls out of date, and it's difficult to root-cause CI failures.

To fix the immediate issue, we could manually read through the .github/workflows/e2e-nvidia-l40s-x4.yml file in instructlab and make any synchronization changes here.

I'm not sure how we would improve this long-term. Maybe there are more code items or YAML blocks we could centralize to https://github.com/instructlab/ci-actions/ .