Commit Graph

2 Commits

Author SHA1 Message Date
Hannah R. b47a870284 fix(mtrl): Replace internal STAGE-based S3 path with customer placeholder (#4904)
The 3rd cell of mtrl_finetuning_example_notebook_v3_prod.ipynb set
S3_TRAINING_DATA using an f-string that referenced an undefined STAGE
variable and an internal `-rftjob-input` bucket pattern left over from
when MTRL was tested in gamma/prod. Executing the cell raised
`NameError: name 'STAGE' is not defined`.

Replace it with a customer-facing placeholder S3 URI, consistent with
the convention used in the sibling sft_finetuning notebook, so users
(and the notebook test engine) can substitute their own dataset path.

Co-authored-by: hrehard <hrehard@amazon.com>
2026-07-30 15:05:30 -07:00
Hannah R. 5a49643c6d Migrate NotebookTestEngine notebooks to the v3-examples suite (#4901)
* feat(notebooks): Seed NotebookTestEngine branch with v3-examples suite

Remove the amazon-sagemaker-examples example notebooks (8 category
folders) and replace them with the SageMaker Python SDK v3-examples
notebook suite from aws/sagemaker-python-sdk master:
- inference-examples, ml-ops-examples, model-customization-examples,
  training-examples at the repo root
- the two loose root notebooks (sagemaker_v3_setup.ipynb,
  iam_role_validation.ipynb) moved into uncategorized/

This layout matches the Notebook Test Engine category rotation
(KNOWN_CATEGORY_FOLDERS + uncategorized catch-all).

* ci: Add Notebook Test Engine PR-check workflow

Add .github/workflows/notebook-tests.yml: on a PR touching any notebook
(**/*.ipynb) targeting the NotebookTestEngine branch, assume the engine
role via GitHub OIDC and start the notebook-test-engine CodeBuild project
in pr_check mode, poll to completion, and reflect the build result as the
PR check (failing notebook rendered + uploaded as a failing-notebooks
artifact). Collaborator gate: collaborators auto-approve, others block on
the manual-approval environment.

* ci: Set MAX_CONCURRENT=18 for PR-check runs

PR runs need a higher concurrency cap than full-scans (project default 15)
because the engine throttles against an account-wide InProgress
processing-job count. With the quota at 20, PR=18 lets a PR claim the
reserved headroom even while a scan holds jobs, leaving ~2 slots for
internally-launched jobs (e.g. Clarify).

* ci: Sync PR-check workflow with default branch (collab github.token fallback)

* fix(ci): Use unpinned refs/pull/N/head for PR checkout

Sync with the default-branch workflow (checkout the PR ref tip).

* Remove workflow file from NotebookTestEngine (it lives on the default branch)

---------

Co-authored-by: hrehard <hrehard@amazon.com>
2026-07-24 10:28:41 -07:00