[mtrl] Fix undefined STAGE in S3_TRAINING_DATA path

[Hannah14295]

Issue

The 3rd code cell of v3-examples/model-customization-examples/mtrl_finetuning_example_notebook_v3_prod.ipynb fails to execute with NameError: name 'STAGE' is not defined.

The cell built the training-data path from an f-string referencing an undefined STAGE variable and an internal -rftjob-input bucket pattern:

S3_TRAINING_DATA = f"s3://{REGION}-{CUSTOMER_ACCOUNT}-{STAGE}-rftjob-input/test-data/prompts/math-100-parquet/"

REGION and CUSTOMER_ACCOUNT are defined earlier in the notebook, but STAGE is not — a leftover from when MTRL was tested internally in gamma/prod. An SA following the MTRL best-practices blog and this notebook hit this error.

Fix

Replace the internal STAGE-based path with a customer-facing placeholder S3 URI, matching the convention already used in the sibling sft_finetuning_example_notebook_pysdk_prod_v3.ipynb:

S3_TRAINING_DATA = "s3://<your-bucket>/<path-to-your-training-data>/"  # TODO: replace with your training data S3 URI

Customers (and the notebook test engine, which substitutes a real dataset S3 URI) can now fill in their own dataset path.

Notes

• Single-line change; notebook validated as well-formed JSON.
• Keeps this notebook in sync with the same fix in aws/amazon-sagemaker-examples ([mtrl] Fix undefined STAGE in S3_TRAINING_DATA path amazon-sagemaker-examples#4904).