Copy pysdk v3 notebooks to default branch and remove 7 obviously redu… by lucasjia-aws · Pull Request #4898 · aws/amazon-sagemaker-examples

lucasjia-aws · Jul 17, 2026

…ndant and outdated versions

Issue #, if available:

Description of changes:

Testing done:

Merge Checklist

Put an x in the boxes that apply. You can also fill these out after creating the PR. If you're unsure about any of them, don't hesitate to ask. We're here to help! This is simply a reminder of what we are going to look for before merging your pull request.

I have verified that my PR does not contain any new notebook/s which demonstrate a SageMaker functionality already showcased by another existing notebook in the repository
I have read the CONTRIBUTING doc and adhered to the guidelines regarding folder placement, notebook naming convention and example notebook best practices
I have updated the necessary documentation, including the README of the appropriate folder as well as the index.rst file
I have tested my notebook(s) and ensured it runs end-to-end
I have linted my notebook(s) and code using python3 -m black -l 100 {path}/{notebook-name}.ipynb

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

…ndant and outdated versions

review-notebook-app · Jul 17, 2026

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

lucasjia-aws · Jul 17, 2026

All 13 comments are automated CodeGuru findings on companion .py scripts (mnist.py, aws_batch_resource_management.py, log_helpers.py, preprocessing.py) that were copied verbatim from upstream aws/sagemaker-python-sdk. Not making changes — breakdown below.

False positives — no such attack surface:

RAG / embedding access controls (preprocessing.py): it's a HuggingFace tokenizer preprocessing script. No RAG, no vector store.
NoSQL injection (mnist.py): MNIST training script, no DynamoDB/MongoDB anywhere.
OS command injection (mnist.py): the flagged line is os.environ["WORLD_SIZE"] = str(world_size) — setting an env var, no shell call.
Path traversal (mnist.py, x2): model_dir is the SageMaker-managed SM_MODEL_DIR env var, not untrusted input.
XSS (aws_batch_resource_management.py, x2): a boto3 helper that logs via json.dumps. No web/HTML output.

Already done in the code — the bot missed it:

Missing random seed (mnist.py): torch.manual_seed(args.seed) / torch.cuda.manual_seed are already called in train().
Missing eval() mode (mnist.py): test() calls model.eval() on its first line.

Style-only suggestions on upstream code:

Cyclomatic complexity / too many return values (log_helpers.py) and torch.inference_mode() vs torch.no_grad() (mnist.py): these files are byte-for-byte copies of the upstream SDK source. Refactoring them here forks them from their source of truth, so leaving as-is.

Bring in migration batches 1-3 (aws#4895, aws#4896, aws#4897) from default.

Copy pysdk v3 notebooks to default branch and remove 7 obviously redu…

87c054e

…ndant and outdated versions