[evakravi]

Problem

ModelBuilder.deploy() takes tags only through **kwargs, so inspect.signature(ModelBuilder.deploy) does not list it and a typo in the name is a silent no-op. Two deploy paths drop it: the fine-tuned model package path reads only self._tags, and the Nova path reads neither. The accumulated tags also have two shapes in one SDK. add_tags compares tag["Key"], so the {"key": ..., "value": ...} form that the sagemaker.core.resources classes take raises KeyError: 'Key', and a sagemaker.core.shapes.Tag raises TypeError: 'Tag' object is not subscriptable, while the fine-tuned and Nova paths convert the stored tags back to lowercase by hand. The SageMaker agent skills carry a commit whose only purpose is to pass {"Key", "Value"} dicts to dodge the KeyError.

Reproduction on origin/master:

>>> _validate_new_tags([{"key": "a", "value": "1"}], [{"Key": "b", "Value": "2"}])
KeyError: 'Key'
>>> _validate_new_tags([Tag(key="a", value="1")], [{"Key": "b", "Value": "2"}])
TypeError: 'Tag' object is not subscriptable

Solution

deploy() gets a named tags: Optional[Tags] = None parameter. format_tags normalizes every accepted form once (a {key: value} mapping, boto-style dicts, core-style dicts, Tag objects, or a mix) into boto-style dicts, raises TypeError for a tag with no key, and add_tags, tag_exists, _validate_new_tags, and remove_tag_with_key go through it. deploy() merges the caller's tags into the builder before dispatch, so the built-model, inference-component, recommendation, fine-tuned, restricted-package, and Nova paths all tag the endpoint from the same list. One helper, _core_endpoint_tags(), replaces the two hand-written case conversions. build(tags=...) from #6510 uses the same normalization.

Tests

New sagemaker-serve/tests/unit/test_model_builder_deploy_tags.py, 13 tests: format_tags on every representation and its TypeError cases, _validate_new_tags with mixed casing and Tag objects, remove_tag_with_key on core casing, add_tags normalization, the tags parameter in the signature, the merge into the builder before the core path, _deploy_core_endpoint normalizing kwargs tags, the recommendation path receiving the builder's tags, _core_endpoint_tags() casing, the restricted-package endpoint being tagged, and the Nova endpoint merging the builder's tags.

cd sagemaker-serve && python -m pytest tests/unit/test_model_builder_deploy_tags.py -q
13 passed

On origin/master the same file reports 13 failed (KeyError: 'Key' in five tests, the rest on the missing parameter and the dropped tags).

Existing suites on this branch: sagemaker-core tests/unit/test_common_utils.py -k tag: 16 passed. sagemaker-serve test_model_builder_build_tags.py, test_model_builder_deploy.py, test_merged_model_deployment.py, test_model_builder_core.py, test_model_builder_methods.py, test_model_builder_utils_methods.py, test_model_builder_register.py, test_inference_recommendation_mixin.py, test_model_builder_additional.py: 188 passed, 6 skipped, the one failure (test_get_client_translators_with_torch_tensor, No module named 'torch') fails identically on origin/master.

black --line-length 100, flake8, and pydocstyle pass on the changed files.

Merge Checklist

• I read the contribution guide.
• The change is backward compatible.
• The commits use the repository commit-message format.
• The change creates no S3 or STS client.
• The change has regression coverage and adds no dependency.

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

[evakravi]

Superseded by #6533, which combines the four changes into one PR and moves the hub-model metadata accessors onto HubContent itself (a HubContentMixin in sagemaker.core.hub_content) instead of a separate class.

[github-actions]

🤖 AI code review

This PR makes ModelBuilder.deploy(tags=...) a named parameter, centralizes tag normalization in format_tags/_normalize_tag so all accepted representations (boto dicts, core-style dicts, Tag objects, {key: value} mappings) become boto-style {"Key","Value"} dicts, and threads a merged tag list through every deploy path (built model, IC, recommendation, fine-tuned, restricted-package, Nova) via add_tags/_core_endpoint_tags. It aligns with prior precedent (PR #6179) and adds focused unit tests. The change is sound; I verified callers of format_tags, _validate_new_tags, and remove_tag_with_key in sagemaker-core, the _deploy_core_endpoint flow, imports, and the existing tag tests. Only one minor edge case worth noting.

_{Reviewed commit 825999b. Automated review; verify before acting.}

[github-actions]

🟡 Low: The dict branch now treats any mapping containing a "Key" or "key" entry as a single tag rather than a {tag_key: tag_value} mapping. This correctly disambiguates a single boto/core tag dict, but it silently changes behavior for a legitimate {tag_key: value} mapping that happens to use "Key"/"key" as one of its tag keys: e.g. format_tags({"Key": "prod", "team": "ml"}) now yields a single tag {"Key": "prod", "Value": None} and drops the "team" entry, whereas before it produced two tags. This is an inherent ambiguity of overloading dict, and unlikely in practice, but it is a new silent data-loss path and may be worth a short note in the docstring or a test guarding the expected precedence.