[992X]

Description of changes:
list_base_models() and base-model resolution now list a Public Hub model only when its HubContentSearchKeywords carry @capability:training_session.

#6506 added a per-Region static list (_TRAINABLE_MODELS_BY_REGION) as a stand-in until the Public Hub published that keyword. The Hub now publishes it in every Region that serves Training Sessions, so this PR removes the list and its fallback in _is_trainable. A model enabled later shows up with no SDK change.

Testing done:

• sagemaker-train/tests/unit/train/training_session: 826 passed. The three static-list tests are replaced by one test: a model without the keyword is not listed.
• Ran list_base_models from this branch against the live Public Hub:

us-west-2  ['google/gemma-4-31b-it', 'google/gemma-4-E2B-it', 'openai/gpt-oss-20b', 'Qwen/Qwen3.6-27B', 'Qwen/Qwen3.8-27B']
us-east-1  ['openai/gpt-oss-20b']
ap-south-1 ['google/gemma-4-31b-it', 'openai/gpt-oss-20b', 'Qwen/Qwen3.6-27B']

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

[github-actions]

🤖 AI code review

This PR removes the transitional _TRAINABLE_MODELS_BY_REGION static allow-list and its fallback in _is_trainable, now that the Public Hub publishes the @capability:training_session keyword in all serving Regions. This is exactly the follow-up that PR #6506's guidance called for ("Remove _TRAINABLE_MODELS_BY_REGION once the Public Hub publishes @capability:training_session keyword", Source: #6506). The _is_trainable signature is simplified, all call sites and the now-unused region local and FrozenSet import are cleaned up, and the three static-list tests are replaced by a negative test. The change is clean and I found no issues.

_{Reviewed commit cc15022. Automated review; verify before acting.}