feat: Enhanced LoRA transforms with flexible path mapping and Pydantic model support

[zhaozuy]

Summary

Introduces enhanced LoRA adapter transforms that address architectural limitations in the current implementation while maintaining full backward compatibility.

Motivation

The current LoRA transform implementation has several limitations:

• Missing support for engine request Pydantic model validation
• Complex inheritance hierarchy that's hard to extend
• Limited flexibility in request path mapping

Changes Made

🆕 Enhanced Transform Architecture:

• BaseLoRAApiTransform: Clean ABC-based architecture with separation of concerns
• LoadLoraApiTransform & UnloadLoraApiTransform: Modular, specialized transform classes
• lora_factory.py: Enhanced decorator factory with flexible configuration
• Comprehensive debug logging throughout transformation pipeline

Enhanced Public API:

• Added new parameters to register_load_adapter_handler() and register_unload_adapter_handler()
• engine_request_lora_name_path & engine_request_lora_src_path: Flexible path mapping
• engine_request_model_cls: Pydantic model support for request validation
• engine_request_defaults: Automatic default value injection

Infrastructure Improvements:

• Enhanced set_value() utility with better typing and optional depth limits

API Examples

Old API (still supported):

@register_load_adapter_handler(
    request_shape={"adapter_name": "body.name", "adapter_path": "body.src"}
)

New API:

@register_load_adapter_handler(
    engine_request_lora_name_path="body.adapter_name",
    engine_request_lora_src_path="body.source_path",
    engine_request_model_cls=MyEngineRequestModel,
    engine_request_defaults={"config.rank": 4, "config.alpha": 16}
)

Backward Compatibility

• Existing request_shape API continues to work unchanged
• Deprecation warnings guide users toward new API
• Gradual migration path with no breaking changes

Testing

• Core implementation complete and functional
• Integration tests are work in progress

Files Changed

• python/model_hosting_container_standards/sagemaker/lora2/ - New enhanced transforms
• python/model_hosting_container_standards/sagemaker/__init__.py - Enhanced decorators
• python/model_hosting_container_standards/common/transforms/utils.py - Utility improvements

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.