Make sure TRT EPs can loads models when initializers in memory

[yuslepukhin]

This PR moves the conversion of initializers in-memory from Graph constructor to early in graph transform before the partitioning. This is done to avoid conversion when subgraphs are constructed.

It also addresses bugs in TRT and NV TRT providers.

Addresses issue: #26653

Graph Initializer Conversion and Handling:

• Added a new method Graph::ConvertInitializersIntoOrtValues() to convert all graph TensorProto initializers into OrtValues and create in-memory external data references, separating this logic from graph construction and making it reusable. (include/onnxruntime/core/graph/graph.h, onnxruntime/core/graph/graph.cc) [1] [2]
• Removed the previous lambda for converting large tensor initializers within the graph constructor, delegating this responsibility to the new method above for clearer separation of concerns. (onnxruntime/core/graph/graph.cc) [1] [2] [3]

Provider Interface Enhancements:

• Introduced move assignment operators for GraphProto and TensorProto in both the provider interface (ProviderHost) and wrapper structs, allowing for more efficient object transfers and assignment. (onnxruntime/core/providers/shared_library/provider_interfaces.h, onnxruntime/core/providers/shared_library/provider_wrappedtypes.h) [1] [2] [3] [4]
• Added iterator interfaces (TensorProto_ConstIterator, TensorProto_Iterator) and corresponding methods to TensorProtos for clean iteration over initializer lists, improving code readability and maintainability. (onnxruntime/core/providers/shared_library/provider_interfaces.h, onnxruntime/core/providers/shared_library/provider_wrappedtypes.h) [1] [2] [3]

Execution Provider Logic Simplification:

• Refactored how initializers are processed in the NVExecutionProvider, using the new initializer conversion and iteration logic to simplify handling of external and in-memory data, and ensuring correct assignment and ownership of user-provided weights. (onnxruntime/core/providers/nv_tensorrt_rtx/nv_execution_provider.cc) [1] [2] [3]

Other Minor Improvements:

• Improved const-correctness and interface consistency for size and iterator methods in TensorProtos. (onnxruntime/core/providers/shared_library/provider_interfaces.h, onnxruntime/core/providers/shared_library/provider_wrappedtypes.h) [1] [2]

[Copilot]

Pull request overview

This PR refactors the initialization of graph initializers by moving the conversion of TensorProto initializers to OrtValues from the Graph constructor to an explicit call during graph transformation (before partitioning). This change ensures that execution providers can work with models that have initializers in memory, addressing issue #26653.

Key Changes:

• Added Graph::ConvertInitializersIntoOrtValues() method to explicitly convert large initializers to OrtValues with in-memory external data references
• Enhanced provider interfaces with move assignment operators and iterator support for TensorProtos
• Refactored TensorRT and NV TensorRT providers to handle initializers more uniformly using the new interfaces

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
include/onnxruntime/core/graph/graph.h	Added declaration for ConvertInitializersIntoOrtValues() method
onnxruntime/core/graph/graph.cc	Implemented new conversion method and removed old lambda-based conversion from constructor
onnxruntime/core/session/inference_session.cc	Added call to convert initializers before partitioning and improved exception handling
onnxruntime/core/session/provider_bridge_ort.cc	Implemented iterator interfaces and move assignment operators for provider bridge
onnxruntime/core/providers/shared_library/provider_interfaces.h	Added TensorProto iterator interfaces and updated method signatures for const-correctness
onnxruntime/core/providers/shared_library/provider_wrappedtypes.h	Updated wrapper types with move semantics and iterator support
onnxruntime/core/providers/tensorrt/tensorrt_execution_provider.cc	Refactored initializer handling to use new iterator-based approach and simplified logic
onnxruntime/core/providers/nv_tensorrt_rtx/nv_execution_provider.cc	Attempted refactoring of initializer handling with critical bugs in variable references
onnxruntime/test/ir/graph_test.cc	Updated test to validate explicit conversion behavior
onnxruntime/test/ir/utils_test.cc	Minor refactoring to use ASSERT_STATUS_OK macro

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.