[Regression]: Phi-3.5-mini-instruct fails to load in-memory initializer for NvTensorRtRtx #26653
Description
When trying to run the Phi-3.5-mini-instruct model with the NvTensorRtRtx ep, below error is coming
2025-11-25 17:33:41.3973916 [E:onnxruntime:, nv_execution_provider.cc:1659 onnxruntime::NvExecutionProvider::GetSupportedList] This is an invalid model. In Node, ("/model/rotemb_caches_subgraph/sin/If", If, "", -1) : ("/model/rotemb_caches_subgraph/Greater/output_0": tensor(bool),) -> ("sin_cache": tensor(float16),) , Error Data of TensorProto ( tensor name: sin_cache_large_split) should be stored in *_ORT_MEM_ADDR_*, but it doesn't exist or is not accessible.
2025-11-25 17:33:41.4274740 [E:onnxruntime:onnxruntime-genai, inference_session.cc:2522 onnxruntime::InferenceSession::Initialize::<lambda_00ab27662768e9e277fdb72241c6ddc2>::operator ()] Exception during initialization: D:\code\anuj_gitlab\onnxruntime\onnxruntime\onnxruntime\core\providers\nv_tensorrt_rtx\nv_execution_provider.cc:1660 onnxruntime::NvExecutionProvider::GetSupportedList [ONNXRuntimeError] : 1 : FAIL : ONNX graph resolve failed: This is an invalid model. In Node, ("/model/rotemb_caches_subgraph/sin/If", If, "", -1) : ("/model/rotemb_caches_subgraph/Greater/output_0": tensor(bool),) -> ("sin_cache": tensor(float16),) , Error Data of TensorProto ( tensor name: sin_cache_large_split) should be stored in *_ORT_MEM_ADDR_*, but it doesn't exist or is not accessible.
I tried with the latest ORT main branch with TRT for RTX 1.2 windows package for here : https://developer.nvidia.com/tensorrt-rtx