ONNX Runtime 发布: ONNX Runtime CUDA Plugin EP 0.1.0

来源摘要

This is the first release of ONNX Runtime CUDA Plugin EP, providing CUDA execution as a separately packaged plugin execution provider. These notes cover commits affecting CUDA Plugin EP core code, CMake integration, and its primary build and package pipeline. Please refer to [QUICK_START.md](https://github.com/microsoft/onnxruntime/blob/main/docs/cuda_plugin_ep/QUICK_START.md) for the usage. ## Highlights ### Plugin Runtime - Introduces the CUDA Plugin EP core and makes it the default CUDA provider implementation ([#27816](https://github.com/microsoft/onnxruntime/pull/27816), [#29544](https://github.com/microsoft/onnxruntime/pull/29544)). - Adds arena allocation, resource accounting, available-resource reporting, and IOBinding synchronization ([#27931](https://github.com/microsoft/onnxruntime/pull/27931), [#28028](https://github.com/microsoft/onnxruntime/pull/28028), [#28103](https://github.com/microsoft/onnxruntime/pull/28103), [#27919](https://github.com/microsoft/onnxruntime/pull/27919)). - Adds provider options for user compute streams, copy behavior, EP-level unified streams, and external allocators, with compatibility back to ONNX Runtime 1.24.4 through version-gated callbacks ([#28603](https://github.com/microsoft/onnxruntime/pull/28603), [#28824](https://github.com/microsoft/onnxruntime/pull/28824)). ### CUDA Graphs and Profiling - Adds CUDA Graph capture and replay, including user compute stream support, synchronization controls, and a kernel sync stream for scratch allocation ([#27958](https://github.com/microsoft/onnxruntime/pull/27958), [#28002](https://github.com/microsoft/onnxruntime/pull/28002), [#28686](https://github.com/microsoft/onnxruntime/pull/28686), [#29221](https://github.com/microsoft/onnxruntime/pull/29221), [#29244](https://github.com/microsof