使用paddle2onnx转换ppyolo_tiny后，存在不支持在gpu上的算子，请问该如何解决？

[gzyao]

模型：ppyolo_tiny_650e_coco.pdparams
命令：

1. python tools/export_model.py
   -c configs/ppyolo/ppyolo_tiny_650e_coco.yml
   -o weights=ppyolo_tiny_650e_coco.pdparams
2. paddle2onnx --model_dir=output_inference/ppyolo_tiny_650e_coco/
   --model_filename=model.pdmodel
   --params_filename=model.pdiparams
   --save_file=ppyolo_tiny.onnx
   测试：
   在cpu上：
   YOLOv8n FPS: 143.21
   PP-YOLO FPS: 185.80
   在gpu上：
   YOLOv8n FPS: 240.72
   PP-YOLO FPS: 93.47
   警告：2025-08-01 09:54:56.701804923 [W:onnxruntime:, transformer_memcpy.cc:74 ApplyImpl] 6 Memcpy nodes are added to the graph PaddlePaddle Graph in PIR mode for CUDAExecutionProvider. It might have negative impact on performance (including unable to run CUDA graph). Set session_options.log_severity_level=1 to see the detail logs before this message.
   2025-08-01 09:54:56.703636717 [W:onnxruntime:, session_state.cc:1166 VerifyEachNodeIsAssignedToAnEp] Some nodes were not assigned to the preferred execution providers which may or may not have an negative impact on performance. e.g. ORT explicitly assigns shape related ops to CPU to improve perf.
   2025-08-01 09:54:56.703655513 [W:onnxruntime:, session_state.cc:1168 VerifyEachNodeIsAssignedToAnEp] Rerunning with verbose output on a non-minimal build will show node assignments.

[github-actions]

This issue is stale because it has been open for 30 days with no activity.