NMS works too slow!

[TolgaOzdmir]

I have trained the YOLOv8 model and converted it to onnx model with easydeploy. The problem is when I looked at the output of the ONNX Runtime Profiler, NMS took most of the time for each inference. Is it a normal thing or what can I do to improve the inference speed of the NMS algorithm or is that a bug?

onnxruntime_profile_model.json

Versions:
CUDA 11.8

mmcv 2.0.1
mmdet 3.2.0
mmengine 0.9.1
mmyolo 0.6.0
onnx 1.14.1
onnx-simplifier 0.4.35
onnxruntime-gpu 1.16.2