我用onnx转成了一个om模型，并用如下命令测试推理速度：
HwHiAiUser@Euler:~/Deploy/yolov8_obb$ python3 -m ais_bench --model=v8_obb-512-640.om --loop=1000 --batchsize=1 

[INFO] acl init success 

[INFO] open device 0 success 

[INFO] load model v8_obb-512-640.om success 

[INFO] create model description success 

[INFO] warm up 1 done 

Inference array Processing:   0%|                                                   | 0/1 [00:00<?, ?it/s] 

loop inference exec: (1000/1000)ython3 -m ais_bench --model=v8_obb-512-640.om --loop=1000 --batchsize=1 

Inference array Processing: 100%|███████████████████████████████████████████| 1/1 [00:09<00:00,  9.19s/it] 

[INFO] -----------------Performance Summary------------------ 

[INFO] NPU_compute_time (ms): min = 8.619999885559082, max = 9.017999649047852, mean = 8.686402983665467, median = 8.675999641418457, percentile(99%) = 8.91700982093811 

[INFO] throughput 1000*batchsize.mean(1)/NPU_compute_time.mean(8.686402983665467): 115.1224507866457

但在模型推理中，以下代码却耗时25ms左右，请问是什么导致了这个差异？

output = model.execute([img_pre])