yolov5 family models conversion failed..

[wxthu]

I wanna accelerate yolov5 inference using tensorrt, However, when I convert pytorch format to tensorrt format, I met the following error. It seems that conversion failed . Could you give me some suggestions?

my code:

# conversion
  model = torch.hub.load('ultralytics/yolov5', 'yolov5s').eval().cuda()
  im = torch.rand(1, 3, 640, 640).cuda()
  model_trt = torch2trt(model, [im])
  
  # execution
  for _ in range(20):
      y = model_trt(im)

  times = 500
  start = time_sync()
  for _ in range(times):
      y = model_trt(im)
  end = time_sync()
  print("Average latency after accis {} ms".format((end - start) * 1000 / times))`
```

tensorrt version: 7.0.0.11
pytorch: 1.8.1
cuda: 10.2
python: 3.7.16

[wxthu]

@koenvandesande

[wxthu]

I found that when I change the batch size from 1 to other value , then conversion and inference could be finished successfully . I am confused .... @chitoku @koenvandesande @oliver-batchelor