mAP is always 0 and there is an error running export.py

[tian123-zxc]

Search before asking

• I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

1. I am using a custom dataset where mAP is always 0 until the end of training. I also tried lowering the version of cuda and pytorch, but it didn't work
   
   

2. After the model is trained I try to convert him to onnx model, but he got the following error, how can I modify it.
   

Additional

No response

[tian123-zxc]

My configuration is as follows：
cuda：11.6
torch：1.12.0
win11

[glenn-jocher]

@tian123-zxc your val set has no labels, i.e. see it says labels 0 in your console output. You need to fix this in your dataset.

Your train set appears correct as your train losses seem fine.