NMS time limit in val.py

### Search before asking - [x] I checked the docs, issues, and discussions and could not find an answer. ### Question When I run val.py with the trained weights, I encounter the NMS time limit. When I try to reduce the batch size, the NMS warning decreases a bit, but the detection results seem to improve in terms of metrics. However, if I want to conduct a comparative experiment, shouldn't the batch size be kept the same? But I also suspect that the appearance of the NMS warning has led to a decline in performance. So I don't know how to make a choice. ### Additional _No response_