Marouan-st

Hi, I do have access to the YOLOv8n torch model from [ultralytics](https://github.com/ultralytics/ultralytics/tree/main/ultralytics). ([ultralytics doc](https://docs.ultralytics.com/usage/python/)) I tried to include the loss computation into my model and export it to onnx as...

I found the int32 tensor and changed its type to int64 and I don't get the ShapeInference error anymore, thanks. Now I get another error when trying to generate the...

> ^ Seems like a bug. I will add a pull-request to address this issue. Hello @baijumeswani, any update regarding this bug?

Ok, thank you for your support. Do you know how I could replace these ReduceMax nodes by supported operations for training? Also, there are ReduceMin nodes in the graph, is...

Hello @baijumeswani, ReduceMax and ReduceMin operations are only used in the loss computation, so the gradient is not really required for these operations. I'm following your suggested approach based on...