Matthias

@AyushExel This came up because of another issue - I wanted to retrieve the metrics from the final validation step (using the best.pt model) in my Python code. However, the...

@Laughing-q @AyushExel Thanks, but now with this change I get the following error from the `train()` call: ```python Traceback (most recent call last): File "/__w/head/head/src/dvc/yolov8-trainer.py", line 86, in main() File...

It's strange indeed, because locally on my dev machine this works. But in the `ultralytics` Docker container, even after running a `git -C /usr/src/ultralytics pull`, the call to `train()` leads...

@AyushExel So I just checked within your `ultralytics` Docker image - the new line `self.metrics_data = self.trainer.validator.metrics` is there in the `train()` method as it should be. According to the...

@AyushExel Thanks, yes I was training on four GPUs. So if I pull now it should work?

@AyushExel Great, thanks. So does this work for multi-GPU training as well, or would I not have access to these metrics then?

@AyushExel Unfortunately no, the issue is still the same, even with **Ultralytics YOLOv8.0.37**: ```python Traceback (most recent call last): File "/__w/head/head/src/dvc/yolov8-trainer.py", line 86, in main() File "/__w/head/head/src/dvc/yolov8-trainer.py", line 36, in...

@AyushExel Update: Running `train()` using only one GPU (but with the same Docker image on the same machine) does work. So it has something to do with multi-GPU training as...

Well, I am not sure whether this is really an edge case - I would assume I am not the only one who wants to get the final validation metrics...

Thanks for your update, I will manually validate the model then. When will you push the release to main?