Hi, I am training a RF-DETR model, but its taking forever to train it took it 20 hours to do just 72 epochs, compared to YOLOV12 accurate which did 300 in 15 hours. Could you stop this training without losing the progress. I’m just trying to experiment on my own for an application in our company, so I don’t want to buy more credits before pitching it to my boss. The link to the model is this Sign in to Roboflow , please stop the training. Thank you very much, also it would be good if I wouldn’t get locked out of the account for going over my credits, I thought both of these would only take 8 hours.
Hey @Honza_Ferenc
Thanks for trying RF-DETR. It’s a pretty new model and so we’re currently making improvements to make that experience better. We’ve just released updates for early stopping and better estimates of training times (and, therefore, credit costs) and are actively working to make the training process itself faster.
Training Times
Compared to the YOLOv12 and other architectures we support, RF-DETR is a larger model. It takes longer to train, but almost always results in better performance (which seems to be the case for your trainings as well). Additionally, it takes longer per epoch but takes fewer epochs to converge.
Credits & Pricing
As for credits and pricing, it looks like you’re on a Growth Trial. While you’re on a trial, you won’t get charged for “Flex Usage” (overages), so you don’t have to worry about that.
If you have any feedback or issues, please let us know and we’ll look into it
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.