I want to use the 2 x T4 GPUs on kaggle to train a segmentation model.
I have my DALI pipeline for training and validation and I’m wondering if I can use the accelerate library to put one instance of each pipeline on each GPU and also split my model across the GPUs.
Did anyone every try that? Some tutorial or tips would be extremely helpful.
P.S: I also use callbacks like Jeremy Howard showed to us in the deep learning course.