This should be investigated. I suspect that some operations are being done in CPU which slows the training and underutilises the GPU.
If my suspicion is correct, identifying the operations that slow down the training on the model and making sure they're run on GPU will speed up the training.
This should be investigated. I suspect that some operations are being done in CPU which slows the training and underutilises the GPU.
If my suspicion is correct, identifying the operations that slow down the training on the model and making sure they're run on GPU will speed up the training.