During training, I noticed that unset bn is performed after each training iteration. What is the purpose of this operation? Additionally, I would like to know how to decouple complex preprocessing steps from the distributed training framework. I am trying to convert the training of efficientvit to the Paddle framework, but I encountered some complex preprocessing steps. Could you provide me with some assistance and guidance in this regard? Thank you very much.
During training, I noticed that unset bn is performed after each training iteration. What is the purpose of this operation? Additionally, I would like to know how to decouple complex preprocessing steps from the distributed training framework. I am trying to convert the training of efficientvit to the Paddle framework, but I encountered some complex preprocessing steps. Could you provide me with some assistance and guidance in this regard? Thank you very much.