Super Data Manipulator: I am still groping at this stage. I can't give too much advice. I can only give a little experience summarized so far: try to expand the data and see how to deal with it faster and better. Faster-How should distributed mechanisms be trained? Model Parallelism or Data Parallelism? How to reduce the network delay and IO time between machines between multiple machines and multiple cards is a problem to be considered. Better-how to ensure that the loss of accuracy is minimized while increasing the speed? How to change can improve the accuracy and MAP of the model is also worth thinking about.
< div id= "showcount" >
