1.pytorch out of memory model suggests cuda runtime error (2): out of memory
Solution: https://ptorch.com/news/160.html
2. In the data reading process, if the situation does not match the dimensions appear before and after the data, how to do?
Solution: (1) try to batch_size size to 1, and prints out each read file path. This scheme applies to the case DataLoader disposable entire file is read (2) the document data set into h5py, disposable read.