manually load a checkpoint

[foldl]

I am investigating how to load state_dict from a checkpoint file. For a Dense module with bias = False, I can see something like this:

hidden0:
     kernel:
         v_row:
             tensor (1024,)
         v_col:
             tensor (4096,)
         v:
             tensor (1,)
         m:
             tensor (1024, 4096)

m shall be the matrix itself. But how about other ones?

[foldl]

These are tensors generated in training stage.

[8bitmp3]

Hi @foldl , just looping in @IvyZX