global step loaded by “load_from_check_point” is “0”
whereas “torch.load” gives the correct global step…
I’m not quite sure if it is a bug tho
global step loaded by “load_from_check_point” is “0”
whereas “torch.load” gives the correct global step…
I’m not quite sure if it is a bug tho
load_from_checkpoint
just loads the model state. global_step is a part of the trainer state.