Hi.
I am trying to finetune my pretrained model on a custom classification dataset. I tried running it on just 30 examples for 500 epochs. However, I noticed that the loss doesn't converge at all. But when I set init_checkpoint(this line) to None , the loss did converge. I don't know if I missed anything. Thanks in advance.
With init_checkpoint set to my pretrained directory
Hi. I am trying to finetune my pretrained model on a custom classification dataset. I tried running it on just 30 examples for 500 epochs. However, I noticed that the loss doesn't converge at all. But when I set
init_checkpoint
(this line) to None , the loss did converge. I don't know if I missed anything. Thanks in advance.With
init_checkpoint
set to my pretrained directoryWith
init_checkpoint
set to None