i am training on Turkish captions. and after 125k iterations i enabled finetuning on Flickr30k dataset. but my validation loss increases constantly each iteration. and also my bleu score dropped after finetune. do you think of any reasons for that.
thanks in advance.
It is overfitting right now. Is your training loss getting closer to 0? How do you choose the number of iterations before your fine tuning for different dataset?
i am training on Turkish captions. and after 125k iterations i enabled finetuning on Flickr30k dataset. but my validation loss increases constantly each iteration. and also my bleu score dropped after finetune. do you think of any reasons for that. thanks in advance.