When the model misreads a word, how many sentences containing the correct word are needed to use for finetune to correct it?
The problem that occurs is that after fine tuning, it adds extra things to the end of sounds and sentences, if penalty is not used, why?
When the model misreads a word, how many sentences containing the correct word are needed to use for finetune to correct it? The problem that occurs is that after fine tuning, it adds extra things to the end of sounds and sentences, if penalty is not used, why?