How only one kanji per image?
#1
by j35t3r - opened
What did you do that there is only one kanji? usually there 20 kanjis when just asking for one.
if your model is trained on one kanji then it would only output one kanji
I did. I just put train batch size = 1. No change.
Then, the dataset must be curated. I have image <---> text pair. I used lora fintetuning.
The text contains several meanings... in my case. I think I change to one.
I'm not sure I exactly understand what you mean.
Batch size refers to how many training examples are processed together in one pass of the model.
Kanjis can have several meanings, that's to be expected.