Model quality
#3
by
raminh921
- opened
Hi,
I tested some images in the test set of the oddadmix/Khanandeh-0.1-Persian-dataset, and the generated responses were almost correct. However, the responses for hezarai/parsynth-ocr-200k were not correct for the five images I tested, Of course I understand the model is not trained on that dataset.
- Does it generalize simply by having a larger and more generalized dataset, or are other techniques required?