What causes the model to only output "<|im_end|>};" on some pages?

[plamb-viso]

I'm loading the model as specified in the demo script and running inference highly similarly:

res = model.chat(tokenizer, str(image_path), ocr_type='format')

On some page images, the only output i get back from the model is:

<|im_end|>};

These images tend to be pages where the entire page is a table and each cell can contain a fair bit of text.

What causes this?

[rahulvalyu]

got the same issue. did you solve it?

[plamb-viso]

I have not, but the model is currently being added to transformers and I'm going to try it again once it's in those APIs