cant compute score

[laolida-w]

I’m encountering an issue with the following line in fleur.py:
outputs = tokenizer.decode(output_ids[0, input_ids.shape[1]:]).strip()
In this case, input_ids has the shape torch.Size([1, 400]) and output_ids has the shape torch.Size([1, 7]). The slicing output_ids[0, input_ids.shape[1]:] results in an out-of-bounds operation, which causes outputs to be empty.

[Yebin46]

Sorry for the late reply. Thank you for showing interest in our work.

The output format may differ if the LLaVA commit version is different.
As I mentioned in my response to #5, could you check if the LLaVA submodule commit ID is ab58dc1?

If it still doesn’t work, please let me know again.