Trinh Vuong

Would you public "the config" file?

Hello, In your data.csv file. The noisy text and corrected text are identical. I wonder, by any chance, if you will release the noisy text without cleaning it. I really...

LLaVa-Next Video frames attention score.

1

I wonder if anyone works on how to use the: output_attentions=True, outputs["attentions"]) To visualize which frame we should pay more attention to in the sequence of 32 frames?

Will you do this for siglip the new LLaVA as well?

2

Will you do this for siglip the new LLaVA as well? Thank you for a nice work.

I can finetune without autocast or linprobe the model but got loss = nan if i finetune with autocast. Anyone have any suggestion to fix this? with torch.cuda.amp.autocast(): output =...