Some useful scripts for InternVL.
I think this can be a handy tool for people who want to do a bit of prompt tuning for InternVL. I think I will build some more useful things for that model family. The example folder contains an example of a simple German invoice. The following image is the thumbnail attention map for that invoice.
If you want to finetune an InternVL2 model yourself (if you really have that much compute you are a blessed rich boy!), you could use the scripts provided in the datasets folder. These should turn the Huggingface dataset into the right format for finetuning.
For more info on how to finetune InterVL2, check out their website.