MILVLG

/

imp-v1-3b

like 201

We are working on the release of training and evaluaation code on Github. When it is done, the funetine script will be provided there. the HF repo focuses on model inference.

run Google Colab ?

Hey king, here is the piece of code you wanted :
loss = F.cross_entropy(model(images=image_tensor, input_ids=input_ids[:, :-1],
attention_mask=torch.ones_like(input_ids[:, :-1]).to(input_ids.device)).logits[:, -nb_toks_anwer:].permute((0, 2, 1)), input_ids[:, -nb_toks_anwer:])

thanks...,

Do you think this training code is correct?

https://colab.research.google.com/drive/1Rg44ZVPf3_cs77UUXmOp_eMzGLoQtml0?usp=sharing

thanks...,

Do you think this training code is correct?

https://colab.research.google.com/drive/1Rg44ZVPf3_cs77UUXmOp_eMzGLoQtml0?usp=sharing

I don't have a lot of time to check but everything is fine I think. You seem to have understood what I did and to understand what you are doing.