Niklas

> Yeah, might need more prompts where the prompt is at the beginning just incase samples are long and we end up truncating them. Since our code currently skips too...

cc @lintangsutawika @haileyschoelkopf - can't add you as reviewers somehow, but would be great if you could take a look. I'm not 100% sure about the results I got 🧐

Yeah so according to the current results evaluating the model as a causallm is better than a prefixlm after it was fine-tuned as a prefixlm. Also note: - In both...

> Just saw this! Unfortunately, I’ll be away from my computer today and tomorrow. > > Also just a note: we should not merge this into the `eval-hackathon` branch, >...

Hey maybe the below helps: ``` sacremoses==0.0.43 pandas==1.1.3 regex==2020.4.4 h5py==2.10.0 filelock==3.0.10 scipy==1.4.1 sentencepiece~=0.1.91 matplotlib==3.2.1 torch==1.6.0 tensorflow==2.3.1 tqdm==4.45.0 numpy==1.18.1 six==1.14.0 packaging==20.1 wandb==0.10.8 psutil==5.7.0 requests==2.23.0 pytorch_lightning==1.0.4 ImageHash==4.1.0 tokenizers~=0.9.2 transformers==3.5.1 # Required due...

If you want to run the models via Data Parallelism you will need to wrap them in `torch.nn.DataParallel` - In its current state the `args.multiGPU `does not work

Yeah there is some preprocessing still happening in `entryU` - Maybe try instead wrapping `self.model, loading_info = BertU.from_pretrained(self.tr_name, img_dim=2048, output_loading_info=True)` the self.model inside `entryU` with `torch.nn.DataParallel`?

Hi, you can find a list of multi-modal models implemented in this codebase [here](https://github.com/Muennighoff/vilio#architectures)

@darthgera123 sorry for the late reply! This is most likely a shape mismatch - What are the dimensions of your images / are you using the HM Dataset?

Already merged via other PR