HTTPError: 401 Client Error: Unauthorized for url: https://huggingface.co/downloaded/roberta-base/resolve/main/vocab.json

[1112321sfdsaf]

Hey, Thanks very much for the excellent work and repo. When I run 'bash run_scripts/pretrain_m3ae.sh', I got the above exception. I tried to click this link and found that the link is unavailable. So, I replaced ' tokenizer=downloaded/roberta-base' in 'pretrain_m3ae.sh' with ' tokenizer=roberta-base', and got a successful run. I would like to know if this operation is allowed.

[zhjohnchan]

Hi there,

Thanks for your attention. The operation is allowed since the downloaded RoBERTa model is the same one.

Best, Zhihong

[1112321sfdsaf]

Thank you very much for your reply.

[1112321sfdsaf]

Hi again, I found that the default batch size per GPU is 32, and my own configuration seems far from that requirement, can you provide how much GPU memory is needed for 32 batches?

[zhjohnchan]

Hi,

I used A100-80G in my experiments.

Best, Zhihong

[1112321sfdsaf]

Thanks for your reply again.

[1112321sfdsaf]

Hi, when preparing the MELINDA dataset using the link introduced in the paper "MELINDA: A Multimodal Dataset for Biomedical Experiment Method Classification", I got a 404 on the page, which means the link does not exist. So, how can I download this dataset to follow your excellent work?

[zhjohnchan]

Hi,

I request the dataset from the authors. I think you can send an e-mail to the author.

Best, Zhihong

[1112321sfdsaf]

Thanks for your suggestion.

[1112321sfdsaf]

Hi, I'm back. 😅

When I download VQA-RAD on the official page, I found that the files are different from the description of this repo. Moreover, reading the dataset paper, it seems just to divide the dataset into a training set and a test set. So, can you provide more details on the data splits including training, validation, and test sets?

Thanks in advance.