Open LinxinS97 opened 2 months ago
Sorry for this missing hardcoded URL; this is the standard pretrain weight for the BERT model.
You can download manually from: https://huggingface.co/google-bert/bert-base-uncased
or
change the code as: StageBertTokenizer.from_pretrained("bert-base-uncased")
.
The former is faster.
In
lavis/models/blip_models/blip_rel_det.py
I met a hardcoded path for Bert in the code, which I cannot find in your instructions. Could you provide this weight (or reference) in your instructions?