Closed jbdel closed 2 years ago
Hi,
Thanks for your question! The mechanism for interfacing between Huggingface transformers and AllenNLP is pretty complicated. I think your best bet is probably just post an un-trained two-layer transformer to Huggingface and point to that model in the config. The alternative of digging around in the guts of AllenNLP doesn't seem very attractive.
its actually a smart work-around, thanks!
JB
Hello,
Thank you for this amazing repo!
I understand
bert_model
is the name of a pretrained BERT model available on HuggingFace.For an ablation study, I would like to train a simple BERT model, from scratch, with a custom config (like, say, 2 layers). Is there a way to make that work? Thank you very much for your reply.