Is there a proper way to import models from the Nvidia site to be used in Hugging Face? As of this moment there are several models from Nvidia that are in HF but the ones I am interested are not there. Basically the downloaded Nvidia models typically contain a vocab.txt, config.json and a .pt file. I tried a combination of MegatronBertConfig/BertConfig, AutoTokenizer and AutoModel/MegatronBertModel with no luck. Any general guidelines would be appreciated. Thank you.
1 Like
I have a similar question / problem. No model in HF hub for the variant I’m interested in. Wondering if there examples of importing models from the NVIDIA website.