HuggingFace is one of the most famous names in the world of NLP. Their pretrained Deep Learning models can do almost everything related to AI and text:
They are providing that by offering a wide set of fine-tuned BERT and GPT-2 based models, thus making a need of training models from scratch obsolete.
Still, the original and the fine-tuned models came with their own limitations determined by the datasets and intents used for their training. Entity detection BERT based models, for example, provide only a strictly defined set of entities that can be detected in a given text.
Now, they are pushing the things to the upper level, promoting zero-shot learning. As described in the following post:
With that approach, at least for text classification purposes, it is not needed to use some fine-tuned BERT model to make classification against any set of arbitrarily chosen classes.
Check this out: https://huggingface.co/zero-shot/
Also, you can try the following notebook in Google Colab at:
Enjoy in the future of knowlede transfer.