Hugging Face launched the first part of its natural language processing (NLP) course on June 15. The course will cover how to use the NLP libraries of the Hugging Face ecosystem– Transformers, Datasets, Tokenizers and Accelerate, and the Hugging Face Hub.
The course is taught by Matthew Carrigan and Lysandre Debut, Machine Learning Engineers at Hugging Face, and Research Engineer at Hugging Face, Sylvain Gugger.
The participants are expected to be fluid in Python language. Candidates should take the course after taking an introductory deep learning course. While prior knowledge of PyTorch or TensorFlow knowledge is not compulsory, some familiarity is desirable.
The course covers topics such as what transformers can do, how they work, lessons on Encoder models and Decoder models, sequence-to-sequence models, and bias and limitations. It also includes an end-of-chapter quiz to help participants assess their understanding. Additionally, the course also introduces participants to using Hugging Face Transformers, fine-tuning a pretrained model, and sharing models and tokenizers.
The first four chapters provide an introduction to the main concepts of the Hugging Face Transformers library. By the end of the course, students will be familiar with how Transformer models work, and will have a fair understanding of using a model from the Hugging Face Hub, fine-tuning it on a dataset, and sharing their results on the hub.
Chapters five to eight will teach the basics of Hugging Face Datasets and Tokensizers, before diving into classic NLP tasks. By the end of these chapters, students will be able to tackle most NLP problems by themselves.
Finally, chapters nine to 12 will dive even deeper, showcasing specialised architectures (memory efficiency, and long sequences), and teaching students how to write custom objects for exotic use cases.
The course is available for free. Check here to know more.