Introducing Hugging Face Transformers

So far, we have discussed in depth the architecture and training properties of LLMs. But the sad truth is that these models are so large it is unlikely that you or I would build one from scratch. Instead, we’ll probably use a pre-trained model. In this section, we’ll see how to do that with the Hugging Face Transformers library (https://github.com/huggingface/transformers). As the name suggests, its focus is the transformer architecture. It supports three different backends—PyTorch, TensorFlow, and JAX (as usual, we’ll focus on PyTorch). It is open source and available for commercial use. The company behind it, Hugging Face, also develops the Hugging Face Hub—a complementary service to the library cloud-based platform. It supports hosting and/or running Git repositories (such as GitHub), transformer models, datasets, and web applications (intended for proof-of-concept (POC) demos of ML applications). With that...