Working with Hugging Face

Via DataCamp

DataCamp based on 616 ratings

Likes

Overview

This course focuses on using the Hugging Face ecosystem, which is widely regarded as one of the top platforms for working with transformer models and natural language processing (NLP) tasks. Learners will explore Hugging Face's `transformers` library to work with pre-trained models like BERT, GPT, and T5 for tasks such as text classification, summarization, and sentiment analysis. Students will also learn how to fine-tune models, deploy them for inference, and utilize the Hugging Face Hub for sharing models and datasets....

Syllabus

Getting Started with Hugging Face

Start your journey with the Hugging Face platform by understanding what Hugging Face is and common use cases. Then, you'll learn about the Hugging Face Hub including models and datasets available, how to search for them, navigate model, or dataset, cards, and download. Lastly, you'll learn about the high-level components of transformers and LLMs.

Building Pipelines with Hugging Face

It's time to dive into the Hugging Face ecosystem! You'll start by learning the basics of the pipeline module and Auto classes from the transformers library. Then, you'll learn at a high level what natural language processing and tokenization is. Finally, you'll start using the pipeline module for several text-based tasks, including text classification.

Building Pipelines for Image and Audio

In this chapter, you'll apply pipeline methodologies to new tasks using image and audio data. Specifically, you will learn ways to process these types of data in preparation for tasks such as classification, question and answering and automatic speech recognition.

Fine-tuning and Embeddings

Explore the different frameworks for fine-tuning, text generation, and embeddings. Start with the basics of fine-tuning a pre-trained model on a specific dataset and task to improve performance. Then, use Auto classes to generate the text from prompts and images. Finally, you will explore how to generate and use embeddings.