Join this intensive 3-day developer course to gain hands-on experience with Large Language Models using the Transformers library and Hugging Face. You'll explore how transformer-based models work in Natural Language Processing applications, discover their practical capabilities and limitations, and learn to fine-tune pre-trained models for your specific use cases. Through practical exercises, you'll master the Trainer API and Keras for model optimization, share your work on the Hugging Face Hub, create custom datasets, and implement semantic search using FAISS and the Datasets library.
Prerequisites: You should have completed our AI Workbench course or possess basic knowledge of programming concepts and syntax in languages such as Python or JavaScript. General familiarity with APIs is also recommended.
Course Topics
Understanding Transformer Models
- Natural Language Processing foundations and applications
- Transformer architecture: capabilities and what they excel at
- How transformers learn and process language
- Encoder models for understanding tasks
- Decoder models for generation tasks
- Sequence-to-sequence models for translation and conversion
- Bias considerations and model limitations
Working with Transformers
- Understanding the pipeline abstraction
- Model selection and loading
- Tokenizers and text preprocessing
- Managing multiple input sequences
- Combining components into complete workflows
Fine-Tuning Pre-Trained Models
- Preparing and preprocessing training data
- Fine-tuning with the Trainer API and Keras frameworks
- Running complete training workflows
- Evaluating fine-tuned model performance
Publishing Models and Tokenizers
- The Hugging Face Hub: overview and benefits
- Downloading and using pre-trained models
- Sharing your own models with the community
- Creating comprehensive model cards
- End-of-section assessment
Working with the Datasets Library
- Importing datasets not yet available on the Hub
- Slicing and filtering data efficiently
- Processing large datasets at scale
- Building custom datasets for your projects
- Semantic search implementation using FAISS
