Advanced Techniques and Interpretability in LLMs

Packt via Coursera

Go to class Write review

Details

Go to class

Provider

Coursera
Pricing

Paid Course
Languages

English
Certificate

Certificate Available
Effort

9 weeks, 1 hour a week
Sessions

Self-Paced
Level

Intermediate
Subtitles

English

Overview

Google, IBM & Meta Certificates – 40% Off

One plan covers every Professional Certificate on Coursera.

Unlock All Certificates

Delve into advanced transformer techniques, including generative AI, fine-tuning, interpretability, and the critical role of tokenization. Learn how to leverage and interpret large language models for a variety of sophisticated NLP tasks. This course covers the next level of transformer applications, focusing on generative AI with models like ChatGPT, advanced fine-tuning strategies, and the interpretability of model outputs. Learners will explore how tokenization shapes model performance, how embeddings can be used for search and transfer learning, and how to apply transformers to tasks such as semantic role labeling and summarization. By completing this course, you will be able to implement, fine-tune, and interpret large language models for complex NLP challenges. The course combines in-depth conceptual discussions with practical demonstrations, guiding learners through the intricacies of advanced transformer techniques and interpretability tools. Each topic is presented with clarity to ensure learners can apply these methods confidently. This course is part two of a three-course Specialization designed to build a complete and cohesive understanding of the subject. While it offers valuable skills on its own, you'll gain the most benefit by progressing through all three courses as a structured learning journey. This course is based on Transformers for Natural Language Processing and Computer Vision, by Denis Rothman. Packt is one of the world's most prolific publishers of cutting-edge technical content. For over two decades we've made it our mission to curate and publish the knowledge of only the very best technical experts. We focus on real-world courses that help our customers get the job done, with coverage that extends across a wide range of established and cutting-edge technical topics. If you're an individual or an organisation that embraces learning by doing, Packt is the perfect fit for you.

Syllabus

The Generative AI Revolution with ChatGPT

This module introduces the architecture and real-world applications of GPT models, highlighting their transformative impact on society and software development. Learners will explore key concepts such as context size, decoder layers, and the use of GPT-4 as an assistant, including hands-on experience with the GPT-4 API and Retrieval Augmented Generation (RAG). By the end, participants will understand how generative AI is revolutionizing productivity and innovation across domains.

Fine-Tuning OpenAI GPT Models

This module guides learners through the process of fine-tuning OpenAI GPT models, including preparing datasets in JSONL format and executing completion tasks with custom models. Learners will gain practical skills to enhance model accuracy for specific applications.

Shattering the Black Box with Interpretable Tools

This module introduces a range of interpretability tools for transformer models, including BertViz, SHAP, LIME, LIT, and OpenAI's GPT-4 explainer. Learners will gain hands-on experience visualizing attention mechanisms, interpreting model outputs, and understanding the internal workings of large language models. By the end, you'll be equipped to make sense of complex AI systems and enhance their transparency.

Investigating the Role of Tokenizers in Shaping Transformer Models

This module delves into the critical role of tokenizers in transformer-based language models, examining various tokenization techniques such as subword, regular expression, and SentencePiece. Learners will explore how tokenization quality impacts model performance and discover strategies for handling out-of-vocabulary words and token-ID mapping.

Leveraging LLM Embeddings as an Alternative to Fine-Tuning

This module introduces the use of large language model (LLM) embeddings as a practical alternative to fine-tuning for tasks such as retrieval-augmented generation (RAG) and question-answering. Learners will gain hands-on experience with embedding-based search, clustering techniques using Ada embeddings, and evaluating model responses without a dedicated knowledge base. The module emphasizes practical implementation and analysis of embedding-driven workflows.

Toward Syntax-Free Semantic Role Labeling with ChatGPT and GPT-4

This module introduces syntax-free approaches to semantic role labeling (SRL) using advanced transformer models like ChatGPT and GPT-4. Learners will explore how these models handle complex sentence structures without relying on traditional syntactic analysis, and examine the challenges and strategies for standardizing input formats in NLP tasks.

Summarization with T5 and ChatGPT

This module guides learners through the architectures and practical applications of T5 and ChatGPT for text summarization, with a focus on legal and financial domains. You will compare model configurations, implement summarization functions, and evaluate the strengths of each approach for different project needs.

Exploring Cutting-Edge LLMs with Vertex AI and PaLM 2

This module introduces learners to the architecture and applications of Google's PaLM 2 and Vertex AI, highlighting their roles in modern natural language processing tasks. Learners will explore advanced activation functions, interface navigation, and practical use cases such as sentiment analysis and code generation. By the end, participants will gain hands-on experience leveraging these tools for real-world AI solutions.

Guarding the Giants: Mitigating Risks in Large Language Models

This module examines the risks associated with large language models, including ethical concerns, memorization, influence operations, harmful content, and cybersecurity vulnerabilities. Learners will explore practical tools and strategies for risk mitigation, such as RAG, RLHF, and advanced tracking systems. By the end, participants will understand how to responsibly deploy and monitor LLMs in real-world scenarios.