CS 781: Large Language Models (LLMs)

Units: 3-0-0-0-9 (L-T-P-C)

 

Pre-requisites: Instructor’s consent and

Must: Statistical Natural Language Processing (CS779), Proficiency in Linear Algebra, Probability and Statistics, Proficiency in Python Programming

Desirable: Introduction to Machine Learning (CS771) or equivalent course, Deep Reinforcement Learn- ing (CS780), Probabilistic Machine Learning (CS772)

 

Level of the course: Ph.D., PG, and 3rd, 4th year UG Students (7xx level)

 

Course Objectives

In recent times, Large Language Models (LLMs) have revolutionized the field of Natural Language Process- ing (NLP). However, the application of LLMs has not just remain limited to NLP but has also advanced  other areas like Biology, Chemistry, Economics, etc. This calls for in-depth understanding of LLMs. This course will introduce the fundamentals of LLMs and go in-depth into various techniques to develop LLMs, scaling laws. It will cover various LLM architectures. It will teach how to fine-tune LLMs using parameter efficient techniques, how LLMs could be used in conjunction with external knowledge sources such as vector databases. We will have a more mathematical and rigorous approach towards understanding LLMs.

 

Course Contents (total 40 lectures):

1. Classical Language Modeling (CLM) [3 lectures]: n-grams, smoothing, class-based, brown clustering, etc.
2. Neural Language Modeling (NLM) [4 lectures]: Word Embeddings, Word2Vec, FeedForward Neural LM, Contextualization, Sub-tokenization and Subword information, etc.
3. Transformers for Language Modeling [3 lectures]: Encoder Models, Encoder-Decoder Model, Decoder Models, Pre-trained LMs (PLMs), objective functions for training, etc.
4. Introduction to Large Language Models (LLMs) [3 lectures]: PLMs vs LLMs, LLM families
5. Scaling Laws [3 lectures]: Kaplan’s law, Chinchilla Law
6. Training LLMs from Scratch [3 lectures]: Selecting the corpus, cleaning and pre-processing, deciding hyper-parameters using scaling laws, training, etc.
7. Providing Human Feedback [3 lectures]: RLHF, DPO, etc.
8. Emergent Properties in LLMs [4 lectures]: Prompting techniques (zero shot, few shot, etc.), Chain of Thought, Tree of Thought, X of Thought, etc.
9. Parameter Efficient Fine-Tuning (PEFT) [5 lectures]: Transfer Learning, Soft-Prompting, Adaptors, LoRA (and variants).
10. Using LLMs with Vector Databases [4 lectures]: Retireval Augmented Generation and related techniques.
11. Understanding LLM inner workings [5 lectures]: Mechanistic Interpretability

 

References:

Since this is new and emerging area, there are no specific references, this course gleans information from a variety of sources like research papers, tutorials, blogs, etc. Relevant references would be suggested in the lectures.