Course

Syllabus

Background on NLP and tasks: Tokenization – Embeddings -Word2vec – RNN, LSTM Attention mechanism- Transformer architecture – Attention approximation-MHA, MQA, GQA – Position embeddings (regular, learned) – Transformer-based architectures – BERT and its derivatives.

Large Language Models: Definition and architecture – Mixture of experts – Context length, temperature – Sampling strategies – Prompting, in-context learning – Chain of thought – Self-consistency.

LLM training: Pre training –Quantization – Hardware optimization -Supervised fine tuning (SFT) – Parameter-efficient fine tuning (LoRA)

LLM tuning: Preference tuning – RLHF over view -Reward modelling – RL approaches (PPO and variants) – DPO.
LLM reasoning: Reasoning models – RL for reasoning – GRPO – Scaling

Agentic LLMs: Retrieval-augmented generation – Advanced RAG techniques – Function calling – Agents – Re Act framework.

LLM evaluation: LLM-as-a-judge overview – Best practices and benefits – Biases and pitfalls. Beyond Transformer-based LLMs- Diffusion LLMs – Adapting Transformer architecture for image – Vision Transformer – multimodal LLMs.

Text Books / References

Text Books

1. Afshine Amidi and Shervine Amidi, “Super Study Guide: Transformers & Large Language Models” First edition, United States, 2024.
2. Tanmoy Chakraborty, Introduction to Large Language Models, Wiley India, 1st Edition, 2025.

References

1. Denis Rothman, “Transformers for Natural Language Processing and Computer “, Packt Publishing, Birmingham, 2024
2. ZhihanLyu, “Applications of Generative AI”, Springer, Switzerland, 2024.
3. Daniel and Martin JH“Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics and Speech Recognition”, Prentice Hall, 2009.
4. Ben Auffarth, “Generative AI with Lang Chain: Build large language model (LLM) apps with Python, Chat GPT and other LLMs”, Packt Publishing, Birmingham, 2023

This course explores the world of Transformers and Large Language Models (LLMs). Students learn the evolution of NLP methods, the core components of the Transformer architecture, along with how they relate to LLMs as well as techniques to enhance model performance for real-world applications. Through a mix of theory and practical insights, this course will equip students with the knowledge to leverage LLMs effectively. This course is ideal for the students with a background in calculus, linear algebra, and basic machine learning concepts.

Course Outcomes: After successful completion of the course, students will be able to

• CO1:  Understandthe foundations of NLP and Transformer architectures, including attention mechanisms, positional embeddings, and encoder/decoder-based models such as BERT and its variants.
• CO2: Analyse the architecture and operational principles of LLMs, including mixture-of-expert’s models, context handling, prompting strategies, sampling methods, and reasoning techniques such as chain-of-thought and self-consistency.
• CO3: Design and implement LLM training and adaptation strategies, including pretraining concepts, supervised fine-tuning, parameter-efficient tuning (LoRA), quantization, hardware optimization, and alignment methods such as RLHF and DPO.
• CO4: Develop LLM-based systems incorporating reasoning, retrieval-augmented generation, agentic frameworks (ReAct, function calling), and evaluate their performance using modern evaluation techniques including LLM-as-a-judge and bias analysis.
• CO5: Critically assess emerging advancements beyond classical Transformer-based LLMs, including diffusion language models, vision transformers, multimodal LLMs, and scalable reasoning architectures.

CO-PO Mapping