Large Language Models : A Hands on Approach
Jan - May, 2026 @ Center for Continuing Education, Indian Institute of Science
Logistics
- Duration: 18 weeks (Jan - May 2026)
- Format: Online, Tue and Thu, 7:30 - 9:00 PM IST
- Contact: TBA
Course Description
LLMs have become mainstay of NLP and are transforming every domain, from software development, research, and business intelligence to education. However, deploying them efficiently remains a specialized engineering challenge.
This course provides an engineering-focused exploration of Large Language Models (LLMs). Participants will go from understanding transformer architectures and GPU internals to mastering fine-tuning, inference optimization, and large-scale deployment across GPUs, clusters, and edge devices. Through a theory-to-practice approach, including case studies, hands-on labs, and projects, learners will cover key topics such as model architecture, fine-tuning techniques, inference optimization, serving strategies, and applications in retrieval-augmented generation (RAG) and agentic systems.
Learning Outcomes
By the end of this course, participants will be able to:
- Understand LLM Architecture: Master transformer architectures, attention mechanisms, and modern LLM variants (GPT-OSS, Qwen, Gemma, etc.).
- Optimize Inference: Implement efficient inference strategies including quantization, KV caching, and serving via inference engines like vLLM.
- Fine-tune Models: Apply various fine-tuning techniques including LoRA, QLoRA, instruction tuning, and preference alignment.
- Build RAG Systems: Design and implement Retrieval-Augmented Generation pipelines.
- Develop AI Agents: Create tool-using agents with the ReAct framework.
- Deploy at Scale: Set up production-ready LLM serving infrastructure with cost optimization.
- Multimodal Models: Work with vision-language models and speech.
- Evaluation: Understand evaluation strategies for LLMs and RAG systems.
Prerequisites
- Proficiency in Python and familiarity with any deep learning framework (PyTorch preferred).
- Basic understanding of neural networks.
- Working knowledge of Linux, Docker, and Git.
- Optional but recommended: experience with GPU computing.