Unleashing LLMs: Training, Finetuning, and Evaluating

There is an incredible amount of research and development in Generative AI, more specifically Large Language models (LLMs). From ChatGPT to Falcon, and countless others, their names swirl around, leaving us eager to uncover their true nature. How are these models created? How to build large language models from scratch? How do they possess the ability to answer virtually any question you throw at them? These burning questions linger in our mind, fueling curiosity. The wait is over! You are in the right place.

Join us for an immersive workshop on Large Language Models (LLMs), the magical engines behind ChatGPT and Bard. Explore the cutting-edge architectures, uncover the secrets of LLMs, and witness the future of language unfold before your eyes. Get ready to embark on an extraordinary journey that will revolutionize your understanding of NLP.

Here are the detailed module wise details-

Module 1: Getting Started with LLMs

• Recap of Generative AI and its state of the art.
  • NLP (ChatGPT)
  • Computer Vision (Mid-Journey)
  • Audio (Voicebox)
  • Video (Make-a-video)
  • Multimodal (GPT4, MUM)
• Introduction to Large Language Models (LLMs)
• Why do we need LLMs?
• Understand different types of LLMs
  • Continuing the text
  • Dialogue Optimized

Module 2: Understand the current state of the art LLMs

• Transformers
• BERT
• GPT and its variants
• ChatGPT
• LIMA
• Falcon
• OpenChat

Module 3: Training and Evaluating LLMs and their Best Practices

• Build vs Buy Pretrained LLM models?
• Understand cost and infrastructure involved
• Discuss the scaling laws
• Explore Parallel and Distributed Computing Strategies
• Understand the data collection and preprocessing steps
• Train LLMs from scratch with best practices
  • Continuing the text
  • Instruction aligned
• Evaluate LLMs 

Module 4: Finetuning and Prompt Engineering

• How can we use LLMs on the downstream tasks?
  • In Context Learning (Prompting)
  • Finetuning
• Learn about prompt engineering and its different techniques.
• Understand different finetuning techniques
• Finetune LLMs on downstream use cases.
• When to use prompting/finetuning?

Module 5: Parameter Efficient Fine Tuning methods

• Understand the challenges on finetuning LLMs
• What is Parameter Efficient Fine Tuning (PEFT)?
• Understanding different PEFT techniques
  • Prefix Tuning
  • LoRA
  • QLoRA
• Finetune LLM on a single GPU using PEFT techniques

Module 6: Master LLM Tools and Frameworks

• Understand LLM tools and frameworks like LlamaIndex, LangChain and Vector DBs and learn when to use them?
• Build chatbots on a private knowledge base.

Pre-requisites:

• System Requirement and Setup
  • Laptop with at least 4-8 GB of RAM
  • We will be using a cloud jupyter notebook powered by GPU for the workshop
• Offline Setup [Optional]
  • GPU good to have!
  • Install Python3.9 or higher version(Resource)
  • Install jupyter notebook (Resource)
• Pre-reads
  • Programming knowledge in Python (Resource)
  • Jupyter Notebook Environment familiarity (Resource)
  • Basics of Machine Learning and Deep Learning (Resource,Resource)
  • Fundamentals of NLP(Resource1,Resource2)
  • Familiarity with Pytorch(Resource)

Note: These are tentative details and are subject to change.