Large Language Model (LLM) Tutorial

Large Language Models (LLMs) are machine learning models trained on vast amount of textual data to generate and understand human-like language. These models can perform a wide range of natural language processing tasks from text generation to sentiment analysis and summarisation.

Basics

Large Language Models (LLMs) are advanced AI systems trained on massive datasets to understand and generate human-like text, powered by deep learning techniques.

• Large Language Models (LLMs)
• Language Models
• Foundation Models
• Seq2Seq Models
• Multimodal LLMs
• LLM Parameters
• Tokens and Context Windows
• Prompts 
• Prompt Engineering
• LLM Hallucinations
• LLM Model Evaluation
• Word Embedding
• Tokenization
• Byte Pair Encoding
• Data sampling using Sliding Window Attention

Transformers

Transformers are the foundational architecture behind most modern LLMs that rely on attention mechanisms to process the entire sequence of the data simultaneously.

• Attention Mechanism
• Self-Attention Mechanism 
• Multi-Head Attention Mechanism 
• Positional Encoding 
• Feed-Forward Neural Network
• Layer Normalization 
• Encoder-Decoder Model
• Masked Attention
• Cross-Attention Mechanism
• Embedding Layers
• Transformers from Scratch using TensorFlow
• Transformers from Scratch using PyTorch
• Transformers vs LLMs

Training and Fine-Tuning

This section explains how LLMs are trained on massive datasets and later adapted for specific tasks using fine-tuning and prompting techniques.

Language Modeling Techniques

• Autoregressive Models
• Masked Language Models
• Causal Language Models

Fine-tuning

• Reinforcement Learning from Human Feedback (RLHF)
• Fine-Tune an LLM from Hugging Face
• Parameter-Efficient Fine-Tuning (PEFT)
• LoRA (Low-Rank Adaptation)
• QLoRA (Quantized Low-Rank Adaptation)
• Prompt Tuning
• Prompt Tuning Techniques
• Instruction Tuning
• Supervised Fine-Tuning (SFT)
• LLM Distillation

Prompting Techniques

• Zero-Shot Prompting
• Few-Shot Prompting
• Chain-of-Thought (CoT) Prompting
• Self-Consistency Prompting
• Zero-Shot Chain-of-Thought Prompting
• ReAct (Reasoning + Acting) Prompting
• Retrieval-Augmented Prompting

Retrieval-Augmented Generation (RAG)

This section explains how RAG combines information retrieval with language models to generate responses using external knowledge sources.

• What is Retrieval-Augmented Generation (RAG)?
• RAG vs Traditional QA
• Fine tuning vs RAG
• Dense Passage Retrieval (DPR)
• Vector Database
• Chunking in RAG
• Agentic RAG
• Mutlimodal RAG
• How to build RAG Pipeline for LLMs?
• Evaluation Metric for RAG

Popular LLMs and Evaluation

This section introduces widely used LLMss and the metrics used to measure their performance.

• GPT
• BERT
• T5
• LLaMA 
• Claude 
• LCM
• Falcon 

Evaluation

• Perplexity
• BLEU and ROUGE
• Benchmarks

Applications

LLMs are used in various real-world applications including:

• Building Chatbot using Gemini
• Building Chatbot using OpenAI
• Building Chatbot using LLama3
• Sentiment analysis using BERT
• Text generation using FNet
• Text2text generation using Hugging Face Model
• Machine Translation using Transformer
• Generate Images from Text using Stable Diffusion
• Building a RAG Application