book

Quick Start Guide to Large Language Models: Strategies and Best Practices for ChatGPT, Embeddings, Fine-Tuning, and Multimodal AI, 2nd Edition

by Sinan Ozdemir

October 2024

Intermediate to advanced

384 pages

13h 7m

English

Addison-Wesley Professional

Read now

Unlock full access

Audience and PrerequisitesHow to Approach This BookOverviewUnique FeaturesSummary

What Are Large Language Models?Popular Modern LLMsApplications of LLMsSummary
IntroductionThe TaskSolution OverviewThe ComponentsPutting It All TogetherThe Cost of Closed-Source ComponentsSummary
IntroductionPrompt EngineeringWorking with Prompts Across ModelsSummary
IntroductionThe Ever-Shifting Performance of Closed-Source AIAI Reasoning versus ThinkingCase Study 1: Retrieval Augmented GenerationCase Study 2: Automated AI AgentsConclusion
IntroductionTransfer Learning and Fine-Tuning: A PrimerA Look at the OpenAI Fine-Tuning APIPreparing Custom Examples with the OpenAI CLISetting Up the OpenAI CLIOur First Fine-Tuned LLMSummary
IntroductionPrompt Injection AttacksInput/Output ValidationBatch PromptingPrompt ChainingCase Study: How Good at Math Is AI?Summary
IntroductionCase Study: Building a Recommendation SystemSummary
IntroductionAligned to Whom and to What End?Alignment as a Bias MitigatorThe Pillars of AlignmentConstitutional AI: A Step Toward Self-AlignmentConclusion
IntroductionCase Study: Visual Q/ACase Study: Reinforcement Learning from FeedbackSummary
IntroductionExample: Anime Genre Multilabel Classification with BERTExample: LaTeX Generation with GPT2Sinan’s Attempt at Wise Yet Engaging Responses: SAWYERSummary
IntroductionDeploying Closed-Source LLMs to ProductionDeploying Open-Source LLMs to ProductionSummary
IntroductionEvaluating Generative TasksEvaluating Understanding TasksConclusionKeep Going!
The LLM already knows about the domain I’m working in. Why should I add any few-shot examples or grounded content as in RAG?I just want to deploy a closed-source API. What are the main things I need to look out for?I really want to deploy an open-source model. What are the main things I need to look out for?Creating and fine-tuning my own model architecture seems hard. What can I do to make it easier?I think my model is susceptible to prompt injections or going off task. How do I correct it?Why didn’t we talk about third-party LLM tools like LangChain?How do I deal with overfitting or underfitting in LLMs?How can I use LLMs for non-English languages? Are there any unique challenges?How can I implement real-time monitoring or logging to understand the performance of my deployed LLM better?What are some things we didn’t talk about in this book?
Transformer ArchitectureAttention MechanismLarge Language Model (LLM)Autoregressive Language ModelsAutoencoding Language ModelsTransfer LearningPrompt EngineeringAlignmentReinforcement Learning from Human Feedback (RLHF)Reinforcement Learning from AI Feedback (RLAIF)CorporaFine-TuningLabeled DataHyperparametersLearning RateBatch SizeTraining EpochsEvaluation MetricsIncremental/Online LearningOverfittingUnderfittingKnowledge DistillationTask-Specific DistillationTask-Agnostic DistillationMultimodal ModelsAlignment
General Chatbots / Retrieval Augmented Generation (RAG)AgentsFine-Tuning a Closed-Source LLMFine-Tuning an Open-Source LLMFine-Tuning a Bi-encoder to Learn New EmbeddingsFine-Tuning an LLM for Following Instructions Using Both LM Training and Reinforcement Learning from Human / AI Feedback (RLHF and RLAIF)Open-Book Question-AnsweringVisual Question-Answering