Large Language Models Bootcamp
•
2 likes•4,318 views
This document provides a 50-hour roadmap for building large language model (LLM) applications. It introduces key concepts like text-based and image-based generative AI models, encoder-decoder models, attention mechanisms, and transformers. It then covers topics like intro to image generation, generative AI applications, embeddings, attention mechanisms, transformers, vector databases, semantic search, prompt engineering, fine-tuning foundation models, orchestration frameworks, autonomous agents, bias and fairness, and recommended LLM application projects. The document recommends several hands-on exercises and lists upcoming bootcamp dates and locations for learning to build LLM applications.
Report
Share
More Related Content
What's hot
What's hot (20)
Similar to Large Language Models Bootcamp
Similar to Large Language Models Bootcamp (20)
Recently uploaded
Recently uploaded (20)
Large Language Models Bootcamp
- 1. Building LLM Applications A 50-hour roadmap A comprehensive roadmap to building large language model applications in ~50 hours
- 2. Types of generative AI models Text-based models Image-based models Foundation models & LLMs Encoder decoder Attention mechanism Transformers model and BERT model Intro to Image Generation Image captioning models Diffusion models Generative AI applications ChatGPT & Bard DALL-E & Midjourney Quick overview of generative AI, LLMs, and foundation models. Learn more about how transformers and attention mechanism works behind the text and image-based models: Introduction to Generative AI Roadmap to become an LLM applications developer Data Science for Everyone https://datasciencedojo.com
- 3. Large language models and foundation models Vector databases, embeddings, and LLM cache Prompts and prompt engineering Context window and token limits Embeddings and vector databases Build custom LLM applications by: Training a new model from scratch Fine-tuning foundation LLMs In-context learning Canonical architecture for an end-to-end LLM application Understand the common use cases of large language models and the fundamental building blocks of such applications. Learners will be introduced to the following topics at a very high level without going into the technical details: Emerging Architectures Roadmap to become an LLM applications developer Data Science for Everyone https://datasciencedojo.com
- 4. Review of classical techniques Review of binary/one-hot, count-based, and TF-IDF techniques for vectorization Capturing local context with n-grams and challenges Semantic encoding techniques Overview of Word2Vec and dense word embeddings Application of Word2Vec in text analytics and NLP tasks Hands-on exercise Creating a TF-IDF and semantic embeddings on a document corpus In this module, we will be reviewing how embeddings have evolved from the simplest one-hot encoding approach to more recent semantic embedding approaches. The module will go over the following topics: Embeddings Roadmap to become an LLM applications developer Data Science for Everyone https://datasciencedojo.com
- 5. Text embeddings Word and sentence embeddings Multilingual sentence embeddings Text similarity measures Dot product, cosine similarity, inner product Hands-on exercise Calculating similarity between sentences using cosine similarity and dot product Attention mechanism and transformer models Neural machine translation (NMT) and sequence-to-sequence models Attention mechanism components Self-attention and multi-head attention Transformer networks: Tokenization, embedding, positional encoding, and transformers block Hands-on exercise Understanding attention mechanisms: Self-attention for contextual word analysis Dive into the world of large language models, discovering the potent mix of text embeddings, attention mechanisms, and the game-changing transformer model architecture. This module consists of: Roadmap to become an LLM applications developer Attention Mechanism and Transformers Data Science for Everyone https://datasciencedojo.com
- 6. Overview The rationale for vector databases Importance of vector databases in LLMs Popular vector databases Indexing techniques Product quantization (PQ), Locality sensitive hashing (LSH), and Hierarchical navigable small world (HNSW) Retrieval techniques Cosine similarity Nearest neighbor search Hands-on exercise Creating a vector store using HNSW Creating, storing, and retrieving embeddings using cosine similarity and nearest neighbors Learn about efficient vector storage and retrieval with vector databases, indexing techniques, retrieval methods, and hands-on exercises: Vector Databases Roadmap to become an LLM applications developer Data Science for Everyone https://datasciencedojo.com
- 7. Understanding and implementing semantic search Introduction and importance of semantic search Distinguishing semantic search from the lexical search Semantic search using text embeddings Exploring advanced concepts and techniques in semantic search Multilingual search Limitations of embeddings and similarity in semantic search Improving semantic search beyond embeddings and similarity Hands-on exercise Building a simple semantic search engine with multilingual capability Understand how semantic search overcomes the fundamental limitation in lexical search i.e. lack of semantics. Learn how to use embeddings and similarity in order to build a semantic search model: Semantic Search Data Science for Everyone https://datasciencedojo.com Roadmap to become an LLM applications developer
- 8. Prompt design and engineering Prompting by instruction Prompting by example Controlling the model output When to stop Being creative vs. predictable Saving and sharing your prompts Use case Ideation Utilizing goal, task, and domain for perfect prompt Example use cases Summarizing (summarizing a technical report) Inferring (sentiment classification, topic extraction) Transforming text (translation, spelling, and grammar correction) Expanding (automatically writing emails) Unleash your creativity and efficiency with prompt engineering. Seamlessly prompt models, control outputs, and generate captivating content across various domains and tasks. This module includes: Prompt Engineering Roadmap to become an LLM applications developer Data Science for Everyone https://datasciencedojo.com
- 9. Fine-tuning foundation LLMs Rationale for fine-tuning Limitations of fine-tuning Parameter efficient fine-tuning Hands-on exercise Fine-tuning and deploying the OpenAI GPT model on Azure Discover the ins and outs of fine-tuning foundation language models (LLMs) through theory discussions, exploring rationale, limitations, and parameter efficient fine-tuning (PEFT): Fine-Tuning Foundation Models Roadmap to become an LLM applications developer Data Science for Everyone https://datasciencedojo.com
- 10. Why are Orchestration Frameworks (OF) needed? Eliminate the need for foundation model retraining Overcoming token limits Connecters for data sources Explore the necessity of orchestration frameworks, tackling issues like foundation model retraining, token limits, data source connectivity, and boilerplate code. Discover popular frameworks, their creators, and open-source availability: Orchestration Frameworks Roadmap to become an LLM applications developer Data Science for Everyone https://datasciencedojo.com
- 11. Introduction toLangChain Schema, models, and prompts Memory and chains Loading, transforming, indexing, and retrieving data Document loader Text splitters Retrievers LangChain use cases Summarization: Summarizing long documents QnA using documents as context Extraction: Getting structured data from unstructured text Evaluation: Evaluating outputs generated from LLM models Querying tabular data without using any extra code Hands-on exercise Using LangChain loader, splitter, and retrievals on a pdf document Build LLM apps using LangChain. Learn about LangChain's key components such as models, prompts, parsers, memory, chains, and QnA. Get hands-on evaluation experience: LangChain Roadmap to become an LLM applications developer Data Science for Everyone https://datasciencedojo.com
- 12. Agents and tools Agent types Conversational agents OpenAI functions agents ReAct agents Plan and execute agents Hands-on exercise: Create and execute some of the following agents Excel agent JSON agent Python Pandas agent Document comparison agent Power BI agent Use LLMs to make decisions about what to do next. Enable these decisions with tools. We’ll learn what they are, how they work, and how to use them within the LangChain library to superpower our LLMs. In this module, we’ll talk about: Autonomous Agents Roadmap to become an LLM applications developer Data Science for Everyone https://datasciencedojo.com
- 13. Ethics, bias, fairness Sources of bias in acquisition/annotation of training data, model building Precautions against safeguarding the model from bias Review some of the regulations/legislation Principles of responsible AI Fairness and eliminating bias Reliability and safety Privacy and data protection Transparency and explainability Accountability and governance Inclusivity and accessibility Review some of the tools available to assess the following in a large language model application Correctness and security Bias, fairness, and explainability of the model Bias can creep in at any stage of the lifecycle of a model. While large language models offer tremendous business value, humans are involved in all stages of the lifecycle of an LLM from acquisition of data to interpretation of insights. In this module, we will learn about the following: Bias, Fairness and Explainablity Roadmap to become an LLM applications developer Data Science for Everyone https://datasciencedojo.com
- 14. Virtual assistant: A dynamic customer service agent designed for the car manufacturing industry. Content generation (Marketing co-pilot): Enhancing your marketing strategies with an intelligent co-pilot. Conversational agent (Legal and compliance assistant): Assisting with legal and compliance matters through interactive conversations. QnA (IRS tax bot): An intelligent bot designed to answer your questions about IRS tax-related topics. Content personalizer: Tailoring content specifically to your preferences and needs. YouTube virtual assistant: Engage in interactive conversations with your favorite YouTube channels and playlists. Recommended Projects Roadmap to become an LLM applications developer Data Science for Everyone https://datasciencedojo.com
- 15. Learn to Build LLM Applications New York December 4-8, 2023 Roadmap to become an LLM applications developer Data Science for Everyone https://datasciencedojo.com Join this 5-day | 40-hour bootcamp to get started with building large language model applications on your enterprise data Seattle September 18-22, 2023 Washington, D.C. October 16-20, 2023 Austin November 6-10, 2023 Singapore January, 2024