Llm Mastery: Hands-On Code, Align And Master Llms

Code LLMs and alignment from scratch with Python and PyTorch. And explore generative AI and deep learning using Origami

What you'll learn

Code and train an LLM from scratch, line by line, understanding every concept in detail

Understand and analyze in depth an advanced LLM architecture based on the Llama system

Code and train an alignment process from scratch, to align an LLM with a preferred form of interaction

Understand in great depth key concepts like the Attention Mechanisms, the Cross Entropy Loss, the way neural networks learn and many more

Explore in depth insights about deep learning and neural networks through the use of Origami

In addition to the coding, every section includes in-depth explanations of key concepts related to these architectures and generative AI

Requirements

Basic knowledge of python. It's enough with the very basics, as we will code every little thing together, line by line

You can code either on a free online platform like Google Colab, or using your local Laptop or Desktop

Lots of enthusiasm. We are going to go really deep into both the code and the concepts around the code. You can grow and stretch your knowledge and experience a lot if you stay focused and motivated throughout the course, let's do it :)

Description

Dive into the most exhilarating and hands-on LLM course you'll ever experience! This isn't just learning—it's an adventure that will transform you from an AI enthusiast into a creator at the bleeding edge of technology. AI expert Javier Ideami, the creator of one of the most successful AI related courses on Udemy, brings you a totally unique new experience around LLM technology.What Makes This Course Unmissable:Code Your Own AI Universe: 80% hands-on coding with Python and Pytorch. Build an LLM from scratch, line by line. Watch AI come alive through your fingertips. And then go beyond and code a compact version of an alignment process, the magic that makes ChatGPT, GPT 4, Claude and Gemini possible.Origami Meets AI: Be part of a world-first! Unravel deep learning mysteries through the art of paper folding.Deep Dive, Gradual Learning Curve: Only basic Python needed. We'll guide you through all the complex concepts around LLMs, from attention mechanisms to cross-entropy and beyond. By the end of the course, you will have gained advanced skills and knowledge about generative AI and LLMs.Mind-Bending Finale: Cap it off with an optional guided meditation using the "generative AI" in your own brain. Mind = Blown!Flexible requirements: Run everything on a humble 4GB GPU or anything more powerful. From Google Colab to your trusty laptop, flexibility is the mantra. On the cloud or local (Windows / Linux / Mac). All platforms and devices will work (with a minimum of 4GB of GPU memory)Course Highlights:Intro to Generative AI: Dive into the mesmerizing world of Generative AI, where machines create and innovate beyond imagination.Code an LLM from Scratch: Code and nurture your very own LLM from scratch. Unlocking an LLM Titan: Dissect an advanced LLM architecture. Peek behind the curtain of the most powerful AI systems on the planet.Alignment. Code the Secret Sauce of the top LLMs: Code a cutting edge LLM alignment process. This is the crucial stage that makes ChatGPT, GPT 4, Claude and Gemini possible and we code together a cutting edge variation of it.Origami AI: Fold your way to neural network nirvana. Experience a world-first fusion of ancient art and cutting-edge science to grasp deep learning like never before.AI Meets Zen: Cap your journey with an optional mind-bending guided meditation. Explore the ultimate generative AI - the one in your own brain - in a profound finale that bridges technology and spirituality.All in One / Why You Can't Miss This:The full package: You code both a small LLM as well as a cutting edge alignment technique. You also go deep into the understanding of a complex LLM architecture. In parallel you dive very deeply into all sorts of complex concepts around LLMs and deep learning, both during the coding and also during the unique Origami experience as well as the initial intro to Generative AI.Uniqueness: Origami + AI = A learning experience you won't find anywhere else. Understand key insights about Deep Learning and Neural Networks through the magic of paper folding.Practical Hands-on Mastery: 80% Practical. Learn and Build, Train and Align.Future-Proof Skills: Position yourself at the forefront of the AI revolution.And there's more: Added to all of that, the course connects you with free tools, articles and infographics produced by Ideami that enrich and accelerate your learning even more. Some of then, like the Loss Landscape explorer app, are unique tools in the world, created by Ideami for you.Accessibility: Complex concepts explained so clearly, you'll feel like an AI whisperer.In summaryThis isn't just a course—it's a ticket to the AI creator's club. By the end, you'll have coded an LLM, understood its deepest secrets, coded an alignment technique, dived deep into profound insights about deep learning and gained practical skills that will make you the AI guru in any room.Ready to code the future, fold profound insights through origami and blow your own mind? Join us on this unparalleled journey to the heart of LLM technology. Enroll now and prepare for the most fun, deep, and transformative tech adventure of your life!

Overview

Section 1: Introduction to Generative AI

Lecture 1 Welcome to the course

Lecture 2 Why we will start by introducing Generative AI concepts

Lecture 3 Introducing myself

Lecture 4 Generative modelling, Evolution of Generative AI and Overview of applications

Lecture 5 Building Blocks of Machine Creativity: Machine Learning Foundations for Gen AI

Lecture 6 Architectures of Machine Imagination, from GANs to Diffusion and beyond

Lecture 7 Machine Creativity Meets Real-World Impact - Applications of Generative AI

Lecture 8 The Ethics of Machine Creativity: Challenges and Considerations in Generative AI

Lecture 9 Worlds Reimagined: Visions of the Future with Generative AI

Lecture 10 Summary and closing thoughts about this intro of GenAI

Section 2: Coding a small LLM from scratch, understanding all the key concepts involved

Lecture 11 Welcome to this section

Lecture 12 Where to do the coding - Intro

Lecture 13 Where to do the coding - Details

Lecture 14 Dealing with challenges, and reminder about coding options

Lecture 15 Setting up the coding environment

Lecture 16 How Jupyter Notebooks work

Lecture 17 Importing the necessary libraries

Lecture 18 Setting up our base files

Lecture 19 Setting up the parameters of the architecture

Lecture 20 Exploring the crucial hyperparameters

Lecture 21 Key parameters for an effective training process

Lecture 22 Introducing Logging

Lecture 23 Setting up logging

Lecture 24 Setting up the tokenizer and related functionality

Lecture 25 Splitting our data and creating our get batch function

Lecture 26 The Transformer Architecture

Lecture 27 Declaring the top layers of the LLM

Lecture 28 The forward function of the LLM

Lecture 29 The Cross Entropy Loss with Pytorch

Lecture 30 The Cross Entropy Loss recreated manually

Lecture 31 From Information to Cross-Entropy - Deep Dive

Lecture 32 Completing and verifying the manual cross entropy loss

Lecture 33 Generating new samples - Intro

Lecture 34 Creating the functionality to generate new samples

Lecture 35 Testing the sample generation functionality

Lecture 36 Coding the blocks of the LLM architecture

Lecture 37 Communication plus Computation

Lecture 38 Providing computational power to the LLM

Lecture 39 The Multi Head Attention Mechanism

Lecture 40 Attention is all you need

Lecture 41 Coding and understanding the attention head

Lecture 42 Understanding attention - deep manual dive

Lecture 43 Review and debugging example

Lecture 44 Evaluating the performance with more precision

Lecture 45 Setting up the Optimizer and Scheduler

Lecture 46 Loading checkpoints for Inference or to restart trainings

Lecture 47 Loading and testing a pre-trained checkpoint

Lecture 48 Coding the learning process - Intro

Lecture 49 The training loop

Lecture 50 Training our LLM

Lecture 51 Keeping in mind the scale of our LLM

Lecture 52 Training the tokenizer

Lecture 53 Encoding our dataset with the tokenizer

Lecture 54 Conclusions and what comes next

Section 3: Understanding the code and concepts of an Advanced LLM

Lecture 55 Welcome to a deep dive through an advanced LLM architecture

Lecture 56 Setting up a new environment and hosting the support files

Lecture 57 Declaring the main parameters of the model

Lecture 58 Main structure and loss calculation

Lecture 59 Advanced generation using extra parameters

Lecture 60 The main blocks of the architecture

Lecture 61 Analyzing the computational layers of the LLM

Lecture 62 An efficient attention implementation, part 1

Lecture 63 An efficient attention implementation, part 2

Lecture 64 Exploring rotary positional embeddings and other supporting functions

Lecture 65 Analyzing the inference code

Lecture 66 Preparing to run inference on the cloud and locally

Lecture 67 Inference on non-aligned vs aligned versions of the model

Lecture 68 Further reflections on the inference results

Section 4: Coding an alignment process from scratch, understanding all the key concepts

Lecture 69 The importance of alignment

Lecture 70 The pretraining and alignment datasets

Lecture 71 Importing the necessary libraries

Lecture 72 Setting up the parameters for the alignment process

Lecture 73 Setting up the chat template for the tokenizing process

Lecture 74 Filtering our alignment dataset

Lecture 75 Pre-processing and Tokenizing the alignment dataset

Lecture 76 Debugging and completing the pre-processing function

Lecture 77 Splitting the alignment data and creating our dataloaders

Lecture 78 Setting up the model and optimizer for the alignment training process

Lecture 79 Setting up our scheduler function

Lecture 80 Coding the training loop of the alignment process

Lecture 81 Coding the alignment loss calculation - part 1

Lecture 82 Understanding how we will favor aligned responses - Deep Dive

Lecture 83 Coding the alignment loss calculation - part 2

Lecture 84 Coding the alignment loss calculation - part 3

Lecture 85 Adding logging, checkpoint saving and launching the training

Lecture 86 Training and testing the alignment, analyzing and expanding the stats

Lecture 87 Adding new code to calculate more precise training and validation losses

Lecture 88 Comparing training and validation charts - Deep Dive

Lecture 89 Alignment wrap-up

Lecture 90 The path towards alignment

Lecture 91 Congrats, summary, and what's next

Section 5: Origami + AI: Learning key insights about neural networks and AI with Origami

Lecture 92 Welcome to this original origami based journey to the core of AI

Lecture 93 In Search of the Magical Mappings of Creativity, using Origami!

Lecture 94 The Search for the Perfect Mapping: datasets and dimensionality

Lecture 95 From Linearity to Complexity: Neural Networks and the Nonlinearities of Life

Lecture 96 Bending the Rules: Non-Linear transformations and the key to complexity

Lecture 97 Not Too Tight, Not Too Loose - Finding the perfect fit

Lecture 98 How increasing the dimensionality impacts the latent complexity of the network

Lecture 99 The Power of Depth: Creating Sophisticated Mappings with AI networks

Lecture 100 From high dimensional manifolds to dynamic and ever changing latent spaces

Lecture 101 Advanced digital representations of the latent complexity of neural networks

Lecture 102 Visualizing the Journey: Loss Landscapes and the Search for Optimal Weights

Lecture 103 Example of the dynamic Loss Landscape of a generative adversarial network

Lecture 104 Lucy - Real Time Visualization of the changing weights of a neural network

Lecture 105 Charting the hidden depths: a recap of our transformative latent space journey

Lecture 106 Summary of the last sections

Section 6: Activating the Generative Model of your own mind

Lecture 107 Introducing our final journey

Lecture 108 A guided visualization experience to exercise the generative model in your head

Lecture 109 Intro to the journey to the center of the neuron

Lecture 110 The container, the salty ocean and the 150000 cortical columns

Lecture 111 Visualizing the pyramidal neuron

Lecture 112 The Synapse, visualizing the input-output interface

Lecture 113 Biological vs Artificial Neurons: Inputs, Outputs, Speed, etc

Lecture 114 Learning in biological and artificial neurons

Lecture 115 Planning, decision making and world models

Lecture 116 Efficiency: sparsity in biological vs artificial networks

Lecture 117 Consciousness: within the neurons

Lecture 118 The future, towards AGI / ASI

Lecture 119 Conclusion and congratulations

AI Enthusiasts and Developers: Individuals with a passion for artificial intelligence and a basic knowledge of Python who want to dive deep into the world of LLMs and generative AI.,Tech Innovators and Creators: Aspiring AI developers who wish to be at the cutting edge of technology, learning to build and understand complex AI systems from the ground up.,Students and Professionals in AI: Those studying or working in AI-related fields who seek to enhance their practical skills and theoretical understanding of LLMs and deep learning.,Curious Minds with Creative Flair: Learners interested in a unique blend of technology and art, who are eager to explore deep learning concepts through innovative methods like origami.,Software engineers: interested in understanding and coding LLMs as well as implementing AI alignment techniques with LLMs,General: People that are curious about how LLMs work and want to understand them in great depth,General: People that want to stretch their generative AI experience using Python and Pytorch to program LLMs and alignment processes,General: People that want to know all the key parts of how ChatGPT, Gemini or Claude work internally, in order to get inspired about new ways of applying this technology,General: Those that want a deep introduction to Generative AI, deep learning and neural networks,General: Those that want to deeply understand how neural networks learn in a fun way and unique way through Origami