Last updated: March 06, 2025

By M. Ali Yousuf

[Disclaimer: The code on this page is NOT written by me and comes from the original author/website mentioned in each Google Colab notebook. I have made modifications, when needed, to make them run from within Google Colab, and some extra code/text added to explain the code.]

[Note: If you are new to the field of AI and have not learned it yet, there is a precursor to this course, "Machine Learning For Non-Computer Science Majors" which you can find here https://sites.google.com/view/aiandml4all/home]

Introduction

This course is a collection of labs to give you some hands-on experience. Please note that I am NOT the author of these examples - I have just collected them here with some additional notes in the notebooks. You will find links to the original versions within the notebooks.  

Generative artificial intelligence (AI) is a type of AI that uses existing data to create new and realistic content. This content can include text, images, audio, video, and more. Generative AI is different from other types of AI because it produces new data, rather than just analyzing it. The goal is to create content that is similar to what humans would create. [Text generated by Google AI].

Generative AI utilizes various machine learning methods to create new data. Here are some common ones:

• Generative Adversarial Networks (GANs): Two neural networks compete, with one generating new data (generator) and the other evaluating its authenticity (discriminator).

• Diffusion Models: These models progressively add noise to data, then learn to reverse the process, essentially denoising the data to create new content.

• Variational Autoencoders (VAEs): These models encode data into a latent space, allowing for the manipulation and generation of similar data.

• Transformers: While not exclusive to generative AI, transformers are powerful neural network architectures that can be used for various tasks, including generating different creative text formats.

References

Some of the sources used:

1. SuperAnnotate - Introduction to diffusion models for machine learning

2. TechTarget SearchEnterpriseAI - What is Generative AI? Everything You Need to Know

3. IBM - What is Generative AI?

4. Medium - Generative AI (Part-1)

5. Encord - An Introduction to Diffusion Models for Machine Learning

A note on Google Colab notebooks

When you will click on any file below, it will open and you will be able to make changes. However, you will NOT be able to save those changes as these are my files and you are only a viewer. To get full access, make a 'copy' by clicking on File -> 'Save a Copy in Drive'. That way, you will have a copy of the file in YOUR google drive, under the folder 'Colab Notebooks' (generally a yellow colored folder symbol). You can edit that version as much as you want and it will be saved on your drive. In case you mess up, go back to the original link to the file in my Google Drive and make another fresh copy!

What is Generative AI? 

• My own presentation on the topic: What is Generative AI (Google Slides)

• Another of my presentations on Generative Art (Google Slides).

• Understanding LLMs from Scratch Using Middle School Math, https://towardsdatascience.com/understanding-llms-from-scratch-using-middle-school-math-e602d27ec876 

• Generative AI exists because of the transformer (Financial Time article, very good): https://ig.ft.com/generative-ai/ 

• The IBM Technology YouTube channel has many high-quality videos, all very short (less than 10 minutes each) describing various aspects of the field. Here are two starting points:

  • What are Generative AI models? (IBM Technology), https://www.youtube.com/watch?v=hfIUstzHs9A&t=430s  

  • How Large Language Models Work (IBM Technology), https://youtu.be/5sLYAQS9sWQ?si=WLtiQsya0x6bQxFb  

Fine-Tuning

• Fine-tune GPT2 For Sequence Classification

• The CPU version takes about 14 hours to run, so run it on Google Colab GPU with max RAM and it takes 16 minutes only! https://colab.research.google.com/drive/1h2lY-zE2O_oAD4uDxQW_o9MAVbw9zeNB?usp=sharing . The GPU version works better! (It consumes around $1 of credit from GC).

• Want to know more about vectorizers before starting the above? Check https://colab.research.google.com/drive/1qJ1pvafSEMhFehZ97qqxDKeech5o3a-O?usp=sharing 

• Want to play more with vector embeddings? See https://colab.research.google.com/drive/1PXmg1erDvxq1Msdww84zxrYwtX2JQA58?usp=sharing 

• Fine-Tune a Vision Language Model Llava 

• This example comes from https://huggingface.co/blog/vlms. 

• The video presentation can be found here: https://www.youtube.com/live/cambXXq9mrs?feature=shared

• The notebook itself is a copy and the link to the original can be found in the notebook, https://colab.research.google.com/drive/1-AR1OC6Csm4rPoWTM8vM8sFI55nye4l_?usp=sharing 

• OpenAI with Python

  • Fine-tune models for better results and efficiency (using Python), https://platform.openai.com/docs/guides/fine-tuning/fine-tuning 

  • Data preparation and analysis for chat model fine-tuning, https://cookbook.openai.com/examples/chat_finetuning_data_prep 

• LitGPT 

• Use, finetune, pretrain, and deploy LLMs Lightning fast. Every LLM is implemented from scratch with no abstractions and full control, making them blazing fast, minimal, and performant at enterprise scale. Choose from 20+ LLMs. Requires HF Token

• https://colab.research.google.com/drive/1Zf8Ta_0SP9gKn91kMxhZAajaWAXzOBNp?usp=sharing 

• Fine Tune with Unsloth and Ollama

  • Watch the video https://www.youtube.com/watch?v=pxhkDaKzBaY [MUST watch the last part of the video that requires work outside Google Notebook and uses Ollama]

  • Google Colab, https://colab.research.google.com/drive/1GY3m_AXubttIaTZ-Yv6FMvAYqgBn99qO?usp=sharing 

• Free and Tiny Models

• Find free (and tiny) models and code here: https://huggingface.co/models?sort=trending&search=tiny Everything here runs on a CPU.

• Google Colab, https://colab.research.google.com/drive/12vz_KCk81gqxRJ1-R67B_7KWPCJ_yTyB?usp=sharing 

• Fine-Tune ChatGPT in Plain English 

• There are various ways to fine-tune a LLM but this one requires a paid chatgpt.com account ($20 per month at the time of this writing). It allows you to fine-tune a model using simple instructions and (optional) configuration. You can upload documents of your own to this system.  https://chatgpt.com/gpts/editor/ 

RAGs - Retrieval-Augmented Generation

RAGs (Retrieval-Augmented Generation)

• What is Retrieval-Augmented Generation and why is it better? https://youtu.be/T-D1OfcDW1M?si=gql4IZ3-i5zp9e_q  

• I want Llama3 to perform 10x with my private knowledge" - Local Agentic RAG w/ llama3, https://youtu.be/u5Vcrwpzoz8?si=3VmJv1oEFCBrJ1qR 

• A Crash Course on Building RAG Systems – Part 1 (With Implementation), https://www.dailydoseofds.com/a-crash-course-on-building-rag-systems-part-1-with-implementations/

• A Crash Course on Building RAG Systems – Part 2 (With Implementation), https://www.dailydoseofds.com/a-crash-course-on-building-rag-systems-part-2-with-implementations/

• A Crash Course on Building RAG Systems – Part 3 (With Implementation), https://www.dailydoseofds.com/a-crash-course-on-building-rag-systems-part-3-with-implementation/ 

• Understanding RAG Part I: Why It’s Needed

• Understanding RAG Part II: How Classic RAG Works 

• Understanding RAG Part III: Fusion Retrieval and Reranking

• Understanding RAG Part IV: RAGAs & Other Evaluation Frameworks

• Understanding RAG Part V: Managing Context Length

• Understanding RAG Part VI: Effective Retrieval Optimization

• Simple RAG 

• This version uses plain text but in just a few lines - making the whole process very clear. 

• We use OpenAI's API again (you need the openai_api_key, which generally comes with a paid account) https://colab.research.google.com/drive/1CombJNNVmzx9FhsnQz-35JUg0roN2zbG?usp=sharing 

• Advanced RAG 

• This version allows you to upload pdf files and convert each line into plain text. 

• We use OpenAI's API again (you need the openai_api_key, which generally comes with a paid account) 

• The Google Colab file can be found here: https://colab.research.google.com/drive/1yXQAzuZam6hd7YS1AU8iGQ1ZNPoGJeUy?usp=sharing 

• RAG using Clarifai

• Large Language Models are not up-to-date, and they also lack domain-specific knowledge, as they are trained for generalized tasks and cannot be used to ask questions about your own data.

• That's where Retrieval-Augmented Generation (RAG) comes in: an architecture that provides the most relevant and contextually important data to the LLMs when answering questions.

• Learn more about the differences between RAG and Fine Tuning here, https://www.youtube.com/watch?v=00Q0G84kq3M 

• For this purpose, we'll use the clarifai system. Please open a free account at: https://www.clarifai.com/ 

• Here is the jupyter notebook that achieves this in only a few lines of code:

• https://colab.research.google.com/drive/1bHIOtwuU3CGvduZsBeqXCzf52FFSzqW0?usp=sharing 

• Building A RAG System with Gemma, MongoDB and Open Source Models

• Authored by Richmond Alake. From https://huggingface.co/learn/cookbook/en/rag_with_hugging_face_gemma_mongodb which is available here as a google colab. Original can be found here: https://colab.research.google.com/github/huggingface/cookbook/blob/main/notebooks/en/rag_with_hugging_face_gemma_mongodb.ipynb#scrollTo=5gCzss27UwWw 

• My copy, https://colab.research.google.com/drive/1brwqjA97_P6P0WIel-y289lUB6FfJoMs?usp=sharing 

• RAG using LangChain and HuggingFace

• This code comes from https://medium.com/@akriti.upadhyay/implementing-rag-with-langchain-and-hugging-face-28e3ea66c5f7

• https://colab.research.google.com/drive/1wQFGHy03ux7As0tPkDENquRZh5MIDzlG?usp=sharing 

Building Your Second Brain AI Assistant Using Agents, LLMs and RAG

• Six lessons and code at https://github.com/decodingml/second-brain-ai-assistant-course/tree/main?tab=readme-ov-file 

Further Readings

1. Advanced RAG Techniques: Elevating Your Retrieval-Augmented Generation Systems, https://github.com/NirDiamant/RAG_Techniques 

2. 5 Python Libraries to Build an Optimized RAG System, https://machinelearningmastery.com/5-python-libraries-build-optimized-rag-system/?utm_source=drip&utm_medium=email&utm_campaign=MLM+Newsletter+January+24%2C+2025&utm_content=Python+Libraries+to+Build+an+Optimized+RAG+System+%E2%80%A2+Integrating+Language+Models+Into+Your+Text+Adventure+Games 

3. Building a Retrieval-Augmented Generation (RAG) System with DeepSeek R1: A Step-by-Step Guide, https://www.marktechpost.com/2025/01/27/building-a-retrieval-augmented-generation-rag-system-with-deepseek-r1-a-step-by-step-guide/ 

4. ColiVara - Make your RAG application 10x smarter, https://github.com/tjmlabs/ColiVara?tab=readme-ov-file