From Traditional ML to LLMs

Name: From Traditional ML to LLMs
Rating: 5.0 (2 reviews)

Bridging the gap from ML basics to advanced LLMs

Created byLiana Isayan

Last updated 9/2024

English

What you'll learn

Leveraging traditional ML knowledge for working with LLMs
Hands-on experience with PyTorch for LLMs
Deeply understand the details of Transformer architecture
Unfold the use-cases of LLMs for different tasks
Discover new evaluation metrics specifically for LLMs
Perform Text Classification and Text Summarization in Python
Get familiar with concepts like RLHF or OpenAI API
Confidence to make the first steps in LLMs

Course content

6 sections • 22 lectures • 39m total length

Welcome!0:58
Bridge traditional ML to LLMs by exploring transformer architecture, use cases, and evaluation metrics, then apply a practical challenge to build classification and text summarization models with PyTorch.
Requirements: What you need to bring from traditional ML?1:47
Bring basic math, deep learning architectures, natural language processing basics, and evaluation metrics to work with large language models, using PyTorch or TensorFlow for classification and text summarization tasks.

Requirements: Basic Math (PCA vs DNN math)1:47
Acquire essential math knowledge from traditional ML to LLMs, including statistics, probability, linear algebra, and calculus basics, with PCA concepts like covariance, SVD, variance, mean, standardization, and projection.
Refresher quiz
Requirements: Deep Learning Models Architectures2:37
Explore famous deep learning architectures: sequence-to-sequence, sequence-to-vector, vector-to-sequence, and encoder-decoder. Apply these to time series forecasting, classification, text generation, and machine translation, including LLMs for summarization and question answering.
Knowledge check
Requirements: NLP (Tokenization, Corpus & Vocab, etc.)1:25
Apply natural language processing fundamentals to large language models by using tokenization, corpus and vocab construction, stop words removal, stemming, and vectorization to create numerical representations for training.
Knowledge check
Requirements: Evaluation metrics1:05
Explore evaluation metrics from traditional ML, including accuracy, F-one score, and confusion matrix, and preview LLM metrics like perplexity and exact match, with emphasis on human feedback.
Knowledge check
Requirements: Tools (Pytorch, Tensorflow)0:36
Learn how building and using large language models, including accessing pre-trained models, relies on PyTorch and TensorFlow, with PyTorch for text classification and summarization.
Knowledge check
Requirements: Summary1:11
Bridge traditional ML knowledge to LLMs by applying linear algebra, calculus, probability theory or statistics, optimization, deep learning architectures, natural language processing concepts, evaluation metrics, and Python programming.

The three types of Transformer architectures1:02
Explore transformer architectures, including encoder-only, decoder-only, and encoder-decoder designs. Identify BERT, GPT, and Gemini as key examples of these architectures in modern large language models.
The Key components of Transformer architectures1:34
Explore the key components of transformer architectures, including the attention mechanism (self-attention and cross-attention), feed-forward networks, layer normalization, skip connections, and dropouts, focusing on model head and body.
What is the difference between encoder only vs decoder only models?1:17
Encoder only transformer1:57
Explore the encoder-only transformer architecture, from input embeddings and positional encoding through bidirectional multi-head self-attention in encoder layers, producing context-aware token embeddings for tasks like text classification and semantic search.
Decoder only transformer2:49
Explore the decoder-only transformer architecture, featuring masked multi-head attention with an upper triangular attention mask and autoregressive past-token focus, ideal for prompts-based next-token prediction and creative music and art generation.
Encoder-Decoder transformer1:53
Explore the encoder-decoder transformer and its cross-attention mechanism, which blends encoder and decoder inputs to predict the next token in sequence-to-sequence tasks like translation.
Knowledge check

The main and advanced tasks LLMs can perform2:45
Explore the main uses of llms for daily tasks, including text classification, generation, summarization, translation, and question answering, with notes on advanced time series forecasting and reinforcement learning applications.
Knowledge check
LLMs' Evaluation metrics3:37
The lecture explains how to evaluate LLMs with perplexity, BLEU, recall-oriented understudy for gisting evaluation, exact match, and F-score with ordering penalties across generation, summarization, translation, and question answering.
Knowledge check
Reinforcement Learning from Human Feedback (RLHF)3:51
Explore reinforcement learning from human feedback (rlhf) and how fine-tuning, prompt engineering, and retrieval augmented generation (rag) shape llms performance by combining human labels, domain adaptation, and external knowledge.
Knowledge check

Intro the dataset and the task(s) description0:56
Explore practical use of large language models by classifying sentiment and summarizing reviews with the Yelp polarity dataset, evaluated via accuracy, F1, rouge, and blue scores.
Text classification and text summarization (codes)2:00
Utilize the HuggingFace transformers pipeline to perform binary classification on reviews and print predictions, then apply text summarization using rouge and bleu metrics.
Coding Assignment
Computation Resources - available options and the best practice1:52
Explore freely available LLM resources like Colab, Kaggle, and Paperspace, noting Colab's NVIDIA Tesla T4 12-hour sessions, Kaggle's NVIDIA Tesla P100 9–12 hours, and Paperspace's 5 GB 6-hour option.
For future consideration: Open AI API, AI applications1:41
Discover how the OpenAI API serves as the interface to access models, including authentication with an access key, and simple code to connect, while noting cost implications.

Requirements

Knowledge of traditional ML basic concepts and Python

Description

Unlock the most recent 'now' of machine learning with this hands-on, fast-paced crash course entitled "From Traditional ML to LLMs."

Your Story: [Hypothetical] Anna, a seasoned ML engineer, had mastered traditional machine learning models, but every job listing screamed "LLMs." The world was moving on, and she needed to keep up. Learning Large Language Models sounded like a daunting leap—until she found a way to bridge her existing skills with the cutting-edge techniques she needed. This course was her solution.

[Hypothetical] Jamal was a data scientist with strong ML experience, but transformers and tokenization seemed like a different universe. He needed to add LLMs to his skill set to stay competitive, and he didn't want theory; he wanted practical, hands-on applications that would help him shine in real-world projects.

My Story: I’ve been where you are—armed with traditional ML knowledge but looking to level up. I struggled with endless tutorials and theories, but through persistence, I got hands-on and found the perfect way to apply my traditional ML expertise to LLMs. I went from logistic regression models to transformer-based LLMs, and now I want to help you do the same. By the end of this course, you'll confidently build and fine-tune LLMs using your existing knowledge, apply PyTorch, and solve real-world text-based challenges.

What You'll Learn: In this course, I won’t just throw theory at you. You'll gain real, actionable skills to bridge the gap from traditional ML to LLMs, helping you tackle practical challenges in the industry. Here’s what you’ll get:

Core skills refreshed and connected to LLMs.
A deep understanding of the famous Transformers.
Practical insights into LLM concepts — from tokenization to RLHF.
A hands-on project-based approach where you'll build a text classification and a summarization model using PyTorch.

How This Course is Structured: I know learning LLMs can feel like stepping into a foreign world. So, I’ve designed this course to be practical and fun—no abstract concepts, just real-world applications. I'll walk you through exercises and examples based on actual ML-to-LLM workflows. Expect quizzes and assignments that you can apply directly to your work.

FAQs:

Do I need to know LLMs already? - Nope! We'll cover everything you need from basic architecture concepts to advanced LLMs.
Will this course work for PyTorch beginners? - Absolutely! We guide you through the necessary steps to build and fine-tune your first models.

Ready to close the gap between traditional ML and the next wave of AI innovation? Jump in and let's get started!

Who this course is for:

Data scientists with good background in traditional ML but lacking any knowledge in LLMs

From Traditional ML to LLMs

What you'll learn

Explore related topics

Course content

Introduction2 lectures • 3min

Requirements: What you need to bring from traditional ML?6 lectures • 9min

The Transformer Architecture6 lectures • 11min

Different tasks and evaluation metrics of LLMs3 lectures • 10min

Your First Practical LLM Project with Python4 lectures • 6min

Summary1 lecture • 1min

Requirements

Description

Who this course is for: