Language Models From First Principles
Master the foundational mechanics of LLMs—from tokenization and attention mechanisms to training dynamics—and build intuitive, mathematical clarity on how modern language models operate from the ground up.
Created and taught by Bhim Prasad Adhikari
6
Lessons
6h
Learning
7
Checks
Why this path
Understanding before implementation
Dive deep into the mechanics of Large Language Models without skipping the essential math or code. This foundational course guides you step-by-step through byte-pair encoding (BPE) tokenization, vector embeddings, multi-head self-attention, loss functions, and pre-training/fine-tuning pipelines. Designed for AI engineers, developers, and researchers looking to move beyond black-box APIs and build a rock-solid, first-principles understanding of modern Generative AI.
What you will understand
- Dive deep into the mechanics of Large Language Models without skipping the essential math or code.
- This foundational course guides you step-by-step through byte-pair encoding (BPE) tokenization, vector embeddings, multi-head self-attention, loss functions, and pre-training/fine-tuning pipelines.
Before you begin
- • Dive deep into the mechanics of Large Language Models without skipping the essential math or code.
Curriculum
A guided sequence of ideas
Module 1
Foundations & Tokenization
Learn how text raw data is mapped to numerical tokens and build a Byte-Pair Encoding (BPE) tokenizer.
Module 2
Neural Network Building Blocks
Understand backpropagation and build a foundational micrograd engine from scratch.
Module 3
Transformer Architecture & Attention
Deep dive into self-attention, positional encoding, and building nanoGPT.