Let's build GPT: from scratch, in code, spelled out.

Name: Let's build GPT: from scratch, in code, spelled out.
Uploaded: 2026-06-16T13:40:28.359Z
Duration: 1 h 56 min 20 s
Channel: Andrej Karpathy
Description: Andrej Karpathy builds a GPT model from scratch in code, following the 'Attention is All You Need' paper and OpenAI's GPT architecture. The video covers autoregressive language modeling, self-attention mechanisms, and practical implementation using PyTorch.

Andrej Karpathy16 June 2026Watch on YouTube

Description

We build a Generatively Pretrained Transformer (GPT), following the paper "Attention is All You Need" and OpenAI's GPT-2 / GPT-3. We talk about connections to ChatGPT, which has taken the world by storm. We watch GitHub Copilot, itself a GPT, help us write a GPT (meta :D!) . I recommend people watch the earlier makemore videos to get comfortable with the autoregressive language modeling framework and basics of tensors and PyTorch nn, which we take for granted in this video. Links: - Google colab for the video: https://colab.research.google.com/drive/1JMLa53HDuA-i7ZBmqV7ZnA3c_fvtXnx-?usp=sharing - GitHub repo for the video: https://github.com/karpathy/ng-video-lecture - Playlist of the whole Zero to Hero series so far: https://www.youtube.com/watch?v=VMj-3S1tku0&list=PLAqhIrjkxbuWI23v9cThsA9GvCAUhRvKZ - nanoGPT repo: https://github.com/karpathy/nanoGPT - my website: https://karpathy.ai - my twitter: https://twitter.com/karpathy - our Discord channel: https://discord.gg/3zy8kqD9Cp Supplementary links: - Attention is All You Need paper: https://arxiv.org/abs/1706.03762 - OpenAI GPT-3 paper: https://arxiv.org/abs/2005.14165 - OpenAI ChatGPT blog post: https://openai.com/blog/chatgpt/ - The GPU I'm training the model on is from Lambda GPU Cloud, I think the best and easiest way to spin up an on-demand GPU instance in the cloud that you can ssh to: https://lambdalabs.com . If you prefer to work in notebooks, I think the easiest path today is Google Colab. Suggested exercises: - EX1: The n-dimensional tensor mastery challenge: Combine the `Head` and `MultiHeadAttention` into one class that processes all the heads in parallel, treating the heads as another batch dimension (answer is in nanoGPT). - EX2: Train the GPT on your own dataset of choice! What other data could be fun to blabber on about? (A fun advanced suggestion if you like: train a GPT to do addition of two numbers, i.e. a+b=c. You may find it helpful to predict the digits of c in reverse order, as the typical addition algorithm (that you're hoping it learns) would proceed right to left too. You may want to modify the data loader to simply serve random problems and skip the generation of train.bin, val.bin. You may want to mask out the loss at the input positions of a+b that just specify the problem using y=-1 in the targets (see CrossEntropyLoss ignore_index). Does your Transformer learn to add? Once you have this, swole doge project: build a calculator clone in GPT, for all of +-*/. Not an easy problem. You may need Chain of Thought traces.) - EX3: Find a dataset that is very large, so large that you can't see a gap between train and val loss. Pretrain the transformer on this data, then initialize with that model and finetune it on tiny shakespeare with a smaller number of steps and lower learning rate. Can you obtain a lower validation loss by the use of pretraining? - EX4: Read some transformer papers and implement one additional feature or change that people seem to use. Does it improve the performance of your GPT? Chapters: 00:00:00 intro: ChatGPT, Transformers, nanoGPT, Shakespeare baseline language modeling, code setup 00:07:52 reading and exploring the data 00:09:28 tokenization, train/val split 00:14:27 data loader: batches of chunks of data 00:22:11 simplest baseline: bigram language model, loss, generation 00:34:53 training the bigram model 00:38:00 port our code to a script Building the "self-attention" 00:42:13 version 1: averaging past context with for loops, the weakest form of aggregation 00:47:11 the trick in self-attention: matrix multiply as weighted aggregation 00:51:54 version 2: using matrix multiply 00:54:42 version 3: adding softmax 00:58:26 minor code cleanup 01:00:18 positional encoding 01:02:00 THE CRUX OF THE VIDEO: version 4: self-attention 01:11:38 note 1: attention as communication 01:12:46 note 2: attention has no notion of space, operates over sets 01:13:40 note 3: there is no communication across batch dimension 01:14:14 note 4: encoder blocks vs. decoder blocks 01:15:39 note 5: attention vs. self-attention vs. cross-attention 01:16:56 note 6: "scaled" self-attention. why divide by sqrt(head_size) Building the Transformer 01:19:11 inserting a single self-attention block to our network 01:21:59 multi-headed self-attention 01:24:25 feedforward layers of transformer block 01:26:48 residual connections 01:32:51 layernorm (and its relationship to our previous batchnorm) 01:37:49 scaling up the model! creating a few variables. adding dropout Notes on Transformer 01:42:39 encoder vs. decoder vs. both (?) Transformers 01:46:22 super quick walkthrough of nanoGPT, batched multi-headed self-attention 01:48:53 back to ChatGPT, GPT-3, pretraining vs. finetuning, RLHF 01:54:32 conclusions Corrections: 00:57:00 Oops "tokens from the _future_ cannot communicate", not "past". Sorry! :) 01:20:05 Oops I should be using the head_size for the normalization, not C

What you'll learn

You learn how to build a GPT model step-by-step, starting from simple bigram models and progressing to full transformer architectures with self-attention.
Self-attention mechanisms are explained through matrix multiplication as weighted aggregation rather than loops, forming the foundation of modern language models.
Multi-headed attention, feedforward layers, residual connections, and layer normalization are essential building blocks that together form a complete transformer.
You implement everything practically in PyTorch and learn how positional encoding, dropout, and scaling affect model performance.

Frequently asked questions

What is the difference between attention and self-attention?

Self-attention works within a single sequence where tokens communicate with other tokens in the same sequence. With cross-attention, tokens from one sequence communicate with tokens from another sequence, which is used in encoder-decoder architectures.

Why do we split self-attention into multiple heads?

Multi-headed attention allows the model to learn different aspects of token relationships simultaneously. Each head can focus on different patterns, which increases the model's representational capacity.

What is the purpose of positional encoding in a transformer?

Positional encoding adds information about token order to the embeddings, since self-attention mechanisms have no inherent notion of position. This helps the model correctly interpret sequence and context.

How does autoregressive language modeling work in GPT?

GPT predicts the next token step-by-step based on all previous tokens. This is achieved through masking so tokens can only communicate with past tokens, not future ones, enabling training and generation.