How karpathy/llm101n works
This repository documents a planned educational course, not an implemented software product; the README says the course is still being developed and the repository is archived. Its principal workflow is curricular: learners progress from language-model and neural-network fundamentals through training and inference to deployment, with a planned Storyteller web app as the outcome. The graph preserves the documented syllabus stages and major capabilities; all relationships represent course progression or stated goals, not inspected runtime calls.
Main entry points: README.md Bigram Model
Components 16
Model Foundations
- Bigram Model
README.md - Micrograd
README.md - N-gram Model
README.md
LLM Architecture
- Attention
README.md - Transformer
README.md - Tokenization
README.md
Training and Inference
- Optimization
README.md - Performance
README.md - Datasets
README.md - Inference
README.md - SFT Fine-tuning
README.md - RL Fine-tuning
README.md
Product and Extensions
- Deployment
README.md - Storyteller App
README.md - Multimodal
README.md
Other components
- Learner
Connections 15
- Learner → Bigram Model: starts course
README.md - Bigram Model → Micrograd: then studies
README.md - Micrograd → N-gram Model: then studies
README.md - N-gram Model → Attention: then studies
README.md - Attention → Transformer: then studies
README.md - Transformer → Tokenization: then studies
README.md - Tokenization → Optimization: then studies
README.md - Optimization → Performance: then studies
README.md - Performance → Datasets: then studies
README.md - Datasets → Inference: then studies
README.md - Inference → SFT Fine-tuning: then studies
README.md - SFT Fine-tuning → RL Fine-tuning: then studies
README.md - RL Fine-tuning → Deployment: then studies
README.md - Deployment → Storyteller App: targets app
README.md - Storyteller App → Multimodal: course continues
README.md