Candle

Part I: Foundations

1. Neural Networks Without the Magic

Why metaphors aren't enough. What this book is, who it's for, and what you'll build by the end.

2. History of Neural Networks

The failures and breakthroughs that shaped the field.

3. Rust for Neural Networks

Rust's explicitness as a teaching tool.

4. Tensors: The Foundation

The fundamental data structure of deep learning.

Part II: Building Blocks

5. Building Your First Neural Network

A complete implementation from scratch.

6. Loss Functions and Optimizers

How networks measure wrongness and minimize it.

7. Backpropagation from Scratch

The algorithm that makes learning possible. Implement it yourself,

8. Activation Functions

The non-linearity that enables complex patterns.

9. The Learning Rate

The most important hyperparameter. Too high and you overshoot; too low and you stall.

Part III: Architectures

10. Convolutional Neural Networks

Image processing through sliding windows of multiplication.

11. Recurrent Neural Networks

Processing sequences with memory. The Elman architecture

12. LSTM: Next Token Prediction

Building a character-level language model. Predict the next letter, generate text.

Part IV: Transformers

13. Tokenizers

Turning text into numbers. Byte-pair encoding, the bridge between words and vectors.

14. Token Embeddings

Turning numbers into vectors. How meaning gets encoded

15. Self-Attention

The core mechanism of modern AI. Queries, keys, values—attention as matrix

16. Transformer

Putting it all together. Build a complete transformer

Appendices

Appendix A: Pretrained Models in Candle

Using Hugging Face models without starting from scratch.

Appendix B: Tensor Shape Error Reference

Debugging the most common mistakes. Shape mismatches decoded and resolved.

Appendix C: Inference Optimizations

Making models faster on laptops. Practical speedups without a GPU.