Deep learning basics: how a model learns a function from data
Neurons, layers, loss, gradients and backpropagation. Teach a small network XOR and see, briefly, how deep learning learns.
Enough if you can read Python
EXPLORE ARTICLES
Articles & dev notes
Implementation guides, design thinking and troubleshooting. Find them by context and code.
ARTICLES & NOTES
9 ARTICLESArticles & dev notes
How LLMs work: predicting the next token
Tokens, probabilities, temperature and the context window. Run a toy model to see how an LLM builds an answer, and why it gets things wrong.
Enough if you can read JavaScript
A look inside the Transformer: computing attention by hand
The Transformer is what computes next-token probabilities inside an LLM. See its overall structure, and compute its core, attention, in a small piece of code.
For readers of “How LLMs work”
LLM models at a glance: kinds and how to choose
Cloud APIs and open models, size, MoE, thinking, multimodal and roles. A brief map of how to sort the many LLM models and how to pick one.
Enough if you have used an LLM
Local LLMs: running a model on your own machine
What a local LLM is and how it differs from a cloud API. Estimate memory, understand quantization, run a model with Ollama, and know where it fits.
Enough if you can use a terminal
Running local LLMs seriously: big models and the KV cache
How to size the memory beyond the weights (the KV cache), split a model across GPUs, handle MoE, serve several users, and choose a serving tool.
For readers of the local LLM introduction
Build a system that answers questions about your documents (local LLM × RAG)
Split documents, search for the relevant parts, and hand them to a local LLM to answer. Build a question-answering system for internal documents in dependency-free Python.
Enough if you can read Python
Running internal-document Q&A in production: permissions, updates, evaluation and operations
Turn document Q&A (RAG) into a production system many people use. Incremental ingestion, permission filtering, better retrieval, evaluation and operations.
For readers of the basic RAG article
Trying LLMs on Colab: run a model on a free GPU
Even without a GPU in your own PC, Google Colab's free GPU can run an LLM. Two ways to run one (Ollama and Transformers), what you may use it for, what is prohibited, and how to size the memory.
Enough if you can run Python code in a notebook
