Portfolio — Will Kurt

A Damn Fine Stable Diffusion Book

My most recent book with Manning Publications. Focuses on showing you how you can use open models to create incredible Generative AI images using Open Source tools. Available for purchase for early access now, expected print publication Q2 2026.

Generative AI Machine Learning Diffusion Open Source AI

Token Explorer

An easy-to-use interface for interactively exploring the sampling process of an LLM. Manually step through how an LLM (defaults to Qwen/Qwen2.5-0.5B) generates text, monitoring properties of your in-progress generation such as probability of tokens and entropy of tokens. Fun tool for understanding exactly how LLMs work during generation.

Python Machine Learning LLM Generative AI

Deeplearning.AI: Getting Structured LLM Output

A project I led at .txt working with Andrew Ng and his team at Deeplearning.ai to create a course teaching people how to get reliable output from their LLM. The course gives an in-depth overview of how to get predictable structure out of an LLM covering libraries like Outlines and Instructor and going into the details of constrained decoding.

LLM Generative AI

The Bunny B1: Powered by SmolLM2

A project using structured generation to build tool use from scratch in support of HuggingFace's release of the Smol-LM2 model. The demo shows how natural language inputs can be automatically transformed into tool/function calls, working well even with an extremely small model.

Open Source AI LLM Transformers

Coalesence

An article that Andrej Karpathy described as "Very cool", detailing how structure generation makes it possible to skip unnecessary steps in LLM inference. This technique, referred to as "Coalesence", allows for 2-5x speed ups in LLM inference.

LLM Research Generative AI Inference Writing

Improving Prompt Consistency with Structured Generations

Collaborative research project with HuggingFace's Leaderboards and Evals research team showing that structured generation allows more consistent results across evaluations as well as better results with fewer examples.

LLM Research Open Source AI Evaluations Writing

Beating GPT-4 with Open Source

Demonstrating that, with open source tools, small open weights models could compete with proprietary LLMs, like GPT-4, in standard function calling (aka "tool use") tasks as measured by the Berkeley Function Calling Leaderboard.

LLM Research Open Source AI Evaluations Writing

Linear Diffusion

An implementation of a diffusion model for generative AI using only linear models as the building blocks. Both an interesting experiment and a helpful guide for understanding the basic architecture of diffusion models.

Generative AI Diffusion Writing

Evenflow

An experimental library in creating self-organizing systems of agents. The python library makes it easy to describe a computational DAG which can self-organize and adapt to changes. In progress work explores automating concurrent execution.

Agents Python Research

Bayesian Statistics the Fun Way

A beginner friendly introduction to Bayesian Statistics. Over 60,000 copies sold worldwide, translated into at least 5 languages, and frequently in the top 20 for the "Data Mining" and "Probability & Statistics" categories on Amazon.

Bayesian Statistics R Writing

Get Programming with Haskell

An intermediate book helping experienced programmers understand the power of functional programming with Haskell. The book introduces fundamental concepts in functional programming and Haskell's advanced type system.

Haskell Functional Programming Writing

Count Bayesie

For the past 10 years I have run and maintained the Count Bayesie blog, writing accessible and intuitive articles on a range of topics in probability and statistics. At its peak the blog had 20,000 monthly visitors and in its life time has accrued well over 1,000,000 pages views. Posting is less frequent (largely because I'm writing other books), but this blog remains near and dear to my heart.

Bayesian Statistics Writing