Isaac Breen

Article

llguidance HATES This One Weird Grammar (Well-behaved grammars)

A tiny counterexample, a debugger, and a simple rule for avoiding awkward greedy-lexer boundaries.

Read

Article

Where constrained decoding time goes

GLRMask vs llguidance, token by token

Read

Article

Why llguidance is fast

llguidance skips dead token-trie subtrees and fills broad lexer regions with precomputed token slices.

Read

Article

Life after ~~parsing~~ grammar constrained generation

Why I spent years building a general grammar-constrained decoder when most people needed JSON.

Read

Article

(archived) The 112-hyphen token

Why one model token can hide a long lexer and parser execution, and how GLRMask compiles that work into a weighted automaton over GLR stacks.

Read

Library

2024–present

GLRMask

A Rust and Python library for fast grammar-constrained LLM decoding over JSON Schema and general context-free grammars.

Read

Article

llguidance HATES This One Weird Grammar

How llguidance's one-byte lexer lookahead can lose a valid token boundary.

Read

Article

Building a mini llguidance

Rebuilding llguidance's token-trie, lexer, and Earley-parser loop in miniature.

Read

Library

2026

A weighted graph-structured stack

A persistent compressed map from stacks to weights, extracted from glrmask.

Read

Interactive

A grammar-constrained model in your browser

TinyStories-1M runs locally; glrmask WebAssembly applies an exact JSON mask to its BPE logits before sampling.

Read

Project

2023–2025

ToastChat

A 2023 SvelteKit chat client with streamed responses, rich Markdown messages, and a live deployment.

Read

Project

2022–2024

einexpr

An experimental Python library for named tensor dimensions and Einstein-style array programming.

Read

Thesis

2019

Comparing Markov Chain Monte Carlo and Variational Methods for Bayesian Inference on Mixtures of Gamma Distributions

My 2019 honours thesis on variational inference for finite mixtures of gamma distributions.

Read