Doomformer
A community experiment that compiles integer neural networks into ZDoom maps, using game geometry and voodoo dolls for arithmetic.
What happened lately
Doomformer timeline
Doomformer turns supported integer neural networks into ZDoom maps executed through ViZDoom with Freedoom assets. Sector lighting and geometry store values; moving voodoo dolls trigger arithmetic. Python builds maps, writes inputs, advances the simulation and reads results for comparison with a separate NumPy integer reference.
The English documentation describes Python 3.11 or later on Linux and macOS, installation from the GitHub repository, and a first-run ACC build requiring development tools and network access. Local inference does not require a model-provider API key. The code and documentation are MIT licensed; bundled data and game assets have separate licenses.
The release feed dates v0.7.1 to October 28, 2026. A new Hacker News discussion appeared October 29. Neither that discussion nor this discovery date establishes the project’s original launch date. AITrending has not reproduced the author’s demonstrations or performance measurements.