Authorship and AI disclosure#

This project was written by an AI under human direction. The split is:

Who

Idea

Mirza Arnaut (@arunoruto), who proposed a Vulkan target in numba/numba#10116

Direction and design decisions

Mirza Arnaut: a Vulkan target for Numba, built on Numba’s target extension API, with numba-cuda and Taichi as references

Implementation: code, tests, benchmarks, documentation

Claude (Anthropic’s Claude Fable 5.1, and Claude Opus 5.5 from 2 October 2026 on, running in Claude Code)

Review

Mirza Arnaut

What this means in practice:

  • Every line of code, every test and every page of this documentation was generated by a language model in October 2026. The human author steered the work, decided what should be built and reviewed the result, but did not write the code.

  • Many implementation-level choices (for example how arrays are represented, or which LLVM passes run) were made by the model while solving problems it ran into. They are explained in How it works and should be judged on their merits, not assumed to be best practice.

  • The benchmark numbers and hardware observations come from real runs on the author’s machine, not from the model’s memory. The reasons given for driver behaviour are inferences from those runs.

  • The project is alpha software. Treat it with the scepticism that unreviewed-by-third-parties, machine-written code deserves.

Relationship to Numba#

This is an independent project and is not endorsed by the Numba maintainers. Numba has an AI tools policy that requires contributions generated with an LLM to be attributed as such. If any part of this project is ever proposed to Numba or a related project, that policy applies and this disclosure has to accompany it.

Future contributions#

Contributions written with AI tools are welcome under the same terms: state which tool produced them, and have a human review them before they are merged.