Boole Inference

Making intelligence as accessible as electricity.

About

Watt Inference compresses large language models after training, reducing their size while preserving up to 90% of the original model's quality. The result is models that are smaller, cheaper, and faster to run without the steep quality loss typical of compression.

Why

Large models are powerful but expensive to run. For intelligence to work like a utility — available everywhere, to everyone — it needs to run efficiently on far less compute than it does today. That's the problem we're solving.

Status

Early stage. This repository will expand to include a compression toolkit, benchmark results, and usage examples.

Contact

founders@booleinference.com(twitter.com/boole_sh)

License

Other