Boole-1.7B, a 1-bit LLM that outperforms Bonsai-1.7B
Source
precisionnumerics.ai
Author
d7chong
Date
Context
Boole-1.7B's advertised 247 MB refers to its model weights, not necessarily the file you download. Precision Numerics' official repository describes a one-bit version of Qwen's Qwen3-1.7B, with weights 13.9 times smaller than the higher-precision original.
The file size depends on the runtime format. The repository lists a 247.1 MB file for its GPU setup and a 292.1 MB GGUF file for llama.cpp, which supports CPU or GPU execution. It attributes the difference to how the embeddingA list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.Full definition → and output matrices are stored. The release therefore offers small model files for local use; its examples use the supported non-thinking mode.
Terms in this piece · Glossary
embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
quantization — Shrinking a model by storing its numbers less precisely — like rounding — so it runs faster and fits on smaller hardware, at a small quality cost.
