Skip to content

cloudcover95/BitNet-mlx

Repository files navigation

BitNet-mlx

Current Ecosystem State

Added vision feature quantization support (src/vision_quant.py).

This enables efficient 1.58-bit ternary processing for image/text tasks such as Instagram story zoom tag recognition.

Integrates directly with VisionTextEngine in JuniorHome for sovereign, low-power on-device vision inference on Apple Silicon.

The core BitNet-mlx library now supports both language and vision modalities with the same ultra-efficient ternary approach.

About

BitNet-mlx is the core math engine of JuniorCloud LLC's sovereign edge stack. Delivers MLX-native 1.58-bit ternary quantization (W \in \{-1, 0, 1\}) and combinatorial TDA in discrete latent space. Features AbsMean scaling, DynamicBitLinear layers, thermal routing, and O(N) design. Yields peak memory efficiency for LLM/VLM on M1-M4 (atm)

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages