Nanbeige4.2-3B MLX quantization
================================

This repository contains a derivative of `Nanbeige/Nanbeige4.2-3B` (Apache-2.0).

Modifications by Jishnu Venugopal (2026):
  * Converted from the original PyTorch checkpoint to Apple's MLX format.
  * Quantized to 4-bit (group size 64) for Apple Silicon.

The original model and its Apache-2.0 license are retained (see LICENSE).
Upstream notices and the full license text are unchanged. This derivative is
redistributed under the same Apache-2.0 license.

The MLX model definition (`nanbeige.py`) is MIT-licensed source code from the
`nanbeige-mlx` project; the weights themselves remain Apache-2.0.