Z3NN001/gemma4-21b-a4b-REAP-it-mlx-Q4

Gemma-4-21B-A4B-it-REAP (MLX 4-bit)

This is a high-speed 4-bit (INT4) conversion of 0xSero/gemma-4-21b-a4b-it-REAP optimized for Apple Silicon using the MLX framework.

Model Highlights

• Architecture: Gemma 4 Multimodal with Active Blocks (A4B).
• Optimization: REAP (Reasoning Enhancement and Active Pruning) for enhanced vision-language reasoning.
• Precision: 4-bit (Optimized for inference speed and low memory footprint).
• Size: ~12 GB (Recommended for any Apple Silicon Mac with 16GB+ RAM).

Conversion Details

The model was converted locally to ensure bit-perfect compatibility with the MLX ecosystem.

• Hardware: Mac Mini (M4) with 32GB RAM.
• Library: mlx-vlm using -q --q-bits 4.
• Format: Native MLX safetensors.

Usage

Installation

pip install mlx-vlm

Credits
Original Model: 0xSero/gemma-4-21b-a4b-it-REAP

MLX Conversion: Z3NN001