Local model quantization

Create smaller GGUF models with a safe preview-first workflow.

arka quantize model-f16.gguf --output model-q4.gguf --type Q4_K_M
arka quantize model-f16.gguf --output model-q4.gguf --type Q4_K_M --run

The first command previews the llama-quantize command. --run is required to execute it, and the output directory is created automatically. Install a llama.cpp build exposing llama-quantize before running conversions.