Qwodel
CLI Reference

Quantize a model from the command line.

qwodel quantize MODEL_PATH [OPTIONS]

Arguments

ArgumentDescriptionRequired
MODEL_PATHPath to the input model (local directory, .gguf file, or HuggingFace model ID).Yes

Options

OptionShortTypeDefaultDescription
--backend-bstrRequiredBackend to use: awq, gguf, coreml.
--format-fstrRequiredQuantization format (e.g., int4, Q4_K_M, float16).
--output-dir-ostr./quantized_modelsDirectory to save the output.
--verbose-vflagFalseEnable verbose/debug logging.

Examples

# GGUF Q4_K_M (recommended default)
qwodel quantize ./llama-3 -b gguf -f Q4_K_M -o ./output

# AWQ INT4 (GPU)
qwodel quantize ./gemma -b awq -f int4 -o ./output

# CoreML INT8 (Apple)
qwodel quantize ./my-model -b coreml -f int8_linear -o ./output

# From HuggingFace Hub
qwodel quantize meta-llama/Llama-2-7b-hf -b gguf -f Q4_K_M -o ./output