llama-cpp-python/examples/low_level_api/quantize.py at main · DoLife/llama-cpp-python · GitHub

25 lines (21 loc) · 906 Bytes

import os

import argparse

import llama_cpp

def main(args):

if not os.path.exists(fname_inp):

raise RuntimeError(f"Input file does not exist ({fname_inp})")

if os.path.exists(fname_out):

raise RuntimeError(f"Output file already exists ({fname_out})")

fname_inp = args.fname_inp.encode("utf-8")

fname_out = args.fname_out.encode("utf-8")

itype = args.itype

return_code = llama_cpp.llama_model_quantize(fname_inp, fname_out, itype)

if return_code != 0:

raise RuntimeError("Failed to quantize model")

if __name__ == "__main__":

parser = argparse.ArgumentParser()

parser.add_argument("fname_inp", type=str, help="Path to input model")

parser.add_argument("fname_out", type=str, help="Path to output model")

parser.add_argument("type", type=int, help="Type of quantization (2: q4_0, 3: q4_1)")

args = parser.parse_args()

main(args)

Expand file tree