llama.cpp/examples/low_level_api/quantize.py
2023-04-05 04:17:26 -04:00

25 lines
906 B
Python

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)