-
Notifications
You must be signed in to change notification settings - Fork 21
/
quantize.py
57 lines (47 loc) · 2.24 KB
/
quantize.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
import argparse
import pathlib
import sillm.utils as utils
from sillm.utils.quantization import quantize_files
from sillm.models.args import ModelArgs
from sillm.core.tokenizer import TransformerTokenizer, SentencePieceTokenizer
if __name__ == "__main__":
# Parse commandline arguments
parser = argparse.ArgumentParser(description="A simple CLI for generating text with SiLLM.")
parser.add_argument("input", type=str, help="The input model directory or file")
parser.add_argument("output", type=str, help="The output model directory or file")
parser.add_argument("--bits", type=int, default=4, help="Quantization bits")
parser.add_argument("--group_size", default=32, help="Quantization group size")
parser.add_argument("-v", "--verbose", default=1, action="count", help="Increase output verbosity")
args = parser.parse_args()
# Initialize logging
log_level = 40 - (10 * args.verbose) if args.verbose > 0 else 0
logger = utils.init_logger(log_level)
# Log commandline arguments
if log_level <= 10:
utils.log_arguments(args.__dict__)
input_path = pathlib.Path(args.input)
output_path = pathlib.Path(args.output)
quantization = {
"bits": args.bits,
"group_size": args.group_size
}
quantize_files(args.input, args.output, **quantization)
config_path = input_path / "config.json"
model_args = ModelArgs.load_file(config_path)
model_args.quantization = quantization
config_path = output_path / "config.json"
model_args.save_config(config_path)
logger.debug(f"Saved model config to {config_path}")
tokenizer = None
tokenizer_path = None
if (input_path / "tokenizer.model").exists():
tokenizer_path = input_path / "tokenizer.model"
tokenizer = SentencePieceTokenizer(str(tokenizer_path), model_args)
elif (input_path / "tokenizer.json").exists():
tokenizer_path = input_path / "tokenizer.json"
tokenizer = TransformerTokenizer(str(input_path), model_args)
if tokenizer is None:
logger.error(f"No tokenizer found in {input_path}")
tokenizer.save(str(output_path))
logger.debug(f"Saved tokenizer to {output_path}")
logger.info(f"Quantized model with group size {args.group_size} and {args.bits} bits")