safetensors_converter.py

import argparse
from pathlib import Path

import torch
from safetensors.torch import save_file


def convert(path: Path, half: bool = False, no_ema: bool = False):
    state_dict = torch.load(path, map_location="cpu")
    if "state_dict" in state_dict:
        state_dict = state_dict["state_dict"]

    to_remove = []
    for k, v in state_dict.items():
        if not isinstance(v, torch.Tensor):
            to_remove.append(k)
        elif no_ema and "ema" in k:
            to_remove.append(k)

    for k in to_remove:
        del state_dict[k]

    if half:
        state_dict = {k: v.half() for k, v in state_dict.items()}

    output_name = path.stem
    if no_ema:
        output_name += "-pruned"
    if half:
        output_name += "-fp16"
    output_path = path.parent / f"{output_name}.safetensors"
    save_file(state_dict, output_path.as_posix())


def main(path: str, half: bool = False, no_ema: bool = False):
    path_ = Path(path).resolve()

    if not path_.exists():
        raise ValueError(f"Invalid path: {path}")

    if path_.is_file():
        to_convert = [path_]
    else:
        to_convert = list(path_.glob("*.ckpt"))

    for file in to_convert:
        print(f"Converting... {file}")
        convert(file, half, no_ema)


def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument("path", type=str, help="Path to checkpoint file or directory.")
    parser.add_argument(
        "--half", action="store_true", help="Convert to half precision."
    )
    parser.add_argument("--no-ema", action="store_true", help="Ignore EMA weights.")
    return parser.parse_args()


if __name__ == "__main__":
    args = parse_args()
    main(args.path, args.half, args.no_ema)