A conceptual safetensors converter for RVC

convert_to_pth.py

import pathlib
import torch
from safetensors import safe_open
import argparse
import json

parser = argparse.ArgumentParser()
parser.add_argument("input", metavar="in", type=str, default="model.safetensors", help="Input .safetensors model path")
parser.add_argument("-m", "--metadata", required=False, metavar="meta", type=str, default="model.json", help="Metadata .json path")
parser.add_argument("output", metavar="out", nargs="?", default=None, type=str, help="Output .pth model path")
cmd_opts = parser.parse_args()

safetensors_path = cmd_opts.input
metadata_path = cmd_opts.metadata if cmd_opts.metadata is not None else safetensors_path.replace('.safetensors', '.json')
pth_path = cmd_opts.output if cmd_opts.output is not None else safetensors_path.replace('.safetensors', '.pth')

state_dict = {}
metadata = {}

with safe_open(safetensors_path, 'rb') as f:
    for k in f.keys():
        state_dict[k] = f.get_tensor(k)

    if not pathlib.Path(metadata_path).exists():
        with open(metadata_path, 'r') as f:
            metadata = json.load(f)
    else:
        meta = f.metadata()
        metadata['config'] = json.loads(meta.get('config'))
        metadata['sr'] = meta.get('sr')
        metadata['f0'] = 1 if meta.get('f0') == 'true' else 0
        metadata['version'] = meta.get('version', 'v1')
        metadata['info'] = meta.get('info')

metadata['weight'] = state_dict

torch.save(metadata, pth_path)

convert_to_safetensors.py

import torch
from safetensors.torch import save_file
import argparse
import json

parser = argparse.ArgumentParser()
parser.add_argument("input", metavar="in", type=str, default="model.pth", help="Input .pth model path")
parser.add_argument("output", metavar="out", nargs="?", default=None, type=str, help="Output .safetensors model path")
parser.add_argument("-m", "--metadata", metavar="meta", required=False, default=None, type=str, help="Output .json metadata path")
cmd_opts = parser.parse_args()

pth_path = cmd_opts.input
safetensors_path = cmd_opts.output if cmd_opts.output is not None else pth_path.replace('.pth', '.safetensors')
metadata_path = cmd_opts.metadata if cmd_opts.metadata is not None else safetensors_path.replace('.safetensors', '.json')

model = torch.load(pth_path)
state_dict = model.get('weight')
config = model.get('config')

metadata_safetensors = {
    'config': json.dumps(config),
    'sr': model.get('sr'),
    'f0': 'true' if model.get('f0') else 'false',
    'version': model.get('version', 'v1'),
    'info': model.get('info')
}

metadata = {k:v for k,v in model.items() if k != 'weight'}

save_file(state_dict, safetensors_path, metadata_safetensors)
with open(metadata_path, 'w') as f:
    json.dump(metadata, f)

put safetensors behind -m json

Can you write the instructions from scratch? I'm not sure because I'm a beginner

Try

python convert_to_pth.py -m metadata.json input_model.safetensors output_model.pth

or simply

python convert_to_pth.py -m metadata.json input_model.safetensors

can i know how to use this code. im 100% beginner

i just want to convert safetensors to pth file