Ilaria_RVC_MOD / model_handler.py
TheStinger's picture
Upload 11 files
2625edb verified
raw
history blame
7.26 kB
import torch
import numpy as np
import huggingface_hub
import zipfile
import os
from collections import OrderedDict
def model_info(model_path):
model = torch.load(model_path, map_location=torch.device('cpu'))
info = {
'config': model['config'],
'info': model['info'],
'epochs': model['info'].split('epoch')[0],
'sr': model['sr'],
'f0': model['f0'],
'size': model['size'] if 'size' in model['weight'] else 'fp32',
}
return info
def merge(path1, path2, alpha1, sr, f0, info, name, version):
try:
def extract(ckpt):
a = ckpt["model"]
opt = OrderedDict()
opt["weight"] = {}
for key in a.keys():
if "enc_q" in key:
continue
opt["weight"][key] = a[key]
return opt
ckpt1 = torch.load(path1, map_location="cpu")
ckpt2 = torch.load(path2, map_location="cpu")
cfg = ckpt1["config"]
if "model" in ckpt1:
ckpt1 = extract(ckpt1)
else:
ckpt1 = ckpt1["weight"]
if "model" in ckpt2:
ckpt2 = extract(ckpt2)
else:
ckpt2 = ckpt2["weight"]
if sorted(list(ckpt1.keys())) != sorted(list(ckpt2.keys())):
return "Fail to merge the models. The model architectures are not the same."
opt = OrderedDict()
opt["weight"] = {}
for key in ckpt1.keys():
# try:
if key == "emb_g.weight" and ckpt1[key].shape != ckpt2[key].shape:
min_shape0 = min(ckpt1[key].shape[0], ckpt2[key].shape[0])
opt["weight"][key] = (
alpha1 * (ckpt1[key][:min_shape0].float())
+ (1 - alpha1) * (ckpt2[key][:min_shape0].float())
).half()
else:
opt["weight"][key] = (
alpha1 * (ckpt1[key].float()) + (1 - alpha1) * (ckpt2[key].float())
).half()
# except:
# pdb.set_trace()
opt["config"] = cfg
"""
if(sr=="40k"):opt["config"] = [1025, 32, 192, 192, 768, 2, 6, 3, 0, "1", [3, 7, 11], [[1, 3, 5], [1, 3, 5], [1, 3, 5]], [10, 10, 2, 2], 512, [16, 16, 4, 4,4], 109, 256, 40000]
elif(sr=="48k"):opt["config"] = [1025, 32, 192, 192, 768, 2, 6, 3, 0, "1", [3, 7, 11], [[1, 3, 5], [1, 3, 5], [1, 3, 5]], [10,6,2,2,2], 512, [16, 16, 4, 4], 109, 256, 48000]
elif(sr=="32k"):opt["config"] = [513, 32, 192, 192, 768, 2, 6, 3, 0, "1", [3, 7, 11], [[1, 3, 5], [1, 3, 5], [1, 3, 5]], [10, 4, 2, 2, 2], 512, [16, 16, 4, 4,4], 109, 256, 32000]
"""
opt["sr"] = sr
opt["f0"] = 1 if f0 == "Yes" else 0
opt["version"] = version
opt["info"] = info
torch.save(opt, "models/" + name + ".pth")
return "models/" + name + ".pth"
except:
return "Fail to merge the models. The model architectures are not the same." # <- L if u see this u suck
def model_quant(model_path, size):
"""
Quantize the model to a lower precision. - this is the floating point version
Args:
model_path: str, path to the model file
size: str, one of ["fp2", "fp4", "fp8", "fp16"]
Returns:
str, message indicating the success of the operation
"""
size_options = ["fp2", "fp4", "fp8", "fp16"]
if size not in size_options:
raise ValueError(f"Size must be one of {size_options}")
model_base = torch.load(model_path, map_location=torch.device('cpu'))
model = model_base['weight']
#model = json.loads(json.dumps(model))
if size == "fp16":
for key in model.keys():
model[key] = model[key].half() # 16-bit floating point
elif size == "fp8":
for key in model.keys():
model[key] = model[key].half().half() # 8-bit floating point <- this is the most common one
elif size == "fp4":
for key in model.keys():
model[key] = model[key].half().half().half() # 4-bit floating point <- ok maybe you're mentally ill if you choose this (very low precision)
elif size == "fp2":
for key in model.keys():
model[key] = model[key].half().half().half().half() # 2-bit floating point <- if you choose this you're a fucking dickhead coming
print(model_path)
output_path = model_path.split('.pth')[0] + f'_{size}.pth'
output_style = {
'weight': model,
'config': model_base['config'],
'info': model_base['info'],
'sr': model_base['sr'],
'f0': model_base['f0'],
'credits': f"Quantized to {size} precision, using Ilaria RVC, (Mikus's script)",
"size": size
}
torch.save(output_style, output_path)
#AmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithraxAmerithrax
# our data isnt safe anymore currently typing this and there is a 100% chance that it'll be stolen and used for training another fucking dogshit language model by a horrible company like openai
# i say this as a person who communicates with microsoft and i will stop mentioning this as they're so closely tied together nowadays
# as fred durst has said - "That's your best friend and your worst enemy - your own brain." - keep your shit local and never trust scumbag companies even if they make the models oss - they're stealing data
# this is probably the only rant i'll have in this entire space and i put it in a notable spot
return "Model quantized successfully" # <- enjoy this fucking hot shit that looks like a steaming turd paired with skibidi toilet and the unibomber
def upload_model(repo, pth, index, token):
"""
Upload a model to the Hugging Face Hub
Args:
repo: str, the name of the repository
pth: str, path to the model file
index: str, the index of the model in the repository
token: str, the API token
Returns:
str, message indicating the success of the operation
"""
readme = f"""
# {repo}
This is a model uploaded by Ilaria RVC, using Mikus's script.
"""
repo_name = repo.split('/')[1]
with zipfile.ZipFile(f'{repo_name}.zip', 'w') as zipf:
zipf.write(pth, os.path.basename(pth))
zipf.write(index, os.path.basename(index))
zipf.writestr('README.md', readme)
huggingface_hub.HfApi().create_repo(token=token, name=repo, exist_ok=True)
huggingface_hub.HfApi().upload_file(token=token, path=f'{repo.split("/")[1]}.zip', repo_id=repo)
os.remove(f'{repo.split("/")[1]}.zip')
return "Model uploaded successfully"