Upload lora-scripts/sd-scripts/tools/convert_diffusers20_original_sd.py with huggingface_hub
Browse files
lora-scripts/sd-scripts/tools/convert_diffusers20_original_sd.py
ADDED
@@ -0,0 +1,163 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# convert Diffusers v1.x/v2.0 model to original Stable Diffusion
|
2 |
+
|
3 |
+
import argparse
|
4 |
+
import os
|
5 |
+
import torch
|
6 |
+
from diffusers import StableDiffusionPipeline
|
7 |
+
|
8 |
+
import library.model_util as model_util
|
9 |
+
from library.utils import setup_logging
|
10 |
+
setup_logging()
|
11 |
+
import logging
|
12 |
+
logger = logging.getLogger(__name__)
|
13 |
+
|
14 |
+
def convert(args):
|
15 |
+
# 引数を確認する
|
16 |
+
load_dtype = torch.float16 if args.fp16 else None
|
17 |
+
|
18 |
+
save_dtype = None
|
19 |
+
if args.fp16 or args.save_precision_as == "fp16":
|
20 |
+
save_dtype = torch.float16
|
21 |
+
elif args.bf16 or args.save_precision_as == "bf16":
|
22 |
+
save_dtype = torch.bfloat16
|
23 |
+
elif args.float or args.save_precision_as == "float":
|
24 |
+
save_dtype = torch.float
|
25 |
+
|
26 |
+
is_load_ckpt = os.path.isfile(args.model_to_load)
|
27 |
+
is_save_ckpt = len(os.path.splitext(args.model_to_save)[1]) > 0
|
28 |
+
|
29 |
+
assert not is_load_ckpt or args.v1 != args.v2, "v1 or v2 is required to load checkpoint / checkpointの読み込みにはv1/v2指定が必要です"
|
30 |
+
# assert (
|
31 |
+
# is_save_ckpt or args.reference_model is not None
|
32 |
+
# ), f"reference model is required to save as Diffusers / Diffusers形式での保存には参照モデルが必要です"
|
33 |
+
|
34 |
+
# モデルを読み込む
|
35 |
+
msg = "checkpoint" if is_load_ckpt else ("Diffusers" + (" as fp16" if args.fp16 else ""))
|
36 |
+
logger.info(f"loading {msg}: {args.model_to_load}")
|
37 |
+
|
38 |
+
if is_load_ckpt:
|
39 |
+
v2_model = args.v2
|
40 |
+
text_encoder, vae, unet = model_util.load_models_from_stable_diffusion_checkpoint(
|
41 |
+
v2_model, args.model_to_load, unet_use_linear_projection_in_v2=args.unet_use_linear_projection
|
42 |
+
)
|
43 |
+
else:
|
44 |
+
pipe = StableDiffusionPipeline.from_pretrained(
|
45 |
+
args.model_to_load, torch_dtype=load_dtype, tokenizer=None, safety_checker=None, variant=args.variant
|
46 |
+
)
|
47 |
+
text_encoder = pipe.text_encoder
|
48 |
+
vae = pipe.vae
|
49 |
+
unet = pipe.unet
|
50 |
+
|
51 |
+
if args.v1 == args.v2:
|
52 |
+
# 自動判定する
|
53 |
+
v2_model = unet.config.cross_attention_dim == 1024
|
54 |
+
logger.info("checking model version: model is " + ("v2" if v2_model else "v1"))
|
55 |
+
else:
|
56 |
+
v2_model = not args.v1
|
57 |
+
|
58 |
+
# 変換して保存する
|
59 |
+
msg = ("checkpoint" + ("" if save_dtype is None else f" in {save_dtype}")) if is_save_ckpt else "Diffusers"
|
60 |
+
logger.info(f"converting and saving as {msg}: {args.model_to_save}")
|
61 |
+
|
62 |
+
if is_save_ckpt:
|
63 |
+
original_model = args.model_to_load if is_load_ckpt else None
|
64 |
+
key_count = model_util.save_stable_diffusion_checkpoint(
|
65 |
+
v2_model,
|
66 |
+
args.model_to_save,
|
67 |
+
text_encoder,
|
68 |
+
unet,
|
69 |
+
original_model,
|
70 |
+
args.epoch,
|
71 |
+
args.global_step,
|
72 |
+
None if args.metadata is None else eval(args.metadata),
|
73 |
+
save_dtype=save_dtype,
|
74 |
+
vae=vae,
|
75 |
+
)
|
76 |
+
logger.info(f"model saved. total converted state_dict keys: {key_count}")
|
77 |
+
else:
|
78 |
+
logger.info(
|
79 |
+
f"copy scheduler/tokenizer config from: {args.reference_model if args.reference_model is not None else 'default model'}"
|
80 |
+
)
|
81 |
+
model_util.save_diffusers_checkpoint(
|
82 |
+
v2_model, args.model_to_save, text_encoder, unet, args.reference_model, vae, args.use_safetensors
|
83 |
+
)
|
84 |
+
logger.info("model saved.")
|
85 |
+
|
86 |
+
|
87 |
+
def setup_parser() -> argparse.ArgumentParser:
|
88 |
+
parser = argparse.ArgumentParser()
|
89 |
+
parser.add_argument(
|
90 |
+
"--v1", action="store_true", help="load v1.x model (v1 or v2 is required to load checkpoint) / 1.xのモデルを読み込む"
|
91 |
+
)
|
92 |
+
parser.add_argument(
|
93 |
+
"--v2", action="store_true", help="load v2.0 model (v1 or v2 is required to load checkpoint) / 2.0のモデルを読み込む"
|
94 |
+
)
|
95 |
+
parser.add_argument(
|
96 |
+
"--unet_use_linear_projection",
|
97 |
+
action="store_true",
|
98 |
+
help="When saving v2 model as Diffusers, set U-Net config to `use_linear_projection=true` (to match stabilityai's model) / Diffusers形式でv2モデルを保存するときにU-Netの設定を`use_linear_projection=true`にする(stabilityaiのモデルと合わせる)",
|
99 |
+
)
|
100 |
+
parser.add_argument(
|
101 |
+
"--fp16",
|
102 |
+
action="store_true",
|
103 |
+
help="load as fp16 (Diffusers only) and save as fp16 (checkpoint only) / fp16形式で読み込み(Diffusers形式のみ対応)、保存する(checkpointのみ対応)",
|
104 |
+
)
|
105 |
+
parser.add_argument("--bf16", action="store_true", help="save as bf16 (checkpoint only) / bf16形式で保存する(checkpointのみ対応)")
|
106 |
+
parser.add_argument(
|
107 |
+
"--float", action="store_true", help="save as float (checkpoint only) / float(float32)形式で保存する(checkpointのみ対応)"
|
108 |
+
)
|
109 |
+
parser.add_argument(
|
110 |
+
"--save_precision_as",
|
111 |
+
type=str,
|
112 |
+
default="no",
|
113 |
+
choices=["fp16", "bf16", "float"],
|
114 |
+
help="save precision, do not specify with --fp16/--bf16/--float / 保存する精度、--fp16/--bf16/--floatと併用しないでくださ���",
|
115 |
+
)
|
116 |
+
parser.add_argument("--epoch", type=int, default=0, help="epoch to write to checkpoint / checkpointに記録するepoch数の値")
|
117 |
+
parser.add_argument(
|
118 |
+
"--global_step", type=int, default=0, help="global_step to write to checkpoint / checkpointに記録するglobal_stepの値"
|
119 |
+
)
|
120 |
+
parser.add_argument(
|
121 |
+
"--metadata",
|
122 |
+
type=str,
|
123 |
+
default=None,
|
124 |
+
help='モデルに保存されるメタデータ、Pythonの辞書形式で指定 / metadata: metadata written in to the model in Python Dictionary. Example metadata: \'{"name": "model_name", "resolution": "512x512"}\'',
|
125 |
+
)
|
126 |
+
parser.add_argument(
|
127 |
+
"--variant",
|
128 |
+
type=str,
|
129 |
+
default=None,
|
130 |
+
help="読む込むDiffusersのvariantを指定する、例: fp16 / variant: Diffusers variant to load. Example: fp16",
|
131 |
+
)
|
132 |
+
parser.add_argument(
|
133 |
+
"--reference_model",
|
134 |
+
type=str,
|
135 |
+
default=None,
|
136 |
+
help="scheduler/tokenizerのコピー元Diffusersモデル、Diffusers形式で保存するときに使用される、省略時は`runwayml/stable-diffusion-v1-5` または `stabilityai/stable-diffusion-2-1` / reference Diffusers model to copy scheduler/tokenizer config from, used when saving as Diffusers format, default is `runwayml/stable-diffusion-v1-5` or `stabilityai/stable-diffusion-2-1`",
|
137 |
+
)
|
138 |
+
parser.add_argument(
|
139 |
+
"--use_safetensors",
|
140 |
+
action="store_true",
|
141 |
+
help="use safetensors format to save Diffusers model (checkpoint depends on the file extension) / Duffusersモデルをsafetensors形式で保存する(checkpointは拡張子で自動判定)",
|
142 |
+
)
|
143 |
+
|
144 |
+
parser.add_argument(
|
145 |
+
"model_to_load",
|
146 |
+
type=str,
|
147 |
+
default=None,
|
148 |
+
help="model to load: checkpoint file or Diffusers model's directory / 読み込むモデル、checkpointかDiffusers形式モデルのディレクトリ",
|
149 |
+
)
|
150 |
+
parser.add_argument(
|
151 |
+
"model_to_save",
|
152 |
+
type=str,
|
153 |
+
default=None,
|
154 |
+
help="model to save: checkpoint (with extension) or Diffusers model's directory (without extension) / 変換後のモデル、拡張子がある場合はcheckpoint、ない場合はDiffusesモデルとして保存",
|
155 |
+
)
|
156 |
+
return parser
|
157 |
+
|
158 |
+
|
159 |
+
if __name__ == "__main__":
|
160 |
+
parser = setup_parser()
|
161 |
+
|
162 |
+
args = parser.parse_args()
|
163 |
+
convert(args)
|