Support `diffusers` and `stable-diffusion` (pretty much everything)
Browse files- convert.py +90 -34
convert.py
CHANGED
@@ -2,18 +2,18 @@ import argparse
|
|
2 |
import json
|
3 |
import os
|
4 |
import shutil
|
5 |
-
from tempfile import TemporaryDirectory
|
6 |
from collections import defaultdict
|
7 |
from inspect import signature
|
8 |
-
from
|
|
|
9 |
|
10 |
import torch
|
11 |
|
12 |
-
from huggingface_hub import CommitOperationAdd, HfApi, hf_hub_download
|
13 |
from huggingface_hub.file_download import repo_folder_name
|
|
|
14 |
from transformers import AutoConfig
|
15 |
from transformers.pipelines.base import infer_framework_load_model
|
16 |
-
from safetensors.torch import save_file
|
17 |
|
18 |
|
19 |
class AlreadyExists(Exception):
|
@@ -30,15 +30,18 @@ def shared_pointers(tensors):
|
|
30 |
failing.append(names)
|
31 |
return failing
|
32 |
|
|
|
33 |
def check_file_size(sf_filename: str, pt_filename: str):
|
34 |
sf_size = os.stat(sf_filename).st_size
|
35 |
pt_size = os.stat(pt_filename).st_size
|
36 |
|
37 |
if (sf_size - pt_size) / pt_size > 0.01:
|
38 |
-
raise RuntimeError(
|
|
|
39 |
- {sf_filename}: {sf_size}
|
40 |
- {pt_filename}: {pt_size}
|
41 |
-
"""
|
|
|
42 |
|
43 |
|
44 |
def rename(pt_filename: str) -> str:
|
@@ -53,15 +56,14 @@ def convert_multi(model_id: str, folder: str) -> List["CommitOperationAdd"]:
|
|
53 |
data = json.load(f)
|
54 |
|
55 |
filenames = set(data["weight_map"].values())
|
|
|
56 |
for filename in filenames:
|
57 |
-
|
58 |
-
loaded = torch.load(cached_filename)
|
59 |
-
sf_filename = rename(filename)
|
60 |
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
local_filenames.append(
|
65 |
|
66 |
index = os.path.join(folder, "model.safetensors.index.json")
|
67 |
with open(index, "w") as f:
|
@@ -71,17 +73,28 @@ def convert_multi(model_id: str, folder: str) -> List["CommitOperationAdd"]:
|
|
71 |
json.dump(newdata, f)
|
72 |
local_filenames.append(index)
|
73 |
|
74 |
-
operations = [
|
|
|
|
|
75 |
|
76 |
return operations
|
77 |
|
78 |
|
79 |
def convert_single(model_id: str, folder: str) -> List["CommitOperationAdd"]:
|
80 |
-
|
81 |
-
|
82 |
-
|
|
|
|
|
|
|
|
|
83 |
|
84 |
-
|
|
|
|
|
|
|
|
|
|
|
85 |
shared = shared_pointers(loaded)
|
86 |
for shared_weights in shared:
|
87 |
for name in shared_weights[1:]:
|
@@ -90,23 +103,45 @@ def convert_single(model_id: str, folder: str) -> List["CommitOperationAdd"]:
|
|
90 |
# For tensors to be contiguous
|
91 |
loaded = {k: v.contiguous() for k, v in loaded.items()}
|
92 |
|
93 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
94 |
|
95 |
-
check_file_size(local, filename)
|
96 |
|
97 |
-
|
98 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
99 |
|
100 |
def check_final_model(model_id: str, folder: str):
|
101 |
config = hf_hub_download(repo_id=model_id, filename="config.json")
|
102 |
shutil.copy(config, os.path.join(folder, "config.json"))
|
103 |
config = AutoConfig.from_pretrained(folder)
|
104 |
|
105 |
-
_, pt_model = infer_framework_load_model(model_id, config)
|
106 |
-
_, sf_model = infer_framework_load_model(folder, config)
|
107 |
|
108 |
-
|
109 |
-
|
|
|
110 |
|
111 |
pt_params = pt_model.state_dict()
|
112 |
sf_params = sf_model.state_dict()
|
@@ -134,7 +169,6 @@ def check_final_model(model_id: str, folder: str):
|
|
134 |
if "image" in sig.parameters:
|
135 |
kwargs["image"] = pixel_values
|
136 |
|
137 |
-
|
138 |
if torch.cuda.is_available():
|
139 |
pt_model = pt_model.cuda()
|
140 |
sf_model = sf_model.cuda()
|
@@ -146,6 +180,7 @@ def check_final_model(model_id: str, folder: str):
|
|
146 |
torch.testing.assert_close(sf_logits, pt_logits)
|
147 |
print(f"Model {model_id} is ok !")
|
148 |
|
|
|
149 |
def previous_pr(api: "HfApi", model_id: str, pr_title: str) -> Optional["Discussion"]:
|
150 |
try:
|
151 |
discussions = api.get_repo_discussions(repo_id=model_id)
|
@@ -156,7 +191,22 @@ def previous_pr(api: "HfApi", model_id: str, pr_title: str) -> Optional["Discuss
|
|
156 |
return discussion
|
157 |
|
158 |
|
159 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
160 |
pr_title = "Adding `safetensors` variant of this model"
|
161 |
info = api.model_info(model_id)
|
162 |
filenames = set(s.rfilename for s in info.siblings)
|
@@ -174,21 +224,27 @@ def convert(api: "HfApi", model_id: str, force: bool=False) -> Optional["CommitI
|
|
174 |
url = f"https://huggingface.co/{model_id}/discussions/{pr.num}"
|
175 |
new_pr = pr
|
176 |
raise AlreadyExists(f"Model {model_id} already has an open PR check out {url}")
|
177 |
-
elif
|
178 |
-
|
179 |
-
|
180 |
-
|
|
|
|
|
|
|
|
|
181 |
else:
|
182 |
-
|
183 |
|
184 |
if operations:
|
185 |
-
check_final_model(model_id, folder)
|
186 |
new_pr = api.create_commit(
|
187 |
repo_id=model_id,
|
188 |
operations=operations,
|
189 |
commit_message=pr_title,
|
190 |
create_pr=True,
|
191 |
)
|
|
|
|
|
|
|
192 |
finally:
|
193 |
shutil.rmtree(folder)
|
194 |
return new_pr
|
|
|
2 |
import json
|
3 |
import os
|
4 |
import shutil
|
|
|
5 |
from collections import defaultdict
|
6 |
from inspect import signature
|
7 |
+
from tempfile import TemporaryDirectory
|
8 |
+
from typing import Dict, List, Optional, Set
|
9 |
|
10 |
import torch
|
11 |
|
12 |
+
from huggingface_hub import CommitInfo, CommitOperationAdd, Discussion, HfApi, hf_hub_download
|
13 |
from huggingface_hub.file_download import repo_folder_name
|
14 |
+
from safetensors.torch import load_file, save_file
|
15 |
from transformers import AutoConfig
|
16 |
from transformers.pipelines.base import infer_framework_load_model
|
|
|
17 |
|
18 |
|
19 |
class AlreadyExists(Exception):
|
|
|
30 |
failing.append(names)
|
31 |
return failing
|
32 |
|
33 |
+
|
34 |
def check_file_size(sf_filename: str, pt_filename: str):
|
35 |
sf_size = os.stat(sf_filename).st_size
|
36 |
pt_size = os.stat(pt_filename).st_size
|
37 |
|
38 |
if (sf_size - pt_size) / pt_size > 0.01:
|
39 |
+
raise RuntimeError(
|
40 |
+
f"""The file size different is more than 1%:
|
41 |
- {sf_filename}: {sf_size}
|
42 |
- {pt_filename}: {pt_size}
|
43 |
+
"""
|
44 |
+
)
|
45 |
|
46 |
|
47 |
def rename(pt_filename: str) -> str:
|
|
|
56 |
data = json.load(f)
|
57 |
|
58 |
filenames = set(data["weight_map"].values())
|
59 |
+
local_filenames = []
|
60 |
for filename in filenames:
|
61 |
+
pt_filename = hf_hub_download(repo_id=model_id, filename=filename)
|
|
|
|
|
62 |
|
63 |
+
sf_filename = rename(pt_filename)
|
64 |
+
sf_filename = os.path.join(folder, sf_filename)
|
65 |
+
convert_file(pt_filename, sf_filename)
|
66 |
+
local_filenames.append(sf_filename)
|
67 |
|
68 |
index = os.path.join(folder, "model.safetensors.index.json")
|
69 |
with open(index, "w") as f:
|
|
|
73 |
json.dump(newdata, f)
|
74 |
local_filenames.append(index)
|
75 |
|
76 |
+
operations = [
|
77 |
+
CommitOperationAdd(path_in_repo=local.split("/")[-1], path_or_fileobj=local) for local in local_filenames
|
78 |
+
]
|
79 |
|
80 |
return operations
|
81 |
|
82 |
|
83 |
def convert_single(model_id: str, folder: str) -> List["CommitOperationAdd"]:
|
84 |
+
pt_filename = hf_hub_download(repo_id=model_id, filename="pytorch_model.bin")
|
85 |
+
|
86 |
+
sf_name = "model.safetensors"
|
87 |
+
sf_filename = os.path.join(folder, sf_name)
|
88 |
+
convert_file(pt_filename, sf_filename)
|
89 |
+
operations = [CommitOperationAdd(path_in_repo=sf_name, path_or_fileobj=sf_filename)]
|
90 |
+
return operations
|
91 |
|
92 |
+
|
93 |
+
def convert_file(
|
94 |
+
pt_filename: str,
|
95 |
+
sf_filename: str,
|
96 |
+
):
|
97 |
+
loaded = torch.load(pt_filename)
|
98 |
shared = shared_pointers(loaded)
|
99 |
for shared_weights in shared:
|
100 |
for name in shared_weights[1:]:
|
|
|
103 |
# For tensors to be contiguous
|
104 |
loaded = {k: v.contiguous() for k, v in loaded.items()}
|
105 |
|
106 |
+
dirname = sf_filename.rsplit(os.path.sep, 1)[0]
|
107 |
+
os.makedirs(dirname, exist_ok=True)
|
108 |
+
save_file(loaded, sf_filename, metadata={"format": "pt"})
|
109 |
+
check_file_size(sf_filename, pt_filename)
|
110 |
+
reloaded = load_file(sf_filename)
|
111 |
+
for k in loaded:
|
112 |
+
pt_tensor = loaded[k]
|
113 |
+
sf_tensor = reloaded[k]
|
114 |
+
if not torch.equal(pt_tensor, sf_tensor):
|
115 |
+
raise RuntimeError(f"The output tensors do not match for key {k}")
|
116 |
|
|
|
117 |
|
118 |
+
def create_diff(pt_infos: Dict[str, List[str]], sf_infos: Dict[str, List[str]]) -> str:
|
119 |
+
errors = []
|
120 |
+
for key in ["missing_keys", "mismatched_keys", "unexpected_keys"]:
|
121 |
+
pt_set = set(pt_infos[key])
|
122 |
+
sf_set = set(sf_infos[key])
|
123 |
+
|
124 |
+
pt_only = pt_set - sf_set
|
125 |
+
sf_only = sf_set - pt_set
|
126 |
+
|
127 |
+
if pt_only:
|
128 |
+
errors.append(f"{key} : PT warnings contain {pt_only} which are not present in SF warnings")
|
129 |
+
if sf_only:
|
130 |
+
errors.append(f"{key} : SF warnings contain {sf_only} which are not present in PT warnings")
|
131 |
+
return "\n".join(errors)
|
132 |
+
|
133 |
|
134 |
def check_final_model(model_id: str, folder: str):
|
135 |
config = hf_hub_download(repo_id=model_id, filename="config.json")
|
136 |
shutil.copy(config, os.path.join(folder, "config.json"))
|
137 |
config = AutoConfig.from_pretrained(folder)
|
138 |
|
139 |
+
_, (pt_model, pt_infos) = infer_framework_load_model(model_id, config, output_loading_info=True)
|
140 |
+
_, (sf_model, sf_infos) = infer_framework_load_model(folder, config, output_loading_info=True)
|
141 |
|
142 |
+
if pt_infos != sf_infos:
|
143 |
+
error_string = create_diff(pt_infos, sf_infos)
|
144 |
+
raise ValueError(f"Different infos when reloading the model: {error_string}")
|
145 |
|
146 |
pt_params = pt_model.state_dict()
|
147 |
sf_params = sf_model.state_dict()
|
|
|
169 |
if "image" in sig.parameters:
|
170 |
kwargs["image"] = pixel_values
|
171 |
|
|
|
172 |
if torch.cuda.is_available():
|
173 |
pt_model = pt_model.cuda()
|
174 |
sf_model = sf_model.cuda()
|
|
|
180 |
torch.testing.assert_close(sf_logits, pt_logits)
|
181 |
print(f"Model {model_id} is ok !")
|
182 |
|
183 |
+
|
184 |
def previous_pr(api: "HfApi", model_id: str, pr_title: str) -> Optional["Discussion"]:
|
185 |
try:
|
186 |
discussions = api.get_repo_discussions(repo_id=model_id)
|
|
|
191 |
return discussion
|
192 |
|
193 |
|
194 |
+
def convert_generic(model_id: str, folder: str, filenames: Set[str]) -> List["CommitOperationAdd"]:
|
195 |
+
operations = []
|
196 |
+
|
197 |
+
extensions = set([".bin", ".ckpt"])
|
198 |
+
for filename in filenames:
|
199 |
+
prefix, ext = os.path.splitext(filename)
|
200 |
+
if ext in extensions:
|
201 |
+
pt_filename = hf_hub_download(model_id, filename=filename)
|
202 |
+
sf_in_repo = f"{filename}.safetensors"
|
203 |
+
sf_filename = os.path.join(folder, sf_in_repo)
|
204 |
+
convert_file(pt_filename, sf_filename)
|
205 |
+
operations.append(CommitOperationAdd(path_in_repo=sf_in_repo, path_or_fileobj=sf_filename))
|
206 |
+
return operations
|
207 |
+
|
208 |
+
|
209 |
+
def convert(api: "HfApi", model_id: str, force: bool = False) -> Optional["CommitInfo"]:
|
210 |
pr_title = "Adding `safetensors` variant of this model"
|
211 |
info = api.model_info(model_id)
|
212 |
filenames = set(s.rfilename for s in info.siblings)
|
|
|
224 |
url = f"https://huggingface.co/{model_id}/discussions/{pr.num}"
|
225 |
new_pr = pr
|
226 |
raise AlreadyExists(f"Model {model_id} already has an open PR check out {url}")
|
227 |
+
elif info.library_name == "transformers":
|
228 |
+
if "pytorch_model.bin" in filenames:
|
229 |
+
operations = convert_single(model_id, folder)
|
230 |
+
elif "pytorch_model.bin.index.json" in filenames:
|
231 |
+
operations = convert_multi(model_id, folder)
|
232 |
+
else:
|
233 |
+
raise RuntimeError(f"Model {model_id} doesn't seem to be a valid pytorch model. Cannot convert")
|
234 |
+
check_final_model(model_id, folder)
|
235 |
else:
|
236 |
+
operations = convert_generic(model_id, folder, filenames)
|
237 |
|
238 |
if operations:
|
|
|
239 |
new_pr = api.create_commit(
|
240 |
repo_id=model_id,
|
241 |
operations=operations,
|
242 |
commit_message=pr_title,
|
243 |
create_pr=True,
|
244 |
)
|
245 |
+
print(f"Pr created at {new_pr.pr_url}")
|
246 |
+
else:
|
247 |
+
print("No files to convert")
|
248 |
finally:
|
249 |
shutil.rmtree(folder)
|
250 |
return new_pr
|