-
-
Notifications
You must be signed in to change notification settings - Fork 119
/
GenericTrainer.py
398 lines (313 loc) · 15.1 KB
/
GenericTrainer.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
import gc
import json
import os
import shutil
import subprocess
import sys
import time
from contextlib import nullcontext
from datetime import datetime
from pathlib import Path
from typing import Callable
import torch
from PIL.Image import Image
from torch import nn
from torch.cuda.amp import GradScaler
from torch.nn import Parameter
from torch.optim import Optimizer
from torch.utils.tensorboard import SummaryWriter
from torchvision.transforms.functional import pil_to_tensor
from tqdm import tqdm
from modules.dataLoader.MgdsStableDiffusionFineTuneDataLoader import MgdsStableDiffusionFineTuneDataLoader
from modules.model.BaseModel import BaseModel
from modules.modelLoader.BaseModelLoader import BaseModelLoader
from modules.modelSampler.BaseModelSampler import BaseModelSampler
from modules.modelSaver.BaseModelSaver import BaseModelSaver
from modules.modelSetup.BaseModelSetup import BaseModelSetup
from modules.trainer.BaseTrainer import BaseTrainer
from modules.util import path_util, create
from modules.util.TrainProgress import TrainProgress
from modules.util.args.TrainArgs import TrainArgs
from modules.util.callbacks.TrainCallbacks import TrainCallbacks
from modules.util.commands.TrainCommands import TrainCommands
from modules.util.enum.DataType import DataType
from modules.util.enum.ModelFormat import ModelFormat
from modules.util.enum.TimeUnit import TimeUnit
class GenericTrainer(BaseTrainer):
model_loader: BaseModelLoader
model_setup: BaseModelSetup
data_loader: MgdsStableDiffusionFineTuneDataLoader
model_saver: BaseModelSaver
model_sampler: BaseModelSampler
model: BaseModel
optimizer: Optimizer
previous_sample_time: float
sample_queue: list[Callable]
parameters: list[Parameter]
tensorboard_subprocess: subprocess.Popen
tensorboard: SummaryWriter
def __init__(self, args: TrainArgs, callbacks: TrainCallbacks, commands: TrainCommands):
super(GenericTrainer, self).__init__(args, callbacks, commands)
tensorboard_log_dir = os.path.join(args.workspace_dir, "tensorboard")
os.makedirs(Path(tensorboard_log_dir).absolute(), exist_ok=True)
self.tensorboard = SummaryWriter(os.path.join(tensorboard_log_dir, self.__get_string_timestamp()))
if args.tensorboard:
tensorboard_executable = os.path.join(os.path.dirname(sys.executable), "tensorboard")
self.tensorboard_subprocess = subprocess.Popen(
[
tensorboard_executable,
"--logdir",
tensorboard_log_dir,
"--port",
"6006",
"--samples_per_plugin=images=100"
]
)
self.one_step_trained = False
def start(self):
if self.args.clear_cache_before_training and self.args.latent_caching:
self.__clear_cache()
if self.args.train_dtype.enable_tf():
torch.backends.cuda.matmul.allow_tf32 = True
torch.backends.cudnn.allow_tf32 = True
self.model_loader = self.create_model_loader()
self.model_setup = self.create_model_setup()
self.callbacks.on_update_status("loading the model")
self.model = self.model_loader.load(
self.args.model_type,
self.args.weight_dtype.torch_dtype(),
self.args.base_model_name,
self.args.extra_model_name
)
self.callbacks.on_update_status("running model setup")
self.model_setup.setup_train_device(self.model, self.args)
self.model_setup.setup_model(self.model, self.args)
self.model_setup.setup_eval_device(self.model)
self.__gc()
self.callbacks.on_update_status("creating the data loader/caching")
self.data_loader = self.create_data_loader(
self.model, self.model.train_progress
)
self.model_saver = self.create_model_saver()
self.model_sampler = self.create_model_sampler(self.model)
self.previous_sample_time = -1
self.sample_queue = []
self.parameters = list(self.model_setup.create_parameters(self.model, self.args))
def __gc(self):
torch.cuda.synchronize()
gc.collect()
torch.cuda.empty_cache()
def __get_string_timestamp(self):
return datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
def __clear_cache(self):
print(
f'Clearing cache directory {self.args.cache_dir}! '
f'You can disable this if you want to continue using the same cache.'
)
if os.path.isdir(self.args.cache_dir):
for filename in os.listdir(self.args.cache_dir):
path = os.path.join(self.args.cache_dir, filename)
if os.path.isdir(path) and filename.startswith('epoch-'):
shutil.rmtree(path)
def __enqueue_sample_during_training(self, fun: Callable):
self.sample_queue.append(fun)
def __execute_sample_during_training(self):
for fun in self.sample_queue:
fun()
self.sample_queue = []
def __sample_loop(self, train_progress: TrainProgress, sample_definitions: list[dict], folder_postfix: str = ""):
for i, sample_definition in enumerate(sample_definitions):
safe_prompt = path_util.safe_filename(sample_definition['prompt'])
sample_dir = os.path.join(
self.args.workspace_dir,
"samples",
f"{str(i)} - {safe_prompt}{folder_postfix}",
)
sample_path = os.path.join(
sample_dir,
f"{self.__get_string_timestamp()}-training-sample-{train_progress.global_step}-{train_progress.epoch}-{train_progress.epoch_step}.png"
)
def on_sample(image: Image):
self.tensorboard.add_image(f"sample{str(i)} - {safe_prompt}", pil_to_tensor(image),
train_progress.global_step)
self.callbacks.on_sample(image)
self.model_sampler.sample(
prompt=sample_definition["prompt"],
resolution=(sample_definition["height"], sample_definition["width"]),
seed=sample_definition["seed"],
destination=sample_path,
text_encoder_layer_skip=self.args.text_encoder_layer_skip,
force_last_timestep=self.args.rescale_noise_scheduler_to_zero_terminal_snr,
on_sample=on_sample,
)
self.__gc()
def __sample_during_training(self, train_progress: TrainProgress, sample_definitions: list[dict] = None):
self.__gc()
self.callbacks.on_update_status("sampling")
self.model_setup.setup_eval_device(self.model)
if not sample_definitions:
with open(self.args.sample_definition_file_name, 'r') as f:
sample_definitions = json.load(f)
if self.model.ema:
self.model.ema.copy_ema_to(self.parameters, store_temp=True)
self.__sample_loop(train_progress, sample_definitions)
if self.model.ema:
self.model.ema.copy_temp_to(self.parameters)
# ema-less sampling, if an ema model exists
if self.model.ema:
self.__sample_loop(train_progress, sample_definitions, " - no-ema")
self.model_setup.setup_train_device(self.model, self.args)
self.__gc()
def backup(self):
self.__gc()
self.callbacks.on_update_status("creating backup")
backup_path = os.path.join(self.args.workspace_dir, "backup", self.__get_string_timestamp())
print("Creating Backup " + backup_path)
try:
self.model_saver.save(
self.model,
self.args.model_type,
ModelFormat.INTERNAL,
backup_path,
torch.float32
)
except:
print("Could not save backup. Check your disk space!")
try:
if os.path.isdir(backup_path):
shutil.rmtree(backup_path)
except:
print("Could not delete partial backup")
pass
self.model_setup.setup_train_device(self.model, self.args)
self.__gc()
def __needs_sample(self, train_progress: TrainProgress):
return self.action_needed("sample", self.args.sample_after, self.args.sample_after_unit, train_progress)
def __needs_backup(self, train_progress: TrainProgress):
return self.action_needed(
"backup", self.args.backup_after, self.args.backup_after_unit, train_progress, start_at_zero=False
)
def __needs_gc(self, train_progress: TrainProgress):
return self.action_needed("gc", 5, TimeUnit.MINUTE, train_progress, start_at_zero=False)
def __is_update_step(self, train_progress: TrainProgress) -> bool:
return self.action_needed(
"update_step", self.args.gradient_accumulation_steps, TimeUnit.STEP, train_progress, start_at_zero=False
)
def train(self):
train_device = torch.device(self.args.train_device)
train_progress = self.model.train_progress
if self.args.only_cache:
self.callbacks.on_update_status("caching")
self.model_setup.setup_eval_device(self.model)
self.__gc()
cached_epochs = [False] * self.args.latent_caching_epochs
for epoch in tqdm(range(train_progress.epoch, self.args.epochs, 1), desc="epoch"):
if not cached_epochs[epoch % self.args.latent_caching_epochs]:
self.data_loader.ds.start_next_epoch()
cached_epochs[epoch % self.args.latent_caching_epochs] = True
return
lr_scheduler = create.create_lr_scheduler(
optimizer=self.model.optimizer,
learning_rate_scheduler=self.args.learning_rate_scheduler,
warmup_steps=self.args.learning_rate_warmup_steps,
num_cycles=self.args.learning_rate_cycles,
max_epochs=self.args.epochs,
approximate_epoch_length=self.data_loader.ds.approximate_length(),
global_step=train_progress.global_step
)
if self.args.train_dtype.enable_loss_scaling() and self.args.weight_dtype == DataType.FLOAT_32:
scaler = GradScaler()
else:
scaler = None
# False if the model gradients are all None, True otherwise
# This is used to schedule sampling only when the gradients don't take up any space
has_gradient = False
accumulated_loss = 0.0
for epoch in tqdm(range(train_progress.epoch, self.args.epochs, 1), desc="epoch"):
self.callbacks.on_update_status("starting epoch/caching")
self.model_setup.setup_eval_device(self.model)
self.__gc()
self.data_loader.ds.start_next_epoch()
self.model_setup.setup_train_device(self.model, self.args)
self.__gc()
current_epoch_length = len(self.data_loader.dl) + train_progress.epoch_step
for epoch_step, batch in enumerate(tqdm(self.data_loader.dl, desc="step")):
if self.__needs_sample(train_progress):
self.__enqueue_sample_during_training(
lambda: self.__sample_during_training(train_progress)
)
if self.__needs_gc(train_progress):
self.__gc()
sample_command = self.commands.get_and_reset_sample_command()
if sample_command:
self.__enqueue_sample_during_training(
lambda: self.__sample_during_training(train_progress, [sample_command])
)
if not has_gradient:
self.__execute_sample_during_training()
if self.__needs_backup(train_progress):
self.backup()
self.callbacks.on_update_status("training")
if self.args.train_dtype.enable_mixed_precision() and self.args.weight_dtype == DataType.FLOAT_32:
forward_context = torch.autocast(train_device.type, dtype=self.args.train_dtype.torch_dtype())
else:
forward_context = nullcontext()
with forward_context:
model_output_data = self.model_setup.predict(self.model, batch, self.args, train_progress)
loss = self.model_setup.calculate_loss(self.model, batch, model_output_data, self.args)
loss = loss / self.args.gradient_accumulation_steps
if scaler:
scaler.scale(loss).backward()
else:
loss.backward()
has_gradient = True
accumulated_loss += loss.item()
if self.__is_update_step(train_progress):
if scaler:
scaler.unscale_(self.model.optimizer)
nn.utils.clip_grad_norm_(self.parameters, 1)
scaler.step(self.model.optimizer)
scaler.update()
else:
nn.utils.clip_grad_norm_(self.parameters, 1)
self.model.optimizer.step()
self.model.optimizer.zero_grad(set_to_none=True)
has_gradient = False
lr_scheduler.step()
self.tensorboard.add_scalar(
"learning_rate", lr_scheduler.get_last_lr()[0], train_progress.global_step
)
self.tensorboard.add_scalar("loss", accumulated_loss, train_progress.global_step)
accumulated_loss = 0.0
self.model_setup.after_optimizer_step(self.model, self.args, train_progress)
if self.model.ema:
self.model.ema.step(
self.parameters,
train_progress.global_step // self.args.gradient_accumulation_steps
)
self.one_step_trained = True
train_progress.next_step(self.args.batch_size)
self.callbacks.on_update_progress(train_progress, current_epoch_length, self.args.epochs)
if self.commands.get_stop_command():
return
train_progress.next_epoch()
self.callbacks.on_update_progress(train_progress, current_epoch_length, self.args.epochs)
if self.commands.get_stop_command():
return
def end(self):
if self.one_step_trained:
if self.args.backup_before_save:
self.backup()
self.callbacks.on_update_status("saving the final model")
if self.model.ema:
self.model.ema.copy_ema_to(self.parameters, store_temp=True)
self.model_saver.save(
model=self.model,
model_type=self.args.model_type,
output_model_format=self.args.output_model_format,
output_model_destination=self.args.output_model_destination,
dtype=self.args.output_dtype.torch_dtype()
)
self.tensorboard.close()
if self.args.tensorboard:
self.tensorboard_subprocess.kill()