-
Notifications
You must be signed in to change notification settings - Fork 8
/
train_ldm_discrete.py
495 lines (396 loc) · 19.7 KB
/
train_ldm_discrete.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
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
import ml_collections
import torch
from torch import multiprocessing as mp
from uvit_datasets import get_dataset, multiscale_collate_fn, MultiscaleDistSampler, MultiscaleBatchSampler
from torchvision.utils import make_grid, save_image
import utils
import einops
from torch.utils._pytree import tree_map
import accelerate
from torch.utils.data import DataLoader
from tqdm.auto import tqdm
from dpm_solver_pp import NoiseScheduleVP, DPM_Solver
import tempfile
from tools.fid_score import calculate_fid_given_paths
from absl import logging
import builtins
import os
import wandb
import libs.autoencoder
import numpy as np
from functools import partial
def stable_diffusion_beta_schedule(linear_start=0.00085, linear_end=0.0120, n_timestep=1000):
_betas = (
torch.linspace(linear_start ** 0.5, linear_end ** 0.5, n_timestep, dtype=torch.float64) ** 2
)
return _betas.numpy()
def get_skip(alphas, betas):
N = len(betas) - 1
skip_alphas = np.ones([N + 1, N + 1], dtype=betas.dtype)
for s in range(N + 1):
skip_alphas[s, s + 1:] = alphas[s + 1:].cumprod()
skip_betas = np.zeros([N + 1, N + 1], dtype=betas.dtype)
for t in range(N + 1):
prod = betas[1: t + 1] * skip_alphas[1: t + 1, t]
skip_betas[:t, t] = (prod[::-1].cumsum())[::-1]
return skip_alphas, skip_betas
def stp(s, ts: torch.Tensor): # scalar tensor product
if isinstance(s, np.ndarray):
s = torch.from_numpy(s).type_as(ts)
extra_dims = (1,) * (ts.dim() - 1)
return s.view(-1, *extra_dims) * ts
def mos(a, start_dim=1, w=None): # mean of square
ans = a.pow(2).flatten(start_dim=start_dim)
if w is not None:
w = w.reshape(-1, 1)
ans = ans * w
return ans.mean(dim=-1)
class Schedule(object): # discrete time
def __init__(self, _betas, reweight_schedule=None, multi_times=1, device='cuda'):
r""" _betas[0...999] = betas[1...1000]
for n>=1, betas[n] is the variance of q(xn|xn-1)
for n=0, betas[0]=0
"""
self.reweight_schedule = reweight_schedule
self.multi_times = multi_times
self._betas = _betas
self.betas = np.append(0., _betas)
self.alphas = 1. - self.betas
self.N = len(_betas)
assert isinstance(self.betas, np.ndarray) and self.betas[0] == 0
assert isinstance(self.alphas, np.ndarray) and self.alphas[0] == 1
assert len(self.betas) == len(self.alphas)
# skip_alphas[s, t] = alphas[s + 1: t + 1].prod()
self.skip_alphas, self.skip_betas = get_skip(self.alphas, self.betas)
self.cum_alphas = self.skip_alphas[0] # cum_alphas = alphas.cumprod()
self.cum_betas = self.skip_betas[0]
self.snr = self.cum_alphas / self.cum_betas
logging.info(f'snr {self.snr}')
if self.reweight_schedule in ['upsampler', ]:
from mamba_attn_diff.models.upsample_guidance import get_tau
t_continuous = list(range(1, self.N + 1))
def snr_func(ts):
return torch.tensor(self.snr[ts])
adjusted_timesteps_indices = get_tau(
torch.tensor(self.snr[t_continuous]), snr_func, t_continuous,
m = 1./multi_times,
return_indices = True,
)
adjusted_timesteps = np.array([t_continuous[i] for i in adjusted_timesteps_indices])
self.adjusted_timesteps = adjusted_timesteps
print('adjusted_timesteps ', adjusted_timesteps )
if self.reweight_schedule == 'upsampler':
logging.info('upsampler')
self.snr = torch.tensor(self.snr).to(device)
def tilde_beta(self, s, t):
return self.skip_betas[s, t] * self.cum_betas[s] / self.cum_betas[t]
def sample(self, x0): # sample from q(xn|x0), where n is uniform
if self.reweight_schedule == 'upsampler':
select_id = np.random.choice( 2 , (len(x0),))
ori_n = np.random.choice( len(self.adjusted_timesteps) , (len(x0),))
n = self.adjusted_timesteps[ ori_n ]
n = n * select_id + ori_n * (1 - select_id)
eps = torch.randn_like(x0)
xn = stp(self.cum_alphas[n] ** 0.5, x0) + stp(self.cum_betas[n] ** 0.5, eps)
elif self.reweight_schedule == 'continuous':
n = np.random.rand(len(x0)) * self.N
eps = torch.randn_like(x0)
n_floor = np.floor(n).astype(np.int64)
n_ceil = np.ceil(n).astype(np.int64)
n_rate = (n - n_floor) / (n_ceil - n_floor + 1e-6)
inter_cum_alphas = self.cum_alphas[n_floor] + n_rate * (self.cum_alphas[n_ceil] - self.cum_alphas[n_floor])
inter_cum_betas = self.cum_betas[n_floor] + n_rate * (self.cum_betas[n_ceil] - self.cum_betas[n_floor])
xn = stp(inter_cum_alphas ** 0.5, x0) + stp(inter_cum_betas ** 0.5, eps)
else:
n = np.random.choice(list(range(1, self.N + 1)), (len(x0),))
eps = torch.randn_like(x0)
xn = stp(self.cum_alphas[n] ** 0.5, x0) + stp(self.cum_betas[n] ** 0.5, eps)
return torch.tensor(n, device=x0.device, dtype=x0.dtype), eps, xn.to(x0.dtype)
def __repr__(self):
return f'Schedule({self.betas[:10]}..., {self.N})'
def LSimple(x0, nnet, schedule, is_snr=False, **kwargs):
n, eps, xn = schedule.sample(x0) # n in {1, ..., 1000}
eps_pred = nnet(xn, n, **kwargs)
eps_pred = eps_pred.sample if not isinstance(eps_pred, torch.Tensor) else eps_pred
w = None
if is_snr:
n = n.to(torch.long).clamp(min=0, max=len(schedule.snr))
snr = schedule.snr[n]
if isinstance(snr, np.ndarray):
snr = torch.from_numpy(snr).type_as(x0).to(x0.device)
w = snr.clamp(max=5)
return mos(eps - eps_pred, w=w)
def train(config):
if config.get('benchmark', False):
torch.backends.cudnn.benchmark = True
torch.backends.cudnn.deterministic = False
if config.nnet.get('latent_size', False):
assert config.z_shape[-1] == config.nnet.latent_size
mp.set_start_method('spawn')
gradient_accumulation_steps = config.get('gradient_accumulation_steps', 1)
try:
from accelerate.accelerator import is_deepspeed_available
is_not_deepspeed_available = not is_deepspeed_available()
except ImportError:
is_not_deepspeed_available = False
if is_not_deepspeed_available:
logging.info('Using default accelerator')
accelerator = accelerate.Accelerator(
gradient_accumulation_steps = gradient_accumulation_steps,
)
else:
deepspeed_plugin = accelerate.DeepSpeedPlugin(zero_stage=2, gradient_accumulation_steps=gradient_accumulation_steps)
accelerator = accelerate.Accelerator(
deepspeed_plugin=deepspeed_plugin
)
logging.info('Using deepspeed accelerator')
device = accelerator.device
accelerate.utils.set_seed(config.seed, device_specific=True)
logging.info(f'Process {accelerator.process_index} using device: {device}')
config.mixed_precision = accelerator.mixed_precision
config = ml_collections.FrozenConfigDict(config)
assert config.train.batch_size % accelerator.num_processes == 0
mini_batch_size = config.train.batch_size // accelerator.num_processes
if accelerator.is_main_process:
os.makedirs(config.ckpt_root, exist_ok=True)
os.makedirs(config.sample_dir, exist_ok=True)
accelerator.wait_for_everyone()
if accelerator.is_main_process:
wandb.init(dir=os.path.abspath(config.workdir), project=f'uvit_{config.dataset.name}', config=config.to_dict(),
name=config.hparams, job_type='train', mode='offline')
utils.set_logger(log_level='info', fname=os.path.join(config.workdir, 'output.log'))
logging.info(config)
else:
utils.set_logger(log_level='error')
builtins.print = lambda *args: None
logging.info(f'Run on {accelerator.num_processes} devices')
dataset = get_dataset(**config.dataset)
assert os.path.exists(dataset.fid_stat), "Please download the fid stat file first {}".format(dataset.fid_stat)
train_dataset = dataset.get_split(split='train', labeled=config.train.mode == 'cond')
if isinstance(train_dataset.resolution, list) or isinstance(train_dataset.resolution, tuple):
data_sampler = MultiscaleDistSampler(
resolution=train_dataset.resolution,
mini_batch_size=mini_batch_size,
dataset=train_dataset,
shuffle=True,
num_replicas=accelerator.num_processes,
rank=accelerator.process_index,
)
batch_data_sampler = MultiscaleBatchSampler(
sampler=data_sampler,
resolution=train_dataset.resolution,
batch_size=mini_batch_size,
drop_last=True,
)
train_dataset_loader = DataLoader(
train_dataset, batch_size=mini_batch_size, drop_last=True,
sampler=data_sampler, #batch_sampler=batch_data_sampler,
num_workers=config.train.get('num_workers', 8), #8,
pin_memory=True,
persistent_workers=(True if config.train.get('num_workers', 8) > 0 else False),
)
else:
train_dataset_loader = DataLoader(
train_dataset, batch_size=mini_batch_size, drop_last=True, shuffle=True,
num_workers=config.train.get('num_workers', 8), #8,
pin_memory=True,
persistent_workers=(True if config.train.get('num_workers', 8) > 0 else False),
)
train_state = utils.initialize_train_state(config, device)
ckpt_path = train_state.resume(config.ckpt_root, exclude_accelerate=True)
if hasattr(train_state.nnet, 'enable_gradient_checkpointing') and config.get('gradient_checkpointing', False):
train_state.nnet.enable_gradient_checkpointing()
# nnet, nnet_ema, optimizer, train_dataset_loader = accelerator.prepare(
# train_state.nnet, train_state.nnet_ema, train_state.optimizer, train_dataset_loader)
nnet, optimizer, train_dataset_loader = accelerator.prepare(
train_state.nnet, train_state.optimizer, train_dataset_loader)
nnet_ema = train_state.nnet_ema.to(device).to(torch.float32)
dtype = torch.bfloat16
lr_scheduler = train_state.lr_scheduler
if ckpt_path is not None:
# train_state.resume(config.ckpt_root)
accelerator.load_state(ckpt_path)
autoencoder = libs.autoencoder.get_model(config.autoencoder.pretrained_path)
autoencoder.to(device)
autoencoder.requires_grad_(False)
@ torch.cuda.amp.autocast()
def encode(_batch):
return autoencoder.encode(_batch)
@ torch.cuda.amp.autocast()
def decode(_batch):
return autoencoder.decode(_batch)
def get_data_generator():
while True:
for data in tqdm(train_dataset_loader, disable=not accelerator.is_main_process, desc='epoch'):
yield data
data_generator = get_data_generator()
_betas = stable_diffusion_beta_schedule()
_schedule = Schedule(
_betas,
reweight_schedule=config.get('reweight_schedule', None),
multi_times=config.nnet.get('multi_times', 1),
device=device,
)
logging.info(f'use {_schedule}')
max_grad_norm = config.get('max_grad_norm', False)
logging.info(f'Using max_grad_norm={max_grad_norm}')
def train_step(_batch):
_metrics = dict()
optimizer.zero_grad()
y = _batch[1] if config.train.mode == 'cond' else None
_z = _batch[0] if config.train.mode == 'cond' else _batch
if 'nested_feature' in config.dataset.name:
_z = autoencoder.scale_factor * _z
elif 'feature' in config.dataset.name:
_z = autoencoder.sample(_z)
else:
_z = encode(_z)
_z = _z.to(dtype)
loss = LSimple(_z, nnet, _schedule, is_snr=config.train.get('is_snr', False), y=y)
_metrics['loss'] = accelerator.gather(loss.detach()).mean()
accelerator.backward(loss.mean())
if max_grad_norm:
params_to_clip = filter(lambda p: p.requires_grad, nnet.parameters())
accelerator.clip_grad_norm_(params_to_clip, max_grad_norm)
optimizer.step()
lr_scheduler.step()
train_state.ema_update(config.get('ema_rate', 0.9999))
train_state.step += 1
return dict(lr=train_state.optimizer.param_groups[0]['lr'], **_metrics)
def dpm_solver_sample(_n_samples, _sample_steps, **kwargs):
_z_init = torch.randn(_n_samples, *config.z_shape, device=device)
noise_schedule = NoiseScheduleVP(schedule='discrete', betas=torch.tensor(_betas, device=device).float())
def model_fn(x, t_continuous):
t = t_continuous * _schedule.N
eps_pre = nnet_ema(x, t, **kwargs)
eps_pre = eps_pre.sample if not isinstance(eps_pre, torch.Tensor) else eps_pre
return eps_pre
dpm_solver = DPM_Solver(model_fn, noise_schedule, predict_x0=True, thresholding=False)
_z = dpm_solver.sample(_z_init, steps=_sample_steps, eps=1. / _schedule.N, T=1.)
return decode(_z)
def eval_step(n_samples, sample_steps):
logging.info(f'eval_step: n_samples={n_samples}, sample_steps={sample_steps}'
f'mini_batch_size={config.sample.mini_batch_size}')
def sample_fn(_n_samples):
if config.train.mode == 'uncond':
kwargs = dict()
elif config.train.mode == 'cond':
kwargs = dict(y=dataset.sample_label(_n_samples, device=device))
else:
raise NotImplementedError
return dpm_solver_sample(_n_samples, sample_steps, **kwargs)
with tempfile.TemporaryDirectory() as temp_path:
path = config.sample.path or temp_path
if accelerator.is_main_process:
os.makedirs(path, exist_ok=True)
utils.sample2dir(accelerator, path, n_samples, config.sample.mini_batch_size, sample_fn, dataset.unpreprocess)
_fid = 0
if accelerator.is_main_process:
_fid = calculate_fid_given_paths((dataset.fid_stat, path))
logging.info(f'step={train_state.step} fid{n_samples}={_fid}')
with open(os.path.join(config.workdir, 'eval.log'), 'a') as f:
print(f'step={train_state.step} fid{n_samples}={_fid}', file=f)
wandb.log({f'fid{n_samples}': _fid}, step=train_state.step)
_fid = torch.tensor(_fid, device=device)
_fid = accelerator.reduce(_fid, reduction='sum')
return _fid.item()
logging.info(f'Start fitting, step={train_state.step}, mixed_precision={config.mixed_precision}')
step_fid = []
while train_state.step < config.train.n_steps:
nnet.train()
batch = tree_map(lambda x: x.to(device), next(data_generator))
metrics = train_step(batch)
nnet.eval()
if accelerator.is_main_process and train_state.step % config.train.log_interval == 0:
logging.info(utils.dct2str(dict(step=train_state.step, **metrics)))
logging.info(config.workdir)
wandb.log(metrics, step=train_state.step)
if train_state.step % config.train.eval_interval == 0:
torch.cuda.empty_cache()
if accelerator.is_main_process:
logging.info('Save a grid of images...')
with torch.no_grad():
if config.train.mode == 'uncond':
_n_samples = 5 if config.get('z_shape', False) and (config.z_shape[-1] >= 128) else 5*10
samples_list = []
for i in range(0, _n_samples, config.sample.mini_batch_size):
samples = dpm_solver_sample(_n_samples=config.sample.mini_batch_size, _sample_steps=50)
samples_list.append(samples)
samples_list = torch.cat(samples_list, dim=0)
samples = samples_list[:_n_samples]
elif config.train.mode == 'cond':
_n_samples = 5*10
y = einops.repeat(torch.arange(5, device=device) % dataset.K, 'nrow -> (nrow ncol)', ncol=10)
samples = dpm_solver_sample(_n_samples=_n_samples, _sample_steps=50, y=y)
else:
raise NotImplementedError
if accelerator.is_main_process:
samples = make_grid(dataset.unpreprocess(samples), min(10, _n_samples) )
save_image(samples, os.path.join(config.sample_dir, f'{train_state.step}.png'))
wandb.log({'samples': wandb.Image(samples)}, step=train_state.step)
torch.cuda.empty_cache()
accelerator.wait_for_everyone()
if train_state.step % config.train.save_interval == 0 or train_state.step == config.train.n_steps:
torch.cuda.empty_cache()
logging.info(f'Save and eval checkpoint {train_state.step}...')
save_path = os.path.join(config.ckpt_root, f'{train_state.step}.ckpt')
if accelerator.local_process_index == 0:
# train_state.save(os.path.join(config.ckpt_root, f'{train_state.step}.ckpt'))
os.makedirs(save_path, exist_ok=True)
torch.save(train_state.step, os.path.join(save_path, 'step.pth'))
torch.save(train_state.lr_scheduler.state_dict(), os.path.join(save_path, 'lr_scheduler.pth'))
torch.save(nnet_ema.state_dict(), os.path.join(save_path, 'nnet_ema.pth'))
accelerator.save_state(output_dir=save_path)
accelerator.wait_for_everyone()
fid = eval_step(n_samples=min(10000, config.sample.get('intermidiate_n_samples', 10000) ), sample_steps=50) # calculate fid of the saved checkpoint
step_fid.append((train_state.step, fid))
torch.cuda.empty_cache()
accelerator.wait_for_everyone()
logging.info(f'Finish fitting, step={train_state.step}')
logging.info(f'step_fid: {step_fid}')
step_best = sorted(step_fid, key=lambda x: x[1])[0][0]
logging.info(f'step_best: {step_best}')
train_state.load(os.path.join(config.ckpt_root, f'{step_best}.ckpt'))
del metrics
accelerator.wait_for_everyone()
eval_step(n_samples=config.sample.n_samples, sample_steps=config.sample.sample_steps)
from absl import flags
from absl import app
from ml_collections import config_flags
import sys
from pathlib import Path
FLAGS = flags.FLAGS
config_flags.DEFINE_config_file(
"config", None, "Training configuration.", lock_config=False)
flags.mark_flags_as_required(["config"])
flags.DEFINE_string("workdir", None, "Work unit directory.")
def get_config_name():
argv = sys.argv
for i in range(1, len(argv)):
if argv[i].startswith('--config='):
return Path(argv[i].split('=')[-1]).stem
def get_hparams():
argv = sys.argv
lst = []
for i in range(1, len(argv)):
assert '=' in argv[i]
if argv[i].startswith('--config.') and not argv[i].startswith('--config.dataset.path'):
hparam, val = argv[i].split('=')
hparam = hparam.split('.')[-1]
if hparam.endswith('path'):
val = Path(val).stem
lst.append(f'{hparam}={val}')
hparams = '-'.join(lst)
if hparams == '':
hparams = 'default'
return hparams
def main(argv):
config = FLAGS.config
config.config_name = get_config_name()
config.hparams = get_hparams()
config.workdir = FLAGS.workdir or os.path.join('workdir', config.config_name, config.hparams)
config.ckpt_root = os.path.join(config.workdir, 'ckpts')
config.sample_dir = os.path.join(config.workdir, 'samples')
train(config)
if __name__ == "__main__":
app.run(main)