-
Notifications
You must be signed in to change notification settings - Fork 21
/
chassis.py
350 lines (278 loc) · 12.8 KB
/
chassis.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
from sys import stderr
import torch as t
from tensorboardX import SummaryWriter
# this SummaryWriter doesn't work with torch_xla, causes crash
# from torch.utils.tensorboard import SummaryWriter
import data
import autoencoder_model as ae
import mfcc_inverter as mi
import checkpoint as ckpt
import util
import netmisc
import librosa
import os.path
import time
try:
import torch_xla
import torch_xla.core.xla_model as xm
import torch_xla.distributed.parallel_loader as pl
except ModuleNotFoundError:
pass
class GPULoaderIter(object):
def __init__(self, loader, device):
self.loader_iter = iter(loader)
self.device = device
def __iter__(self):
return self
def __next__(self):
items = next(self.loader_iter)
return tuple(item.to(self.device) if isinstance(item, t.Tensor) else
item for item in items)
def reduce_add(vlist):
return t.stack(vlist).sum(dim=0)
def reduce_mean(vlist):
return t.stack(vlist).mean(dim=0)
class Chassis(object):
"""
Coordinates the construction of the model, dataset, optimizer,
checkpointing state, and GPU/TPU iterator wrappers.
Provides a single function for training the model from the constructed
setup.
"""
def __init__(self, device, index, hps, dat_file):
self.is_tpu = (hps.hw in ('TPU', 'TPU-single'))
if self.is_tpu:
num_replicas = xm.xrt_world_size()
rank = xm.get_ordinal()
elif hps.hw == 'GPU':
if not t.cuda.is_available():
raise RuntimeError('GPU requested but not available')
num_replicas = 1
rank = 0
elif hps.hw == 'CPU':
num_replicas = 1
rank = 0
else:
raise ValueError(f'Chassis: Invalid device "{hps.hw}" requested')
self.replica_index = index
self.state = ckpt.Checkpoint(hps, dat_file, train_mode=True,
ckpt_file=hps.get('ckpt_file', None),
num_replicas=num_replicas, rank=rank)
hps = self.state.hps
if not self.is_tpu or xm.is_master_ordinal():
print('Hyperparameters:\n', file=stderr)
print('\n'.join(f'{k} = {v}' for k, v in hps.items()), file=stderr)
self.learning_rates = dict(zip(hps.learning_rate_steps,
hps.learning_rate_rates))
if self.state.model.bn_type == 'vae':
self.anneal_schedule = dict(zip(hps.bn_anneal_weight_steps,
hps.bn_anneal_weight_vals))
self.ckpt_path = util.CheckpointPath(hps.ckpt_template, not self.is_tpu
or xm.is_master_ordinal())
self.softmax = t.nn.Softmax(1) # input to this is (B, Q, N)
self.hw = hps.hw
if hps.hw == 'GPU':
self.device_loader = GPULoaderIter(self.state.data.loader, device)
self.state.to(device)
else:
para_loader = pl.ParallelLoader(self.state.data.loader, [device])
self.device_loader = para_loader.per_device_loader(device)
self.num_devices = xm.xrt_world_size()
self.state.to(device)
self.state.init_torch_generator()
if not self.is_tpu or xm.is_master_ordinal():
self.writer = SummaryWriter(log_dir=hps.log_dir)
else:
self.writer = None
def train(self):
hps = self.state.hps
ss = self.state
current_stats = {}
writer_stats = {}
# for resuming the learning rate
sorted_lr_steps = sorted(self.learning_rates.keys())
lr_index = util.greatest_lower_bound(sorted_lr_steps, ss.data.global_step)
ss.update_learning_rate(self.learning_rates[sorted_lr_steps[lr_index]])
if ss.model.bn_type != 'none':
sorted_as_steps = sorted(self.anneal_schedule.keys())
as_index = util.greatest_lower_bound(sorted_as_steps,
ss.data.global_step)
ss.model.objective.update_anneal_weight(self.anneal_schedule[sorted_as_steps[as_index]])
if ss.model.bn_type in ('vqvae', 'vqvae-ema'):
ss.model.init_codebook(self.data_iter, 10000)
start_time = time.time()
for batch_num, batch in enumerate(self.device_loader):
wav, mel, voice, jitter, position = batch
global_step = len(ss.data.dataset) * position[0] + position[1]
# print(f'replica {self.replica_index}, batch {batch_num}', file=stderr)
# stderr.flush()
if (batch_num % hps.save_interval == 0 and batch_num != 0):
self.save_checkpoint(position)
if hps.skip_loop_body:
continue
lr_index = util.greatest_lower_bound(sorted_lr_steps, global_step)
ss.update_learning_rate(self.learning_rates[sorted_lr_steps[lr_index]])
# if ss.data.global_step in self.learning_rates:
# ss.update_learning_rate(self.learning_rates[ss.data.global_step])
if ss.model.bn_type == 'vae' and ss.step in self.anneal_schedule:
ss.model.objective.update_anneal_weight(self.anneal_schedule[ss.data.global_step])
ss.optim.zero_grad()
quant, self.target, loss = self.state.model.run(wav, mel, voice, jitter)
self.probs = self.softmax(quant)
self.mel_enc_input = mel
# print(f'after model.run', file=stderr)
# stderr.flush()
loss.backward()
# print(f'after loss.backward()', file=stderr)
# stderr.flush()
if batch_num % hps.progress_interval == 0:
pars_copy = [p.data.clone() for p in ss.model.parameters()]
# print(f'after pars_copy', file=stderr)
# stderr.flush()
if self.is_tpu:
xm.optimizer_step(ss.optim)
else:
ss.optim.step()
ss.optim_step += 1
if ss.model.bn_type == 'vqvae-ema' and ss.data.global_step == 10000:
ss.model.bottleneck.update_codebook()
tprb_m = self.avg_prob_target()
if batch_num % hps.progress_interval == 0:
iterator = zip(pars_copy, ss.model.named_parameters())
uw_ratio = { np[0]: t.norm(c - np[1].data) / c.norm() for c, np
in iterator }
writer_stats.update({ 'uwr': uw_ratio })
if self.is_tpu:
count = torch_xla._XLAC._xla_get_replication_devices_count()
loss_red, tprb_red = xm.all_reduce('sum', [loss, tprb_m],
scale=1.0 / count)
# loss_red = xm.all_reduce('all_loss', loss, reduce_mean)
# tprb_red = xm.all_reduce('all_tprb', tprb_m, reduce_mean)
else:
loss_red = loss
tprb_red = tprb_m
writer_stats.update({
'loss_r': loss_red,
'tprb_r': tprb_red,
'optim_step': ss.optim_step
})
current_stats.update({
'optim_step': ss.optim_step,
'gstep': global_step,
# 'gstep': ss.data.global_step,
'epoch': position[0],
'step': position[1],
# 'loss': loss,
'lrate': ss.optim.param_groups[0]['lr'],
# 'tprb_m': tprb_m,
# 'pk_d_m': avg_peak_dist
})
current_stats.update(ss.model.objective.metrics)
if ss.model.bn_type in ('vae'):
current_stats['free_nats'] = ss.model.objective.free_nats
current_stats['anneal_weight'] = \
ss.model.objective.anneal_weight.item()
if ss.model.bn_type in ('vqvae', 'vqvae-ema', 'ae', 'vae'):
current_stats.update(ss.model.encoder.metrics)
if self.is_tpu:
xm.add_step_closure(
self.train_update,
args=(writer_stats, current_stats))
else:
self.train_update(writer_stats, current_stats)
# if not self.is_tpu or xm.is_master_ordinal():
# if batch_num in range(25, 50) or batch_num in range(75, 100):
stderr.flush()
elapsed = time.time() - start_time
# print(f'{elapsed}, worker {self.replica_index}, batch {batch_num}', file=stderr)
# stderr.flush()
def train_update(self, writer_stats, stdout_stats):
if self.replica_index == 0:
netmisc.print_metrics(stdout_stats, self.replica_index, 100)
if self.writer:
self.writer.add_scalars('metrics', { k: writer_stats[k].item() for k
in ('loss_r', 'tprb_r') }, writer_stats['optim_step'])
self.writer.add_scalars('uw ratio', writer_stats['uwr'], writer_stats['optim_step'])
self.writer.flush()
def save_checkpoint(self, position):
global_step = len(self.state.data.dataset) * position[0] + position[1]
ckpt_file = self.ckpt_path.path(global_step.item())
self.state.save(ckpt_file, position[0], position[1])
if not self.is_tpu or xm.is_master_ordinal():
print('Saved checkpoint to {}'.format(ckpt_file), file=stderr)
stderr.flush()
def avg_max(self):
"""Average max value for the predictions. As the prediction becomes
more peaked, this should go up"""
max_val, max_ind = t.max(self.probs, dim=1)
mean = t.mean(max_val)
return mean
def avg_prob_target(self):
"""Average probability given to target"""
target_probs = t.gather(self.probs, 1, self.target.long().unsqueeze(1))
mean = t.mean(target_probs)
return mean
class DataContainer(t.nn.Module):
def __init__(self, my_values):
super().__init__()
for key in my_values:
setattr(self, key, my_values[key])
def forward(self):
pass
class InferenceChassis(object):
"""
Coordinates construction of model and dataset for running inference
"""
def __init__(self, device, index, hps, dat_file):
self.output_dir = hps.output_dir
self.n_replicas = hps.dec_n_replicas
try:
self.data_write_tmpl = hps.data_write_tmpl
except AttributeError:
self.data_write_tmpl = None
self.state = ckpt.InferenceState(hps, dat_file, hps.ckpt_file)
self.state.model.wavenet.set_n_replicas(self.n_replicas)
self.state.model.eval()
self.sample_rate = hps.sample_rate
if hps.hw in ('GPU', 'CPU'):
self.device_loader = GPULoaderIter(self.state.data.loader, device)
self.state.to(device)
else:
import torch_xla.core.xla_model as xm
import torch_xla.distributed.parallel_loader as pl
para_loader = pl.ParallelLoader(self.state.data.loader, [device])
self.device_loader = para_loader.per_device_loader(device)
self.num_devices = xm.xrt_world_size()
self.state.to(device)
def infer(self, model_scr=None):
n_quant = self.state.model.wavenet.n_quant
for batch in self.device_loader:
wav, mel, voice_idx, jitter_idx, file_paths, position = batch
if self.data_write_tmpl:
dc = t.jit.script(DataContainer({
'mel': mel,
'wav': wav,
'voice': voice_idx,
'jitter': jitter_idx
}))
dc.save(self.data_write_tmpl)
print('saved {}'.format(self.data_write_tmpl))
out_template = os.path.join(self.output_dir,
os.path.basename(os.path.splitext(file_paths[0])[0])
+ '.{}.wav')
if model_scr:
with t.no_grad():
wav = model_scr(wav, mel, voice_idx, jitter_idx)
else:
wav = self.state.model(wav, mel, voice_idx, jitter_idx)
wav_orig, wav_sample = wav[0,...], wav[1:,...]
# save results to specified files
for i in range(self.n_replicas):
wav_final = util.mu_decode_torch(wav_sample[i], n_quant)
path = out_template.format('rep' + str(i))
librosa.output.write_wav(path, wav_final.cpu().numpy(), self.sample_rate)
wav_final = util.mu_decode_torch(wav_orig, n_quant)
path = out_template.format('orig')
librosa.output.write_wav(path, wav_final.cpu().numpy(), self.sample_rate)
print('Wrote {}'.format(
out_template.format('0-'+str(self.n_replicas-1))))