mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-16 04:32:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

Replace with autocast(...) with with autocast("cuda", ...)

Browse Source
This commit is contained in:
Li Peng 2024-11-28 18:28:26 +08:00
parent 2d9825aa29
commit 30ba83a7b2
7 changed files with 21 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
egs/libritts/CODEC/encodec/encodec.py
View File

@ -148,7 +148,7 @@ class Encodec(nn.Module):
) )
# calculate losses # calculate losses
with autocast(enabled=False): with autocast("cuda", enabled=False):
gen_stft_adv_loss = self.generator_adversarial_loss(outputs=y_hat) gen_stft_adv_loss = self.generator_adversarial_loss(outputs=y_hat)
if self.multi_period_discriminator is not None: if self.multi_period_discriminator is not None:
@ -272,7 +272,7 @@ class Encodec(nn.Module):
speech_hat.contiguous().detach(), speech_hat.contiguous().detach(),
) )
# calculate losses # calculate losses
with autocast(enabled=False): with autocast("cuda", enabled=False):
( (
disc_stft_real_adv_loss, disc_stft_real_adv_loss,
disc_stft_fake_adv_loss, disc_stft_fake_adv_loss,

8
egs/libritts/CODEC/encodec/train.py
View File

@ -466,7 +466,7 @@ def train_one_epoch(
loss_info["samples"] = batch_size loss_info["samples"] = batch_size
try: try:
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
d_weight = train_discriminator( d_weight = train_discriminator(
params.lambda_adv, params.lambda_adv,
params.cur_epoch, params.cur_epoch,
@ -502,7 +502,7 @@ def train_one_epoch(
scaler.scale(disc_loss).backward() scaler.scale(disc_loss).backward()
scaler.step(optimizer_d) scaler.step(optimizer_d)
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
g_weight = train_discriminator( g_weight = train_discriminator(
params.lambda_adv, params.lambda_adv,
params.cur_epoch, params.cur_epoch,
@ -846,7 +846,7 @@ def scan_pessimistic_batches_for_oom(
) = prepare_input(params, batch, device) ) = prepare_input(params, batch, device)
try: try:
# for discriminator # for discriminator
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
( (
disc_stft_real_adv_loss, disc_stft_real_adv_loss,
disc_stft_fake_adv_loss, disc_stft_fake_adv_loss,
@ -876,7 +876,7 @@ def scan_pessimistic_batches_for_oom(
optimizer_d.zero_grad() optimizer_d.zero_grad()
loss_d.backward() loss_d.backward()
# for generator # for generator
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
( (
commit_loss, commit_loss,
gen_stft_adv_loss, gen_stft_adv_loss,

8
egs/libritts/TTS/vits/train.py
View File

@ -456,7 +456,7 @@ def train_one_epoch(
loss_info["samples"] = batch_size loss_info["samples"] = batch_size
try: try:
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
# forward discriminator # forward discriminator
loss_d, stats_d = model( loss_d, stats_d = model(
text=tokens, text=tokens,
@ -475,7 +475,7 @@ def train_one_epoch(
scaler.scale(loss_d).backward() scaler.scale(loss_d).backward()
scaler.step(optimizer_d) scaler.step(optimizer_d)
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
# forward generator # forward generator
loss_g, stats_g = model( loss_g, stats_g = model(
text=tokens, text=tokens,
@ -748,7 +748,7 @@ def scan_pessimistic_batches_for_oom(
) = prepare_input(batch, tokenizer, device, train_speaker_map) ) = prepare_input(batch, tokenizer, device, train_speaker_map)
try: try:
# for discriminator # for discriminator
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
loss_d, stats_d = model( loss_d, stats_d = model(
text=tokens, text=tokens,
text_lengths=tokens_lens, text_lengths=tokens_lens,
@ -762,7 +762,7 @@ def scan_pessimistic_batches_for_oom(
optimizer_d.zero_grad() optimizer_d.zero_grad()
loss_d.backward() loss_d.backward()
# for generator # for generator
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
loss_g, stats_g = model( loss_g, stats_g = model(
text=tokens, text=tokens,
text_lengths=tokens_lens, text_lengths=tokens_lens,

2
egs/ljspeech/TTS/matcha/train.py
View File

@ -479,7 +479,7 @@ def train_one_epoch(
tokens_lens, tokens_lens,
) = prepare_input(batch, tokenizer, device, params) ) = prepare_input(batch, tokenizer, device, params)
try: try:
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
losses = get_losses( losses = get_losses(
{ {
"x": tokens, "x": tokens,

8
egs/ljspeech/TTS/vits/train.py
View File

@ -396,7 +396,7 @@ def train_one_epoch(
loss_info["samples"] = batch_size loss_info["samples"] = batch_size
try: try:
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
# forward discriminator # forward discriminator
loss_d, stats_d = model( loss_d, stats_d = model(
text=tokens, text=tokens,
@ -414,7 +414,7 @@ def train_one_epoch(
scaler.scale(loss_d).backward() scaler.scale(loss_d).backward()
scaler.step(optimizer_d) scaler.step(optimizer_d)
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
# forward generator # forward generator
loss_g, stats_g = model( loss_g, stats_g = model(
text=tokens, text=tokens,
@ -673,7 +673,7 @@ def scan_pessimistic_batches_for_oom(
) )
try: try:
# for discriminator # for discriminator
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
loss_d, stats_d = model( loss_d, stats_d = model(
text=tokens, text=tokens,
text_lengths=tokens_lens, text_lengths=tokens_lens,
@ -686,7 +686,7 @@ def scan_pessimistic_batches_for_oom(
optimizer_d.zero_grad() optimizer_d.zero_grad()
loss_d.backward() loss_d.backward()
# for generator # for generator
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
loss_g, stats_g = model( loss_g, stats_g = model(
text=tokens, text=tokens,
text_lengths=tokens_lens, text_lengths=tokens_lens,

4
egs/ljspeech/TTS/vits/vits.py
View File

@ -410,7 +410,7 @@ class VITS(nn.Module):
p = self.discriminator(speech_) p = self.discriminator(speech_)
# calculate losses # calculate losses
with autocast(enabled=False): with autocast("cuda", enabled=False):
if not return_sample: if not return_sample:
mel_loss = self.mel_loss(speech_hat_, speech_) mel_loss = self.mel_loss(speech_hat_, speech_)
else: else:
@ -518,7 +518,7 @@ class VITS(nn.Module):
p = self.discriminator(speech_) p = self.discriminator(speech_)
# calculate losses # calculate losses
with autocast(enabled=False): with autocast("cuda", enabled=False):
real_loss, fake_loss = self.discriminator_adv_loss(p_hat, p) real_loss, fake_loss = self.discriminator_adv_loss(p_hat, p)
loss = real_loss + fake_loss loss = real_loss + fake_loss

8
egs/vctk/TTS/vits/train.py
View File

@ -448,7 +448,7 @@ def train_one_epoch(
loss_info["samples"] = batch_size loss_info["samples"] = batch_size
try: try:
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
# forward discriminator # forward discriminator
loss_d, stats_d = model( loss_d, stats_d = model(
text=tokens, text=tokens,
@ -467,7 +467,7 @@ def train_one_epoch(
scaler.scale(loss_d).backward() scaler.scale(loss_d).backward()
scaler.step(optimizer_d) scaler.step(optimizer_d)
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
# forward generator # forward generator
loss_g, stats_g = model( loss_g, stats_g = model(
text=tokens, text=tokens,
@ -740,7 +740,7 @@ def scan_pessimistic_batches_for_oom(
) = prepare_input(batch, tokenizer, device, speaker_map) ) = prepare_input(batch, tokenizer, device, speaker_map)
try: try:
# for discriminator # for discriminator
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
loss_d, stats_d = model( loss_d, stats_d = model(
text=tokens, text=tokens,
text_lengths=tokens_lens, text_lengths=tokens_lens,
@ -754,7 +754,7 @@ def scan_pessimistic_batches_for_oom(
optimizer_d.zero_grad() optimizer_d.zero_grad()
loss_d.backward() loss_d.backward()
# for generator # for generator
with autocast(enabled=params.use_fp16): with autocast("cuda", enabled=params.use_fp16):
loss_g, stats_g = model( loss_g, stats_g = model(
text=tokens, text=tokens,
text_lengths=tokens_lens, text_lengths=tokens_lens,