mirror of https://github.com/coqui-ai/TTS.git
lint fixes
Eren Golge
parent
afdc4bad10
commit
529348d6dc
10
setup.py
10
setup.py
|
|
@ -66,11 +66,11 @@ setup(
|
||||||
package_dir={'': 'tts_namespace'},
|
package_dir={'': 'tts_namespace'},
|
||||||
packages=find_packages('tts_namespace'),
|
packages=find_packages('tts_namespace'),
|
||||||
project_urls={
|
project_urls={
|
||||||
'Documentation': 'https://github.com/mozilla/TTS/wiki',
|
'Documentation': 'https://github.com/mozilla/TTS/wiki',
|
||||||
'Tracker': 'https://github.com/mozilla/TTS/issues',
|
'Tracker': 'https://github.com/mozilla/TTS/issues',
|
||||||
'Repository': 'https://github.com/mozilla/TTS',
|
'Repository': 'https://github.com/mozilla/TTS',
|
||||||
'Discussions': 'https://discourse.mozilla.org/c/tts',
|
'Discussions': 'https://discourse.mozilla.org/c/tts',
|
||||||
},
|
},
|
||||||
cmdclass={
|
cmdclass={
|
||||||
'build_py': build_py,
|
'build_py': build_py,
|
||||||
'develop': develop,
|
'develop': develop,
|
||||||
|
|
|
||||||
108
utils/radam.py
108
utils/radam.py
|
|
@ -1,6 +1,7 @@
|
||||||
import math
|
import math
|
||||||
import torch
|
import torch
|
||||||
from torch.optim.optimizer import Optimizer, required
|
from torch.optim.optimizer import Optimizer
|
||||||
|
|
||||||
|
|
||||||
class RAdam(Optimizer):
|
class RAdam(Optimizer):
|
||||||
|
|
||||||
|
|
@ -9,7 +10,7 @@ class RAdam(Optimizer):
|
||||||
self.buffer = [[None, None, None] for ind in range(10)]
|
self.buffer = [[None, None, None] for ind in range(10)]
|
||||||
super(RAdam, self).__init__(params, defaults)
|
super(RAdam, self).__init__(params, defaults)
|
||||||
|
|
||||||
def __setstate__(self, state):
|
def __setstate__(self, state): # pylint: disable= useless-super-delegation
|
||||||
super(RAdam, self).__setstate__(state)
|
super(RAdam, self).__setstate__(state)
|
||||||
|
|
||||||
def step(self, closure=None):
|
def step(self, closure=None):
|
||||||
|
|
@ -25,19 +26,21 @@ class RAdam(Optimizer):
|
||||||
continue
|
continue
|
||||||
grad = p.grad.data.float()
|
grad = p.grad.data.float()
|
||||||
if grad.is_sparse:
|
if grad.is_sparse:
|
||||||
raise RuntimeError('RAdam does not support sparse gradients')
|
raise RuntimeError(
|
||||||
|
'RAdam does not support sparse gradients')
|
||||||
|
|
||||||
p_data_fp32 = p.data.float()
|
p_data_fp32 = p.data.float()
|
||||||
|
|
||||||
state = self.state[p]
|
state = self.state[p]
|
||||||
|
|
||||||
if len(state) == 0:
|
if not state:
|
||||||
state['step'] = 0
|
state['step'] = 0
|
||||||
state['exp_avg'] = torch.zeros_like(p_data_fp32)
|
state['exp_avg'] = torch.zeros_like(p_data_fp32)
|
||||||
state['exp_avg_sq'] = torch.zeros_like(p_data_fp32)
|
state['exp_avg_sq'] = torch.zeros_like(p_data_fp32)
|
||||||
else:
|
else:
|
||||||
state['exp_avg'] = state['exp_avg'].type_as(p_data_fp32)
|
state['exp_avg'] = state['exp_avg'].type_as(p_data_fp32)
|
||||||
state['exp_avg_sq'] = state['exp_avg_sq'].type_as(p_data_fp32)
|
state['exp_avg_sq'] = state['exp_avg_sq'].type_as(
|
||||||
|
p_data_fp32)
|
||||||
|
|
||||||
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
|
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
|
||||||
beta1, beta2 = group['betas']
|
beta1, beta2 = group['betas']
|
||||||
|
|
@ -53,21 +56,24 @@ class RAdam(Optimizer):
|
||||||
buffered[0] = state['step']
|
buffered[0] = state['step']
|
||||||
beta2_t = beta2 ** state['step']
|
beta2_t = beta2 ** state['step']
|
||||||
N_sma_max = 2 / (1 - beta2) - 1
|
N_sma_max = 2 / (1 - beta2) - 1
|
||||||
N_sma = N_sma_max - 2 * state['step'] * beta2_t / (1 - beta2_t)
|
N_sma = N_sma_max - 2 * \
|
||||||
|
state['step'] * beta2_t / (1 - beta2_t)
|
||||||
buffered[1] = N_sma
|
buffered[1] = N_sma
|
||||||
|
|
||||||
# more conservative since it's an approximated value
|
# more conservative since it's an approximated value
|
||||||
if N_sma >= 5:
|
if N_sma >= 5:
|
||||||
step_size = group['lr'] * math.sqrt((1 - beta2_t) * (N_sma - 4) / (N_sma_max - 4) * (N_sma - 2) / N_sma * N_sma_max / (N_sma_max - 2)) / (1 - beta1 ** state['step'])
|
step_size = group['lr'] * math.sqrt((1 - beta2_t) * (N_sma - 4) / (N_sma_max - 4) * (
|
||||||
|
N_sma - 2) / N_sma * N_sma_max / (N_sma_max - 2)) / (1 - beta1 ** state['step'])
|
||||||
else:
|
else:
|
||||||
step_size = group['lr'] / (1 - beta1 ** state['step'])
|
step_size = group['lr'] / (1 - beta1 ** state['step'])
|
||||||
buffered[2] = step_size
|
buffered[2] = step_size
|
||||||
|
|
||||||
if group['weight_decay'] != 0:
|
if group['weight_decay'] != 0:
|
||||||
p_data_fp32.add_(-group['weight_decay'] * group['lr'], p_data_fp32)
|
p_data_fp32.add_(-group['weight_decay']
|
||||||
|
* group['lr'], p_data_fp32)
|
||||||
|
|
||||||
# more conservative since it's an approximated value
|
# more conservative since it's an approximated value
|
||||||
if N_sma >= 5:
|
if N_sma >= 5:
|
||||||
denom = exp_avg_sq.sqrt().add_(group['eps'])
|
denom = exp_avg_sq.sqrt().add_(group['eps'])
|
||||||
p_data_fp32.addcdiv_(-step_size, exp_avg, denom)
|
p_data_fp32.addcdiv_(-step_size, exp_avg, denom)
|
||||||
else:
|
else:
|
||||||
|
|
@ -77,6 +83,7 @@ class RAdam(Optimizer):
|
||||||
|
|
||||||
return loss
|
return loss
|
||||||
|
|
||||||
|
|
||||||
class PlainRAdam(Optimizer):
|
class PlainRAdam(Optimizer):
|
||||||
|
|
||||||
def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-8, weight_decay=0):
|
def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-8, weight_decay=0):
|
||||||
|
|
@ -84,7 +91,7 @@ class PlainRAdam(Optimizer):
|
||||||
|
|
||||||
super(PlainRAdam, self).__init__(params, defaults)
|
super(PlainRAdam, self).__init__(params, defaults)
|
||||||
|
|
||||||
def __setstate__(self, state):
|
def __setstate__(self, state): # pylint: disable= useless-super-delegation
|
||||||
super(PlainRAdam, self).__setstate__(state)
|
super(PlainRAdam, self).__setstate__(state)
|
||||||
|
|
||||||
def step(self, closure=None):
|
def step(self, closure=None):
|
||||||
|
|
@ -100,19 +107,21 @@ class PlainRAdam(Optimizer):
|
||||||
continue
|
continue
|
||||||
grad = p.grad.data.float()
|
grad = p.grad.data.float()
|
||||||
if grad.is_sparse:
|
if grad.is_sparse:
|
||||||
raise RuntimeError('RAdam does not support sparse gradients')
|
raise RuntimeError(
|
||||||
|
'RAdam does not support sparse gradients')
|
||||||
|
|
||||||
p_data_fp32 = p.data.float()
|
p_data_fp32 = p.data.float()
|
||||||
|
|
||||||
state = self.state[p]
|
state = self.state[p]
|
||||||
|
|
||||||
if len(state) == 0:
|
if not state:
|
||||||
state['step'] = 0
|
state['step'] = 0
|
||||||
state['exp_avg'] = torch.zeros_like(p_data_fp32)
|
state['exp_avg'] = torch.zeros_like(p_data_fp32)
|
||||||
state['exp_avg_sq'] = torch.zeros_like(p_data_fp32)
|
state['exp_avg_sq'] = torch.zeros_like(p_data_fp32)
|
||||||
else:
|
else:
|
||||||
state['exp_avg'] = state['exp_avg'].type_as(p_data_fp32)
|
state['exp_avg'] = state['exp_avg'].type_as(p_data_fp32)
|
||||||
state['exp_avg_sq'] = state['exp_avg_sq'].type_as(p_data_fp32)
|
state['exp_avg_sq'] = state['exp_avg_sq'].type_as(
|
||||||
|
p_data_fp32)
|
||||||
|
|
||||||
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
|
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
|
||||||
beta1, beta2 = group['betas']
|
beta1, beta2 = group['betas']
|
||||||
|
|
@ -126,11 +135,13 @@ class PlainRAdam(Optimizer):
|
||||||
N_sma = N_sma_max - 2 * state['step'] * beta2_t / (1 - beta2_t)
|
N_sma = N_sma_max - 2 * state['step'] * beta2_t / (1 - beta2_t)
|
||||||
|
|
||||||
if group['weight_decay'] != 0:
|
if group['weight_decay'] != 0:
|
||||||
p_data_fp32.add_(-group['weight_decay'] * group['lr'], p_data_fp32)
|
p_data_fp32.add_(-group['weight_decay']
|
||||||
|
* group['lr'], p_data_fp32)
|
||||||
|
|
||||||
# more conservative since it's an approximated value
|
# more conservative since it's an approximated value
|
||||||
if N_sma >= 5:
|
if N_sma >= 5:
|
||||||
step_size = group['lr'] * math.sqrt((1 - beta2_t) * (N_sma - 4) / (N_sma_max - 4) * (N_sma - 2) / N_sma * N_sma_max / (N_sma_max - 2)) / (1 - beta1 ** state['step'])
|
step_size = group['lr'] * math.sqrt((1 - beta2_t) * (N_sma - 4) / (N_sma_max - 4) * (
|
||||||
|
N_sma - 2) / N_sma * N_sma_max / (N_sma_max - 2)) / (1 - beta1 ** state['step'])
|
||||||
denom = exp_avg_sq.sqrt().add_(group['eps'])
|
denom = exp_avg_sq.sqrt().add_(group['eps'])
|
||||||
p_data_fp32.addcdiv_(-step_size, exp_avg, denom)
|
p_data_fp32.addcdiv_(-step_size, exp_avg, denom)
|
||||||
else:
|
else:
|
||||||
|
|
@ -140,68 +151,3 @@ class PlainRAdam(Optimizer):
|
||||||
p.data.copy_(p_data_fp32)
|
p.data.copy_(p_data_fp32)
|
||||||
|
|
||||||
return loss
|
return loss
|
||||||
|
|
||||||
|
|
||||||
class AdamW(Optimizer):
|
|
||||||
|
|
||||||
def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-8, weight_decay=0, warmup = 0):
|
|
||||||
defaults = dict(lr=lr, betas=betas, eps=eps,
|
|
||||||
weight_decay=weight_decay, warmup = warmup)
|
|
||||||
super(AdamW, self).__init__(params, defaults)
|
|
||||||
|
|
||||||
def __setstate__(self, state):
|
|
||||||
super(AdamW, self).__setstate__(state)
|
|
||||||
|
|
||||||
def step(self, closure=None):
|
|
||||||
loss = None
|
|
||||||
if closure is not None:
|
|
||||||
loss = closure()
|
|
||||||
|
|
||||||
for group in self.param_groups:
|
|
||||||
|
|
||||||
for p in group['params']:
|
|
||||||
if p.grad is None:
|
|
||||||
continue
|
|
||||||
grad = p.grad.data.float()
|
|
||||||
if grad.is_sparse:
|
|
||||||
raise RuntimeError('Adam does not support sparse gradients, please consider SparseAdam instead')
|
|
||||||
|
|
||||||
p_data_fp32 = p.data.float()
|
|
||||||
|
|
||||||
state = self.state[p]
|
|
||||||
|
|
||||||
if len(state) == 0:
|
|
||||||
state['step'] = 0
|
|
||||||
state['exp_avg'] = torch.zeros_like(p_data_fp32)
|
|
||||||
state['exp_avg_sq'] = torch.zeros_like(p_data_fp32)
|
|
||||||
else:
|
|
||||||
state['exp_avg'] = state['exp_avg'].type_as(p_data_fp32)
|
|
||||||
state['exp_avg_sq'] = state['exp_avg_sq'].type_as(p_data_fp32)
|
|
||||||
|
|
||||||
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
|
|
||||||
beta1, beta2 = group['betas']
|
|
||||||
|
|
||||||
state['step'] += 1
|
|
||||||
|
|
||||||
exp_avg_sq.mul_(beta2).addcmul_(1 - beta2, grad, grad)
|
|
||||||
exp_avg.mul_(beta1).add_(1 - beta1, grad)
|
|
||||||
|
|
||||||
denom = exp_avg_sq.sqrt().add_(group['eps'])
|
|
||||||
bias_correction1 = 1 - beta1 ** state['step']
|
|
||||||
bias_correction2 = 1 - beta2 ** state['step']
|
|
||||||
|
|
||||||
if group['warmup'] > state['step']:
|
|
||||||
scheduled_lr = 1e-8 + state['step'] * group['lr'] / group['warmup']
|
|
||||||
else:
|
|
||||||
scheduled_lr = group['lr']
|
|
||||||
|
|
||||||
step_size = group['lr'] * math.sqrt(bias_correction2) / bias_correction1
|
|
||||||
|
|
||||||
if group['weight_decay'] != 0:
|
|
||||||
p_data_fp32.add_(-group['weight_decay'] * scheduled_lr, p_data_fp32)
|
|
||||||
|
|
||||||
p_data_fp32.addcdiv_(-step_size, exp_avg, denom)
|
|
||||||
|
|
||||||
p.data.copy_(p_data_fp32)
|
|
||||||
|
|
||||||
return loss
|
|
||||||
|
|
|
||||||
Loading…
Reference in New Issue