Python torch.optim.optimizer.required() Examples
The following are 30
code examples of torch.optim.optimizer.required().
You can vote up the ones you like or vote down the ones you don't like,
and go to the original project or source file by following the links above each example.
You may also want to check out all available functions/classes of the module
torch.optim.optimizer
, or try the search function
.
.
Example #1
| Source File: optimization.py From cmrc2019 with Creative Commons Attribution Share Alike 4.0 International | 7 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #2
| Source File: optimization_openai.py From bert_on_stilts with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, schedule='warmup_linear', warmup=-1, t_total=-1,
b1=0.9, b2=0.999, e=1e-8, weight_decay=0,
vector_l2=False, max_grad_norm=-1, **kwargs):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {}".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {}".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {}".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay, vector_l2=vector_l2,
max_grad_norm=max_grad_norm)
super(OpenAIAdam, self).__init__(params, defaults)
Example #3
| Source File: optimization.py From Bert-Chinese-Text-Classification-Pytorch with MIT License | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01, max_grad_norm=1.0, **kwargs):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if not isinstance(schedule, _LRSchedule) and schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
# initialize schedule object
if not isinstance(schedule, _LRSchedule):
schedule_type = SCHEDULES[schedule]
schedule = schedule_type(warmup=warmup, t_total=t_total)
else:
if warmup != -1 or t_total != -1:
logger.warning("warmup and t_total on the optimizer are ineffective when _LRSchedule object is provided as schedule. "
"Please specify custom warmup and t_total in _LRSchedule object.")
defaults = dict(lr=lr, schedule=schedule,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #4
| Source File: lars.py From pytorch-lars with MIT License | 6 votes |
|
def __init__(self, params, lr=required, momentum=.9,
weight_decay=.0005, eta=0.001, max_epoch=200):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {}".format(lr))
if momentum < 0.0:
raise ValueError("Invalid momentum value: {}".format(momentum))
if weight_decay < 0.0:
raise ValueError("Invalid weight_decay value: {}"
.format(weight_decay))
if eta < 0.0:
raise ValueError("Invalid LARS coefficient value: {}".format(eta))
self.epoch = 0
defaults = dict(lr=lr, momentum=momentum,
weight_decay=weight_decay,
eta=eta, max_epoch=max_epoch)
super(LARS, self).__init__(params, defaults)
Example #5
| Source File: optimization.py From project-purifier with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01, max_grad_norm=1.0, **kwargs):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if not isinstance(schedule, _LRSchedule) and schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
# initialize schedule object
if not isinstance(schedule, _LRSchedule):
schedule_type = SCHEDULES[schedule]
schedule = schedule_type(warmup=warmup, t_total=t_total)
else:
if warmup != -1 or t_total != -1:
logger.warning("warmup and t_total on the optimizer are ineffective when _LRSchedule object is provided as schedule. "
"Please specify custom warmup and t_total in _LRSchedule object.")
defaults = dict(lr=lr, schedule=schedule,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #6
| Source File: optimization.py From FlowDelta with MIT License | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #7
| Source File: optimization.py From GPT2sQA with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01, max_grad_norm=1.0, **kwargs):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if not isinstance(schedule, _LRSchedule) and schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
# initialize schedule object
if not isinstance(schedule, _LRSchedule):
schedule_type = SCHEDULES[schedule]
schedule = schedule_type(warmup=warmup, t_total=t_total)
else:
if warmup != -1 or t_total != -1:
logger.warning("warmup and t_total on the optimizer are ineffective when _LRSchedule object is provided as schedule. "
"Please specify custom warmup and t_total in _LRSchedule object.")
defaults = dict(lr=lr, schedule=schedule,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(GPT2Adam, self).__init__(params, defaults)
Example #8
| Source File: optimization.py From crosentgec with GNU General Public License v3.0 | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #9
| Source File: optimization.py From pytorch_pretrained_BERT with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #10
| Source File: optimization.py From squash-generation with MIT License | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01, max_grad_norm=1.0, **kwargs):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if not isinstance(schedule, _LRSchedule) and schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
# initialize schedule object
if not isinstance(schedule, _LRSchedule):
schedule_type = SCHEDULES[schedule]
schedule = schedule_type(warmup=warmup, t_total=t_total)
else:
if warmup != -1 or t_total != -1:
logger.warning("warmup and t_total on the optimizer are ineffective when _LRSchedule object is provided as schedule. "
"Please specify custom warmup and t_total in _LRSchedule object.")
defaults = dict(lr=lr, schedule=schedule,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #11
| Source File: optimization_openai.py From squash-generation with MIT License | 6 votes |
|
def __init__(self, params, lr=required, schedule='warmup_linear', warmup=-1, t_total=-1,
b1=0.9, b2=0.999, e=1e-8, weight_decay=0,
vector_l2=False, max_grad_norm=-1, **kwargs):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if not isinstance(schedule, _LRSchedule) and schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
# initialize schedule object
if not isinstance(schedule, _LRSchedule):
schedule_type = SCHEDULES[schedule]
schedule = schedule_type(warmup=warmup, t_total=t_total)
else:
if warmup != -1 or t_total != -1:
logger.warning("warmup and t_total on the optimizer are ineffective when _LRSchedule object is provided as schedule. "
"Please specify custom warmup and t_total in _LRSchedule object.")
defaults = dict(lr=lr, schedule=schedule,
b1=b1, b2=b2, e=e, weight_decay=weight_decay, vector_l2=vector_l2,
max_grad_norm=max_grad_norm)
super(OpenAIAdam, self).__init__(params, defaults)
Example #12
| Source File: optimization.py From HPSG-Neural-Parser with MIT License | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #13
| Source File: optimization.py From ERNIE with MIT License | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #14
| Source File: optimization.py From slot_filling_and_intent_detection_of_SLU with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01, max_grad_norm=1.0, **kwargs):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if not isinstance(schedule, _LRSchedule) and schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
# initialize schedule object
if not isinstance(schedule, _LRSchedule):
schedule_type = SCHEDULES[schedule]
schedule = schedule_type(warmup=warmup, t_total=t_total)
else:
if warmup != -1 or t_total != -1:
logger.warning("warmup and t_total on the optimizer are ineffective when _LRSchedule object is provided as schedule. "
"Please specify custom warmup and t_total in _LRSchedule object.")
defaults = dict(lr=lr, schedule=schedule,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #15
| Source File: optimization.py From ChID-Dataset with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #16
| Source File: optimization.py From bert_on_stilts with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #17
| Source File: optimization.py From HPSG-Neural-Parser with MIT License | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #18
| Source File: optimization.py From RCZoo with MIT License | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01, max_grad_norm=1.0, **kwargs):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if not isinstance(schedule, _LRSchedule) and schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
# initialize schedule object
if not isinstance(schedule, _LRSchedule):
schedule_type = SCHEDULES[schedule]
schedule = schedule_type(warmup=warmup, t_total=t_total)
else:
if warmup != -1 or t_total != -1:
logger.warning("warmup and t_total on the optimizer are ineffective when _LRSchedule object is provided as schedule. "
"Please specify custom warmup and t_total in _LRSchedule object.")
defaults = dict(lr=lr, schedule=schedule,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #19
| Source File: optimization.py From unilm with MIT License | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear', b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01, max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError(
"Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError(
"Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError(
"Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError(
"Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError(
"Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #20
| Source File: optimization_openai.py From PPLM with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, schedule='warmup_linear', warmup=-1, t_total=-1,
b1=0.9, b2=0.999, e=1e-8, weight_decay=0,
vector_l2=False, max_grad_norm=-1, **kwargs):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {}".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {}".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {}".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay, vector_l2=vector_l2,
max_grad_norm=max_grad_norm)
super(OpenAIAdam, self).__init__(params, defaults)
Example #21
| Source File: Bert_optimization.py From TriB-QA with MIT License | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #22
| Source File: optimization.py From PPLM with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #23
| Source File: optimization.py From fast-bert with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01, max_grad_norm=1.0, **kwargs):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if not isinstance(schedule, _LRSchedule) and schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
# initialize schedule object
if not isinstance(schedule, _LRSchedule):
schedule_type = SCHEDULES[schedule]
schedule = schedule_type(warmup=warmup, t_total=t_total)
else:
if warmup != -1 or t_total != -1:
logger.warning("warmup and t_total on the optimizer are ineffective when _LRSchedule object is provided as schedule. "
"Please specify custom warmup and t_total in _LRSchedule object.")
defaults = dict(lr=lr, schedule=schedule,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #24
| Source File: optimization.py From lxmert with MIT License | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #25
| Source File: downpour_sgd.py From distbelief with GNU General Public License v3.0 | 6 votes |
|
def __init__(self, params, lr=required, n_push=required, n_pull=required, model=required):
"""__init__
:param params:
:param lr:
:param freq:
:param model:
"""
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {}".format(lr))
defaults = dict(lr=lr,)
self.accumulated_gradients = torch.zeros(ravel_model_params(model).size())
self.n_pull = n_pull
self.n_push = n_push
self.model = model
# this sets the initial model parameters
send_message(MessageCode.ParameterUpdate, ravel_model_params(self.model))
self.idx = 0
listener = DownpourListener(self.model)
listener.start()
super(DownpourSGD, self).__init__(params, defaults)
Example #26
| Source File: optimization.py From KagNet with MIT License | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #27
| Source File: optimization_openai.py From KagNet with MIT License | 6 votes |
|
def __init__(self, params, lr=required, schedule='warmup_linear', warmup=-1, t_total=-1,
b1=0.9, b2=0.999, e=1e-8, weight_decay=0,
vector_l2=False, max_grad_norm=-1, **kwargs):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {}".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {}".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {}".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay, vector_l2=vector_l2,
max_grad_norm=max_grad_norm)
super(OpenAIAdam, self).__init__(params, defaults)
Example #28
| Source File: optimization.py From ZEN with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01, max_grad_norm=1.0, **kwargs):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if not isinstance(schedule, _LRSchedule) and schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
# initialize schedule object
if not isinstance(schedule, _LRSchedule):
schedule_type = SCHEDULES[schedule]
schedule = schedule_type(warmup=warmup, t_total=t_total)
else:
if warmup != -1 or t_total != -1:
logger.warning(
"warmup and t_total on the optimizer are ineffective when _LRSchedule object is provided as schedule. "
"Please specify custom warmup and t_total in _LRSchedule object.")
defaults = dict(lr=lr, schedule=schedule,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #29
| Source File: optimization.py From cudaBERT with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear',
b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01,
max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError("Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError("Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError("Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
Example #30
| Source File: optimization.py From VLP with Apache License 2.0 | 6 votes |
|
def __init__(self, params, lr=required, warmup=-1, t_total=-1, schedule='warmup_linear', b1=0.9, b2=0.999, e=1e-6, weight_decay=0.01, max_grad_norm=1.0):
if lr is not required and lr < 0.0:
raise ValueError(
"Invalid learning rate: {} - should be >= 0.0".format(lr))
if schedule not in SCHEDULES:
raise ValueError("Invalid schedule parameter: {}".format(schedule))
if not 0.0 <= warmup < 1.0 and not warmup == -1:
raise ValueError(
"Invalid warmup: {} - should be in [0.0, 1.0[ or -1".format(warmup))
if not 0.0 <= b1 < 1.0:
raise ValueError(
"Invalid b1 parameter: {} - should be in [0.0, 1.0[".format(b1))
if not 0.0 <= b2 < 1.0:
raise ValueError(
"Invalid b2 parameter: {} - should be in [0.0, 1.0[".format(b2))
if not e >= 0.0:
raise ValueError(
"Invalid epsilon value: {} - should be >= 0.0".format(e))
defaults = dict(lr=lr, schedule=schedule, warmup=warmup, t_total=t_total,
b1=b1, b2=b2, e=e, weight_decay=weight_decay,
max_grad_norm=max_grad_norm)
super(BertAdam, self).__init__(params, defaults)
