File size: 1,095 Bytes
d008243 |
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 |
import math
from typing import List
import torch
from torch.optim import Optimizer
__all__ = ["CosineLRwithWarmup"]
class CosineLRwithWarmup(torch.optim.lr_scheduler._LRScheduler):
def __init__(
self,
optimizer: Optimizer,
warmup_steps: int,
warmup_lr: float,
decay_steps: int,
last_epoch: int = -1,
) -> None:
self.warmup_steps = warmup_steps
self.warmup_lr = warmup_lr
self.decay_steps = decay_steps
super().__init__(optimizer, last_epoch)
def get_lr(self) -> List[float]:
if self.last_epoch < self.warmup_steps:
return [
(base_lr - self.warmup_lr) * self.last_epoch / self.warmup_steps
+ self.warmup_lr
for base_lr in self.base_lrs
]
else:
current_steps = self.last_epoch - self.warmup_steps
return [
0.5
* base_lr
* (1 + math.cos(math.pi * current_steps / self.decay_steps))
for base_lr in self.base_lrs
]
|