Lint schedulers
Browse files
src/axolotl/utils/schedulers.py
CHANGED
@@ -1,7 +1,13 @@
|
|
|
|
|
|
1 |
from torch.optim.lr_scheduler import LRScheduler
|
2 |
|
3 |
|
4 |
class InterpolatingLogScheduler(LRScheduler):
|
|
|
|
|
|
|
|
|
5 |
def __init__(self, optimizer, num_steps, min_lr, max_lr, last_epoch=-1):
|
6 |
"""A scheduler that interpolates learning rates in a logarithmic fashion
|
7 |
|
@@ -19,7 +25,9 @@ class InterpolatingLogScheduler(LRScheduler):
|
|
19 |
self.num_steps = num_steps
|
20 |
self.min_lr = min_lr
|
21 |
self.max_lr = max_lr
|
22 |
-
self.q = (max_lr / min_lr) ** (
|
|
|
|
|
23 |
super().__init__(optimizer, last_epoch)
|
24 |
|
25 |
def get_lr(self):
|
|
|
1 |
+
"""Module for custom LRScheduler class"""
|
2 |
+
|
3 |
from torch.optim.lr_scheduler import LRScheduler
|
4 |
|
5 |
|
6 |
class InterpolatingLogScheduler(LRScheduler):
|
7 |
+
"""
|
8 |
+
A scheduler that interpolates learning rates in a logarithmic fashion
|
9 |
+
"""
|
10 |
+
|
11 |
def __init__(self, optimizer, num_steps, min_lr, max_lr, last_epoch=-1):
|
12 |
"""A scheduler that interpolates learning rates in a logarithmic fashion
|
13 |
|
|
|
25 |
self.num_steps = num_steps
|
26 |
self.min_lr = min_lr
|
27 |
self.max_lr = max_lr
|
28 |
+
self.q = (max_lr / min_lr) ** ( # pylint: disable=invalid-name
|
29 |
+
1 / (num_steps - 1)
|
30 |
+
)
|
31 |
super().__init__(optimizer, last_epoch)
|
32 |
|
33 |
def get_lr(self):
|