mirror of
https://github.com/saymrwulf/pytorch.git
synced 2026-05-14 20:57:59 +00:00
63219f1f9f
Summary: Fixes : https://github.com/pytorch/pytorch/issues/24892 In the paper : https://arxiv.org/pdf/1908.03265.pdf Liyuan Liu et al. suggested a new optimization algorithm with an essence of similar to Adam Algorithm. It has been discussed in the paper that, without warmup heuristic, in the early stage of adaptive optimization / learning algorithms sometimes we can get undesirable large variance which can slow overall convergence process. Authors proposed the idea of rectification of variance of adaptive learning rate when it is expected to be high. Differing from the paper, we selected variance tractability cut-off as 5 instead of 4. This adjustment is common practice, and could be found in the code-repository and also tensorflow swift optim library as well :2f03dd1970/radam/radam.py (L156)f51ee4618d/Sources/TensorFlow/Optimizers/MomentumBased.swift (L638)Pull Request resolved: https://github.com/pytorch/pytorch/pull/58968 Reviewed By: vincentqb Differential Revision: D29310601 Pulled By: iramazanli fbshipit-source-id: b7bd487f72f1074f266687fd9c0c6be264a748a9
16 lines
596 B
Python
16 lines
596 B
Python
from . import swa_utils as swa_utils
|
|
from . import lr_scheduler as lr_scheduler
|
|
from .adadelta import Adadelta as Adadelta
|
|
from .adagrad import Adagrad as Adagrad
|
|
from .adam import Adam as Adam
|
|
from .adamax import Adamax as Adamax
|
|
from .adamw import AdamW as AdamW
|
|
from .asgd import ASGD as ASGD
|
|
from .lbfgs import LBFGS as LBFGS
|
|
from .nadam import NAdam as NAdam
|
|
from .optimizer import Optimizer as Optimizer
|
|
from .radam import RAdam as RAdam
|
|
from .rmsprop import RMSprop as RMSprop
|
|
from .rprop import Rprop as Rprop
|
|
from .sgd import SGD as SGD
|
|
from .sparse_adam import SparseAdam as SparseAdam
|