Skip to content

[Adam/AdamW] Update adamw.py #216

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

[Adam/AdamW] Update adamw.py #216

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions plsc/optimizer/adamw.py
View file
Open in desktop
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

另外，貌似应该把 moment2_max 为 None 也传进去，如：

In [45]: out = _legacy_C_ops.adamw(
    ...:     param,
    ...:     grad,
    ...:     lr,
    ...:     moment1,
    ...:     moment2,
    ...:     None,   # `momen2_max` ，如果不需要 `amsgrad`，则传入 `None`
    ...:     beta1_pow,
    ...:     beta2_pow,
    ...:     None,   # `master_param` 根据实际需要传入参数
    ...:     param,
    ...:     moment1,
    ...:     moment2,
    ...:     None,   # `momen2_max` ，如果不需要 `amsgrad`，则传入 `None`
    ...:     beta1_pow,
    ...:     beta2_pow,
    ...:     None,   # `master_param` 根据实际需要传入参数
    ...:     'epsilon',
    ...:     epsilon,
    ...:     'lazy_mode',
    ...:     False,
    ...:     'min_row_size_to_use_multithread',
    ...:     1000,
    ...:     'beta1',
    ...:     beta1,
    ...:     'beta2',
    ...:     beta2,
    ...:     "with_decay",
    ...:     True,
    ...:     'coeff',
    ...:     0.5,
    ...:     'multi_precision',
    ...:     False,
    ...:     'lr_ratio',
    ...:     1.0,
    ...:     'amsgrad',  # `amsgrad` 参数
    ...:     False,      # `amsgrad` 参数
    ...: )

In [46]: out
Out[46]: 
(Tensor(shape=[102, 105], dtype=float32, place=Place(gpu:0), stop_gradient=True,
        [[ 0.20898449,  0.89920276,  0.67242330, ...,  0.26126957,
           0.79362839,  0.82994431],
         [-0.41922054, -0.49964213, -0.72876191, ...,  0.64584875,
           0.38303095,  0.07835867],
         [ 0.82518733, -0.13006617, -0.18193051, ..., -0.83834726,
          -0.48943013,  0.28921935],
         ...,
         [-0.11833674, -0.87520474,  0.71153826, ...,  0.88105798,
          -0.84247899, -0.03978884],
         [ 0.03530697, -0.51926482, -0.60509771, ..., -0.93831873,
          -0.40703350,  0.06399230],
         [-0.96511489,  0.76393193,  0.27214301, ..., -0.11625432,
          -0.12905845, -0.89011657]]),
 Tensor(shape=[102, 105], dtype=float32, place=Place(gpu:0), stop_gradient=True,
        [[ 0.11505818, -0.12621984, -0.01848486, ...,  0.22509800,
          -0.09953045,  0.56122953],
         [ 0.05467438,  0.15807387, -0.00716698, ..., -0.16195901,
          -0.27719164,  0.73311120],
         [-0.21436734, -0.12942475, -0.19450223, ...,  0.26065761,
          -0.09446586,  0.61139995],
         ...,
         [ 0.01895352, -0.44893426, -0.35492349, ...,  0.78992486,
          -0.46006823,  0.70998996],
         [-0.83486271,  0.08806995, -0.31217605, ..., -0.04586679,
           0.63772619, -0.62238657],
         [ 0.40758044,  0.40442133,  0.29918492, ...,  0.60868609,
           0.73768240,  0.27699226]]),
 Tensor(shape=[102, 105], dtype=float32, place=Place(gpu:0), stop_gradient=True,
        [[0.04723934, 0.06505238, 0.23348683, ..., 0.10994066, 0.22805302,
          0.44461903],
         [0.03345066, 0.11195458, 0.18727647, ..., 0.22885345, 0.15290459,
          0.71300763],
         [0.08683055, 0.22121914, 0.06403468, ..., 0.06969104, 0.13063110,
          0.51002014],
         ...,
         [0.11580127, 0.34216970, 0.38666964, ..., 0.77789527, 0.38964579,
          0.50754905],
         [0.68814623, 0.03124339, 0.27777275, ..., 0.22106472, 0.46024293,
          0.40622491],
         [0.32853597, 0.28381625, 0.27526388, ..., 0.47823176, 0.61193144,
          0.35178134]]),
 Tensor(Not initialized),
 Tensor(shape=[], dtype=float32, place=Place(gpu:0), stop_gradient=True,
        0.01142096),
 Tensor(shape=[], dtype=float32, place=Place(gpu:0), stop_gradient=True,
        0.03978442),
 Tensor(Not initialized))

对应的 moment2_max_out 为 Tensor(Not initialized) ～

Original file line number Diff line number Diff line change
Expand Up @@ -117,7 +117,7 @@ def step(self):
sub_exp_avg_sq = paddle.gather(
exp_avg_sq, index, axis=axis)

_, _, _, _, _, _ = _C_ops.adamw(
_, _, _, _, _, *_ = _C_ops.adamw(
sub_p, grad,
paddle.to_tensor(lr), sub_exp_avg, sub_exp_avg_sq,
beta1_pow, beta2_pow, master_param, sub_p, sub_exp_avg,
Expand All @@ -133,7 +133,7 @@ def step(self):
exp_avg_sq.scatter_(index, sub_exp_avg_sq)

else:
_, _, _, _, _, _ = _C_ops.adamw(
_, _, _, _, _, *_ = _C_ops.adamw(
p, grad,
paddle.to_tensor(lr), exp_avg, exp_avg_sq, beta1_pow,
beta2_pow, master_param, p, exp_avg, exp_avg_sq,
Expand Down