Skip to content

[Adam] Fix complex x amsgrad support #104989

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

Closed
wants to merge 5 commits into from
Closed

[Adam] Fix complex x amsgrad support #104989

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
aa45948
[foreach][Adam] Fix complex x amsgrad support
janeyx99 Jul 11, 2023
cb4d97b
Update on "[foreach][Adam] Fix complex x amsgrad support"
janeyx99 Jul 11, 2023
ecfc627
Update on "[foreach][Adam] Fix complex x amsgrad support"
janeyx99 Jul 11, 2023
6bdbc5f
Update on "[foreach][Adam] Fix complex x amsgrad support"
janeyx99 Jul 11, 2023
b94ebcc
Update on "[foreach][Adam] Fix complex x amsgrad support"
janeyx99 Jul 21, 2023
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
6 changes: 4 additions & 2 deletions test/optim/test_optim.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1085,8 +1085,10 @@ def test_adam(self):
constructor_accepts_foreach=True,
)
self._test_complex_2d(optim.Adam)
self._test_complex_2d(functools.partial(optim.Adam, foreach=True))
self._test_complex_2d(functools.partial(optim.Adam, foreach=True, weight_decay=0.2))
self._test_complex_2d(functools.partial(optim.Adam, foreach=False))
self._test_complex_2d(functools.partial(optim.Adam, foreach=False, amsgrad=True))
self._test_complex_2d(functools.partial(optim.Adam, weight_decay=0.2))
self._test_complex_2d(functools.partial(optim.Adam, weight_decay=0.2, amsgrad=True))

with self.assertRaisesRegex(
ValueError, "Invalid beta parameter at index 0: 1.0"
Expand Down
16 changes: 13 additions & 3 deletions torch/optim/adam.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -355,6 +355,8 @@ def _single_tensor_adam(params: List[Tensor],
grad = torch.view_as_real(grad)
exp_avg = torch.view_as_real(exp_avg)
exp_avg_sq = torch.view_as_real(exp_avg_sq)
if amsgrad:
max_exp_avg_sqs[i] = torch.view_as_real(max_exp_avg_sqs[i])
param = torch.view_as_real(param)

# Decay the first and second moment running average coefficient
Expand All @@ -375,10 +377,12 @@ def _single_tensor_adam(params: List[Tensor],
if amsgrad:
# Maintains the maximum of all 2nd moment running avg. till now
if differentiable:
max_exp_avg_sqs_i = max_exp_avg_sqs[i].clone()
max_exp_avg_sq = max_exp_avg_sqs[i].clone()
else:
max_exp_avg_sqs_i = max_exp_avg_sqs[i]
max_exp_avg_sqs[i].copy_(torch.maximum(max_exp_avg_sqs_i, exp_avg_sq))
max_exp_avg_sq = max_exp_avg_sqs[i]

max_exp_avg_sqs[i].copy_(torch.maximum(max_exp_avg_sq, exp_avg_sq))

# Uses the max. for normalizing running avg. of gradient
# Folds in (admittedly ugly) 1-elem step_size math here to avoid extra param-set-sized read+write
# (can't fold it into addcdiv_ below because addcdiv_ requires value is a Number, not a Tensor)
Expand All @@ -400,13 +404,18 @@ def _single_tensor_adam(params: List[Tensor],
if amsgrad:
# Maintains the maximum of all 2nd moment running avg. till now
torch.maximum(max_exp_avg_sqs[i], exp_avg_sq, out=max_exp_avg_sqs[i])

# Use the max. for normalizing running avg. of gradient
denom = (max_exp_avg_sqs[i].sqrt() / bias_correction2_sqrt).add_(eps)
else:
denom = (exp_avg_sq.sqrt() / bias_correction2_sqrt).add_(eps)

param.addcdiv_(exp_avg, denom, value=-step_size)

# Lastly, switch back to complex view
if amsgrad and torch.is_complex(params[i]):
max_exp_avg_sqs[i] = torch.view_as_complex(max_exp_avg_sqs[i])


def _multi_tensor_adam(params: List[Tensor],
grads: List[Tensor],
Expand Down Expand Up @@ -456,6 +465,7 @@ def _multi_tensor_adam(params: List[Tensor],
device_grads = [torch.view_as_real(x) if torch.is_complex(x) else x for x in device_grads]
device_exp_avgs = [torch.view_as_real(x) if torch.is_complex(x) else x for x in device_exp_avgs]
device_exp_avg_sqs = [torch.view_as_real(x) if torch.is_complex(x) else x for x in device_exp_avg_sqs]
device_max_exp_avg_sqs = [torch.view_as_real(x) if torch.is_complex(x) else x for x in device_max_exp_avg_sqs]
device_params = [torch.view_as_real(x) if torch.is_complex(x) else x for x in device_params]

# update steps
Expand Down