Skip to content

Add support for passing args and kwargs to per-sample loss functions #786

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 2 commits into from
Closed

Add support for passing args and kwargs to per-sample loss functions #786

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c8f61d0
Add support for passing additional kwargs to per-sample loss functions
david-stan Sep 12, 2025
6c5e6e8
Refactor `__call__` methods to use `*args` for flexibility in argumen…
david-stan Sep 12, 2025
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: 3 additions & 3 deletions opacus/utils/adaptive_clipping/adaptive_clipping_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -207,14 +207,14 @@ def __init__(
self.clipbound_learning_rate = clipbound_learning_rate
self.initial_noise_multiplier = initial_noise_multiplier

def __call__(self, input, target) -> DPTensorFastGradientAdaptiveClipping:
def __call__(self, *args, **kwargs) -> DPTensorFastGradientAdaptiveClipping:
"""
Redefining the forward function to compute per-sample loss and wrap it in DPTensorFastGradientAdaptiveClipping
"""

loss_per_sample = self.criterion(
input,
target,
*args,
**kwargs,
)
return DPTensorFastGradientAdaptiveClipping(
self.module,
Expand Down
4 changes: 2 additions & 2 deletions opacus/utils/fast_gradient_clipping_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,12 +106,12 @@ def __init__(
self.loss_reduction = loss_reduction
self.criterion.reduction = "none"

def __call__(self, input, target, shape=None) -> DPTensorFastGradientClipping:
def __call__(self, *args, shape=None, **kwargs) -> DPTensorFastGradientClipping:
"""
Redefining the forward function to compute per-sample loss and wrap it in DPTensorFastGradientClipping
"""

loss_per_sample = self.criterion(input, target)
loss_per_sample = self.criterion(*args, **kwargs)

if shape is not None and loss_per_sample.shape[0] == shape[0] * shape[1]:
# Note that the privacy unit for generative NLP tasks is per sequence.
Expand Down
Loading