Skip to content

Update: Fast GeLU Approximation #744

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

Merged
awni merged 5 commits into from
Feb 27, 2024
Merged

Update: Fast GeLU Approximation #744

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
f20853d
add: fast gelu approx
nkasmanoff Feb 26, 2024
5d94568
fix docs
nkasmanoff Feb 26, 2024
66fa314
Update gelu_fast_approx function documentation
nkasmanoff Feb 26, 2024
0252bc6
Update python/mlx/nn/layers/activations.py
nkasmanoff Feb 26, 2024
8e27752
fix: test gelu
nkasmanoff Feb 27, 2024
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
8 changes: 6 additions & 2 deletions python/mlx/nn/layers/activations.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -185,11 +185,15 @@ def gelu_fast_approx(x):

.. math::

x = x \sigma\left(1.773 x\right)
x = x \sigma\left(1.702 x\right)

where :math:`\sigma(\cdot)` is the logistic sigmoid.

References:
- https://github.com/hendrycks/GELUs
- https://arxiv.org/abs/1606.08415
"""
return x * mx.sigmoid(1.773 * x)
return x * mx.sigmoid(1.702 * x)
Copy link
Copy Markdown
Member

@awni awni Feb 26, 2024

Choose a reason for hiding this comment

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

IIRC @angeloskath added this, maybe you had another implementation in mind?



def glu(x: mx.array, axis: int = -1) -> mx.array:
Expand Down
2 changes: 1 addition & 1 deletion python/tests/test_nn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -665,7 +665,7 @@ def test_gelu(self):
y_hat1 = nn.gelu_approx(x)
y_hat2 = nn.gelu_fast_approx(x)
self.assertLess(mx.abs(y - y_hat1).max(), 0.0003)
self.assertLess(mx.abs(y - y_hat2).max(), 0.02)
self.assertLess(mx.abs(y - y_hat2).max(), 0.025)

def test_sin_pe(self):
m = nn.SinusoidalPositionalEncoding(16, min_freq=0.01)
Expand Down