Skip to content

[Bugfix] Make Gelu Activations consistent across frameworks #753

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
vrdn-23 wants to merge 10 commits into
base: main
Choose a base branch
from
Open

[Bugfix] Make Gelu Activations consistent across frameworks #753

Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
a3f6719
[Bugfix] Make Gelu Activations consistent across frameworks
vrdn-23 Nov 6, 2025
aae4307
simplify a bit further
vrdn-23 Nov 6, 2025
2ca266c
style change
vrdn-23 Nov 6, 2025
5ca56db
Merge remote-tracking branch 'origin/main' into vrdn-23/fix-gelu-acti…
vrdn-23 Jan 18, 2026
4f8c82b
Merge branch 'main' into vrdn-23/fix-gelu-activation
vrdn-23 Jan 20, 2026
c29a699
Merge remote-tracking branch 'origin/vrdn-23/fix-gelu-activation' int…
vrdn-23 Jan 20, 2026
d365667
Use `GeluErf` (for gelu) and `Gelu` (for the rest)
alvarobartt Jan 21, 2026
c8b5348
Merge branch 'main' into vrdn-23/fix-gelu-activation
alvarobartt Jan 21, 2026
088b46f
Fix `HiddenAct::{Gelu,GeluErf}`
alvarobartt Jan 21, 2026
d8685ef
Remove `rename_all` in favour of per-field renames
alvarobartt Jan 21, 2026
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
10 changes: 8 additions & 2 deletions backends/candle/src/layers/linear.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,18 +3,23 @@ use candle::{Device, Result, Tensor};
use serde::Deserialize;

#[derive(Debug, Deserialize, PartialEq, Clone)]
#[serde(rename_all = "lowercase")]
pub enum HiddenAct {
#[serde(alias = "gelu_pytorch_tanh")]
#[serde(rename = "gelu")]
GeluErf,
#[serde(alias = "gelu_new", alias = "gelu_pytorch_tanh")]
Gelu,
#[serde(alias = "relu")]
Relu,
#[serde(alias = "silu")]
Silu,
#[serde(rename = "swiglu")]
Swiglu,
}

impl HiddenAct {
pub fn forward(&self, x: &Tensor) -> Result<Tensor> {
match self {
Self::GeluErf => x.gelu_erf(),
Self::Gelu => x.gelu(),
Self::Relu => x.relu(),
Self::Silu => x.silu(),
Expand Down Expand Up @@ -84,6 +89,7 @@ impl Linear {

if let Some(act) = &self.act {
match act {
HiddenAct::GeluErf => x.gelu_erf(),
HiddenAct::Gelu => x.gelu(),
HiddenAct::Relu => x.relu(),
HiddenAct::Silu => x.silu(),
Expand Down
Loading