Skip to content

feat(pt): Add support for SiLU activation function in gradient calculations #5055

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
merged 3 commits into from
Nov 17, 2025
+103 −23
Merged

feat(pt): Add support for SiLU activation function in gradient calculations #5055

Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
dde7188
feat: Add support for SiLU activation function in gradient calculations
OutisLi Nov 15, 2025
620d15d
feat: Enhance activation function support in DPTabulate and tests
OutisLi Nov 15, 2025
61c3b9e
refactor: use `get_activation_fn` from `deepmd.dpmodel.utils.network`…
OutisLi Nov 16, 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
12 changes: 12 additions & 0 deletions deepmd/pt/utils/tabulate.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -84,6 +84,7 @@ def __init__(
"relu6": 4,
"softplus": 5,
"sigmoid": 6,
"silu": 7,
}

activation = activation_fn.activation
Expand Down Expand Up @@ -468,6 +469,11 @@ def grad(xbar: torch.Tensor, y: torch.Tensor, functype: int) -> torch.Tensor:
elif functype == 6:
return y * (1 - y)

elif functype == 7:
# silu'(x) = sigmoid(x) * (1 + x * (1 - sigmoid(x)))
sig = torch.sigmoid(xbar)
return sig + xbar * sig * (1 - sig)

else:
raise ValueError(f"Unsupported function type: {functype}")

Expand Down Expand Up @@ -495,6 +501,12 @@ def grad_grad(xbar: torch.Tensor, y: torch.Tensor, functype: int) -> torch.Tenso
elif functype == 6:
return y * (1 - y) * (1 - 2 * y)

elif functype == 7:
sig = torch.sigmoid(xbar)
d_sig = sig * (1 - sig)
# silu''(x) = 2 * d_sig + x * d_sig * (1 - 2 * sig)
return 2 * d_sig + xbar * d_sig * (1 - 2 * sig)

else:
return -torch.ones_like(xbar)

Expand Down
Loading