Skip to content

[ascend] optimize gated rmsnorm #324

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
wanfengcxz wants to merge 2 commits into
base: main
Choose a base branch
from
Open

[ascend] optimize gated rmsnorm #324

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b874acf
[ascend] optimize gated rmsnorm
wanfengcxz Apr 20, 2026
f6ac78b
[ascend] fix import error
wanfengcxz Apr 20, 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
14 changes: 5 additions & 9 deletions dlinfer/vendor/ascend/triton_ops/rms_norm_gated.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@
import torch.nn as nn
import triton
import triton.language as tl
import torch.nn.functional as F

MAX_CORES = 65535

Expand Down Expand Up @@ -164,12 +165,7 @@ def reset_parameters(self):
torch.nn.init.ones_(self.weight)

def forward(self, x, z=None):
return rmsnorm_fn(
x,
self.weight,
self.bias,
z=z,
eps=self.eps,
group_size=self.group_size,
norm_before_gate=self.norm_before_gate,
)
input_dtype = x.dtype
x = torch.ops.npu.npu_rms_norm(x, self.weight, self.eps)[0]
out = x * F.silu(z.to(torch.float32))
return out.to(input_dtype)
Loading