Skip to content

[Spyre-Next] Reworked rms_norm #873

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
tjohnson31415 merged 2 commits into from
Mar 27, 2026
Merged

[Spyre-Next] Reworked rms_norm #873

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d648ddc
Removed transpose + contiguous()
bohnstingl Mar 27, 2026
3b4c4b6
Removed variance_size_override path
bohnstingl Mar 27, 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
6 changes: 1 addition & 5 deletions vllm_spyre_next/vllm_spyre_next/custom_ops/rms_norm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -130,8 +130,6 @@ def forward_spyre(
if x.shape[-1] != hidden_size:
raise ValueError(f"Expected hidden_size to be {hidden_size}, but found: {x.shape[-1]}")

x = x.transpose(-1, -2).contiguous()

variance_epsilon = torch.full(
x.shape, variance_epsilon, dtype=torch.float16, device=x.device
)
Expand All @@ -147,11 +145,9 @@ def forward_spyre(

x_var = x[:, :, :variance_size_override]
Comment thread
bohnstingl marked this conversation as resolved.
Outdated

# After transpose, hidden dim is now dim=0
variance = x_var.pow(2).mean(dim=0, keepdim=True)
Comment thread
bohnstingl marked this conversation as resolved.
variance = x_var.pow(2).mean(dim=-1, keepdim=True)

x = x * torch.rsqrt(variance + variance_epsilon)
x = x.transpose(-1, -2).contiguous()

if weight is not None:
x = x * weight
Expand Down
Loading