Skip to content

Fix for Llama4 Maverick performance drop #904

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
wpyszka merged 3 commits into from
Feb 2, 2026
Merged

Fix for Llama4 Maverick performance drop #904

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
2c837d2
Revert part of #758 to fix Llama4 Maverick
jkaniecki Jan 30, 2026
c4943dd
Update hpu_model_runner.py
jkaniecki Feb 2, 2026
634e1f2
Update hpu_model_runner.py
jkaniecki Feb 2, 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
8 changes: 5 additions & 3 deletions vllm_gaudi/v1/worker/hpu_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -380,12 +380,14 @@ def patch_llama4_get_attn_scale(model):
continue

attn = layer.self_attn
orig = attn._get_attn_scale

def _get_attn_scale_for_hpu(self, positions, _orig=orig):
def _get_attn_scale_for_hpu(self, positions):
if self.qk_norm is not None:
positions = positions.flatten()
return _orig(positions)
floor = torch.floor((positions + 1.0) / self.floor_scale)
attn_scale = torch.log(floor + 1.0) * self.attn_scale + 1.0

return attn_scale.unsqueeze(-1)

attn._get_attn_scale = types.MethodType(_get_attn_scale_for_hpu, attn)

Expand Down