Skip to content

fix: apply QK norm in MultiheadAttention hook methods #85

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
orrzohar wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix: apply QK norm in MultiheadAttention hook methods #85

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 4 additions & 2 deletions src/cerebras/modelzoo/layers/AttentionLayer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -590,11 +590,13 @@ def apply_rotary_position_embedding(
return vector

def process_q_before_logits_calc(self, q):
# May get overriden by other attention schemas
if self.q_norm is not None:
q = self.q_norm(q)
return q

def process_k_before_logits_calc(self, k):
# May get overriden by other attention schemas
if self.k_norm is not None:
k = self.k_norm(k)
return k

def process_v_before_logits_calc(self, v):
Expand Down