Skip to content

fix(spin): fix tied embedding corruption in silent_run() and unconditional untie/fuse in run() #319

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

fix(spin): fix tied embedding corruption in silent_run() and unconditional untie/fuse in run() #319

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
9 changes: 4 additions & 5 deletions angelslim/compressor/transform/rotation/spin.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -156,6 +156,7 @@ def _get(key, default):

def silent_run(self):
"""only set linears and Rotations"""
self._untie_word_embeddings()
self._apply_fused_ln()
if "R1" in self.spin_config.rotation:
self._apply_r1(no_transform=True)
Expand All @@ -167,11 +168,11 @@ def silent_run(self):
self._apply_r4(no_transform=True)

def run(self):
self._untie_word_embeddings()
# fuse norm
self._apply_fused_ln()
# add rotation
if "R1" in self.spin_config.rotation:
self._untie_word_embeddings()
# fuse norm
self._apply_fused_ln()
self._apply_r1()
if "R2" in self.spin_config.rotation:
self._apply_r2()
Expand Down Expand Up @@ -389,8 +390,6 @@ def _apply_fused_ln(self):
"""
self.logger.info("Applying fused layer norm to a linear layer")

self._untie_word_embeddings()

norm_layers = self.quant_model.get_rotation_mapping_layers(
None,
norm_mapping=self.norm_mapping,
Expand Down