Skip to content

Port auto-detect absorb layers for TEQ #1895

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
merged 3 commits into from
Jul 4, 2024
Merged

Port auto-detect absorb layers for TEQ #1895

Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3996d8d
add ut
yiliu30 Jun 28, 2024
d8ce3e0
disable some check
yiliu30 Jul 1, 2024
9cb3a63
Merge branch 'master' into teq_absorb
yiliu30 Jul 3, 2024
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
Prev Previous commit
Next Next commit
disable some check 
Signed-off-by: yiliu30 <[email protected]>
  • Loading branch information
@yiliu30
yiliu30 committed Jul 1, 2024
commit d8ce3e0d4e2ef890e02c6b93895a29351d2a2b26
8 changes: 5 additions & 3 deletions neural_compressor/torch/algorithms/weight_only/teq.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -63,7 +63,7 @@ def _detect_absorb_to_layer(self, model, folding, example_inputs):

tg = GraphTrace()
detected_absorb_layers, _ = tg.get_absorb_to_layer(model, example_inputs, supported_layers)
else:
else: # pragma: no cover
for name, module in model.named_modules():
if module.__class__.__name__ in supported_layers:
detected_absorb_layers[name] = [name]
Expand Down Expand Up @@ -209,7 +209,9 @@ def _absorb_scales(self, layer, scale, layer_name=""):
scale = scale.view(scale.shape[0], 1)
layer.weight *= scale

elif layer.__class__.__name__ == "LlamaRMSNorm" or layer.__class__.__name__ == "T5LayerNorm": ##quite tricky
elif (
layer.__class__.__name__ == "LlamaRMSNorm" or layer.__class__.__name__ == "T5LayerNorm"
): # pragma: no cover
layer.weight *= scale

else: # pragma: no cover
Expand Down Expand Up @@ -241,7 +243,7 @@ def _scale_layer_weight(self, layer, scale): ##input channel
@torch.no_grad()
def transform(self):
"""Apply alpha/scale."""
if not self._post_initialized:
if not self._post_initialized: # pragma: no cover
self._post_init()
for ln_name, layer_names in self.absorb_to_layer.items():
module = get_module(self.model, ln_name)
Expand Down