Skip to content
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

Fix layerwise targets #36

Merged
merged 1 commit into from
Jul 24, 2024
Merged

Fix layerwise targets #36

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
7 changes: 4 additions & 3 deletions src/llmcompressor/modifiers/quantization/gptq/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,6 +94,7 @@ class GPTQModifier(Modifier):

sequential_update: Optional[bool] = False
targets: Union[str, List[str], None] = None
sequential_targets: Union[str, List[str], None] = None
block_size: int = 128
quantize: Union[bool, Dict] = True
dampening_frac: Optional[float] = 0.01
Expand Down Expand Up @@ -177,11 +178,11 @@ def on_initialize(self, state: "State", **kwargs) -> bool:
modifiable_model = state.model
calibration_dataloader = state.data.calib

if self.targets is None:
if self.sequential_targets is None:
# if no targets are provided, default to the modules that shouldn't be
# split by FSDP. For Transformers models this is equivalent to the
# decoder layers (ie LlamaDecoderLayer)
self.targets = get_no_split_params(modifiable_model)
self.sequential_targets = get_no_split_params(modifiable_model)

self.initialize_compression(modifiable_model, calibration_dataloader)
self.apply_compression(calibration_dataloader)
Expand Down Expand Up @@ -215,7 +216,7 @@ def compressible_layers(self) -> Dict:
f"{type(self.model)} instead"
)

return get_layers(self.targets, self.model)
return get_layers(self.sequential_targets, self.model)

def initialize_compression(
self,
Expand Down
Loading