Skip to content

Additional fixes for autoquant serialization #1486

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 1 commit into from
Jan 3, 2025
Merged

Additional fixes for autoquant serialization #1486

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
11 changes: 6 additions & 5 deletions torchao/quantization/autoquant.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -411,7 +411,7 @@ class AQInt8DynamicallyQuantizedLinearWeight(AQMixin, LinearActivationQuantizedT
AutoQuantizable version of Int8DynamicallyQuantizedLinearWeight
"""

layout: Layout = PlainLayout()
aq_layout: Layout = PlainLayout()

@classmethod
def from_float(cls, weight):
Expand Down Expand Up @@ -442,7 +442,7 @@ def get_weight_block_size(x):
target_dtype = torch.int8
eps = torch.finfo(torch.float32).eps
zero_point_dtype = torch.int64
_layout = cls.layout
_layout = cls.aq_layout
block_size = get_weight_block_size(weight)

weight = to_affine_quantized_intx(
Expand Down Expand Up @@ -616,12 +616,13 @@ class AQInt4G32WeightOnlyQuantizedLinearWeight(AffineQuantizedTensor, AQMixin):
"""

group_size: int = 32
layout: Layout = TensorCoreTiledLayout(inner_k_tiles=8)
# can't override the `layout` attribute
aq_layout: Layout = TensorCoreTiledLayout(inner_k_tiles=8)

@classmethod
def from_float(cls, weight):
group_size = cls.group_size
_layout = cls.layout
_layout = cls.aq_layout

if weight.shape[-1] % group_size != 0:
return weight
Expand Down Expand Up @@ -681,7 +682,7 @@ class AQInt4G128WeightOnlyQuantizedMarlinSparseLinearWeight(
AQInt4G32WeightOnlyQuantizedLinearWeight
):
group_size: int = 128
layout: Layout = MarlinSparseLayout()
aq_layout: Layout = MarlinSparseLayout()


class AQGemliteInt4G32WeightOnlyQuantizedLinearWeight(AffineQuantizedTensor, AQMixin):
Expand Down
Loading