Skip to content

[Bugfix] Fix for ROCM compressed tensor support #11561

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
Dec 27, 2024
Merged

[Bugfix] Fix for ROCM compressed tensor support #11561

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 7 additions & 3 deletions ...el_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_fp8.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,10 +41,12 @@ def process_weights_after_loading(self, layer) -> None:
)

if current_platform.is_rocm():
input_scale = getattr(layer, 'input_scale', None)

weight, max_w_scale, input_scale = normalize_e4m3fn_to_e4m3fnuz(
weight=weight,
weight_scale=max_w_scale,
input_scale=layer.input_scale)
input_scale=input_scale)
if input_scale is not None:
layer.input_scale = Parameter(input_scale,
requires_grad=False)
Expand All @@ -57,11 +59,13 @@ def process_weights_after_loading(self, layer) -> None:
weight = layer.weight

if current_platform.is_rocm():
input_scale = getattr(layer, 'input_scale', None)

weight, weight_scale, input_scale = \
normalize_e4m3fn_to_e4m3fnuz(
weight=weight,
weight_scale=layer.weight_scale,
input_scale=layer.input_scale)
input_scale=input_scale)
if input_scale is not None:
layer.input_scale = Parameter(input_scale,
requires_grad=False)
Expand All @@ -76,7 +80,7 @@ def process_weights_after_loading(self, layer) -> None:
raise ValueError(f"Unknown quantization strategy {self.strategy}")

# INPUT SCALE
if self.is_static_input_scheme:
if self.is_static_input_scheme and hasattr(layer, 'input_scale'):
layer.input_scale = Parameter(layer.input_scale.max(),
requires_grad=False)
else:
Expand Down
Loading