Skip to content

Fix AffineQuantizedMinMaxObserver #4189

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
Freed-Wu wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix AffineQuantizedMinMaxObserver #4189

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: 3 additions & 6 deletions torchao/quantization/observer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -153,10 +153,7 @@ def forward(self, input: torch.Tensor):
input_detached = input_detached.view(shape_for_reduction)
min_val = torch.amin(input_detached, dim=reduction_dims, keepdim=self.keepdim)
max_val = torch.amax(input_detached, dim=reduction_dims, keepdim=self.keepdim)
if not hasattr(self, "min_val") or not hasattr(self, "max_val"):
self.min_val = min_val
self.max_val = max_val
else:
if hasattr(self, "min_val") and hasattr(self, "max_val"):
assert self.min_val.shape == min_val.shape, (
f"Can't update existing min_val - shape mismatch, self.min_val:{self.min_val.shape} != min_val:{min_val.shape}"
)
Expand All @@ -165,8 +162,8 @@ def forward(self, input: torch.Tensor):
)
min_val = torch.min(self.min_val, min_val)
max_val = torch.max(self.max_val, max_val)
self.min_val.copy_(min_val)
self.max_val.copy_(max_val)
self.min_val = min_val
self.max_val = max_val
# returning original input
return input

Expand Down