Skip to content

Fix yolox kmac #35

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
sarhakor wants to merge 2 commits into
base: main
Choose a base branch
from
+29 −13
Open

Fix yolox kmac #35

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1ab7159
[fix] handle scalar KMAC returns in YOLOX
sarhakor Aug 1, 2026
5b54bd4
[fix] preserve CUDA device context for YOLOX inference and MAC comput…
sarhakor Aug 1, 2026
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
30 changes: 19 additions & 11 deletions compressai_vision/model_wrappers/yolox.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,6 +28,7 @@
# ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.


from contextlib import nullcontext
from enum import Enum
from pathlib import Path
from typing import Dict, List
Expand Down Expand Up @@ -184,19 +185,26 @@ def input_to_features(self, x, device: str) -> Dict:
def features_to_output(self, x: Dict, device: str):
"""Complete the downstream task from the intermediate deep features"""

self.model = self.model.to(device).eval()
target_device = torch.device(device)
device_context = (
torch.cuda.device(target_device)
if target_device.type == "cuda"
else nullcontext()
)

if self.split_id == self.SPLIT_L13:
return self._feature_at_l13_to_output(
x["data"], x["org_input_size"], x["input_size"], device
)
elif self.split_id == self.SPLIT_L37:
return self._feature_at_l37_to_output(
x["data"], x["org_input_size"], x["input_size"], device
)
else:
self.logger.error(f"Not supported split points {self.split_id}")
with device_context:
self.model = self.model.to(target_device).eval()

if self.split_id == self.SPLIT_L13:
return self._feature_at_l13_to_output(
x["data"], x["org_input_size"], x["input_size"], target_device
)
if self.split_id == self.SPLIT_L37:
return self._feature_at_l37_to_output(
x["data"], x["org_input_size"], x["input_size"], target_device
)

self.logger.error(f"Not supported split points {self.split_id}")
raise NotImplementedError

@torch.no_grad()
Expand Down
12 changes: 10 additions & 2 deletions compressai_vision/utils/measure_complexity.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -719,7 +719,11 @@ def calc_complexity_nn_part1_yolox(vision_model, img):

C, H, W = img.shape[1:]

kmacs, _ = measure_kmacs(partial_model, img)
if img.is_cuda:
with torch.cuda.device(img.device):
kmacs = measure_kmacs(partial_model, img)
else:
kmacs = measure_kmacs(partial_model, img)

pixels = reduce(operator.mul, [p_size for p_size in img.shape])
return kmacs, pixels
Expand All @@ -743,7 +747,11 @@ def calc_complexity_nn_part2_yolox(vision_model, dec_features):
C, H, W = input_tensor.shape[1:]
partial_model = YoloxPart2(vision_model, vision_model.split_id)

kmacs, _ = measure_kmacs(partial_model, input_tensor)
if input_tensor.is_cuda:
with torch.cuda.device(input_tensor.device):
kmacs = measure_kmacs(partial_model, input_tensor)
else:
kmacs = measure_kmacs(partial_model, input_tensor)

pixels = reduce(operator.mul, input_tensor.shape)

Expand Down