torch.empty()
for speed improvements
#9025
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
π οΈ PR Summary
Made with β€οΈ by Ultralytics Actions
π Summary
Optimization of tensor initialization in YOLOv5 by using
torch.empty
instead oftorch.zeros
.π Key Changes
torch.zeros
withtorch.empty
for initializing tensors in various files includingmodels/common.py
,models/yolo.py
,utils/autobatch.py
,utils/loggers/__init__.py
, andutils/torch_utils.py
.π― Purpose & Impact
torch.empty
allocates memory without initializing it, which can be slightly faster than initializing to zero withtorch.zeros
.