NVIDIA / NVIDIA/apex

Difference between cuda backend syncbn and Python version

Open
#669 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
9k
Forks
1.5k
Avg merge
2d 4h
Merged PRs (30d)
3

Description

“Warning: using Python fallback for SyncBatchNorm, possibly because apex was installed without --cuda_ext. The exception raised when attempting to import the cuda backend was: No module named 'syncbn'“

I wonder the speed difference of both, how slower is the Python verson. Have you test for the specific tasks. Thanks!

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

No file, test, or benchmark task is named. Start by locating the SyncBatchNorm CUDA backend and Python fallback referenced in the warning, then identify representative tasks for comparison. Done means the relative speed difference is measured and documented with enough context to interpret the results.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
machine-learning, performance
Issue type
Documentation
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.