* Fixing bug multi-gpu training This solves this issue: https://github.com/ultralytics/yolov5/issues/6297#issue-1103853348 * Update torch_utils.py for pep8
Henry