pytorch
b3bf08e6 - Log nccl debug level in ProcessGroupNCCL (#52803)

Commit
5 years ago
Log nccl debug level in ProcessGroupNCCL (#52803) Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/52803 This is useful for double checking we have the expected nccl_debug level when debugging problematic jobs. New logs: When default is warn: ``` NCCL_ASYNC_ERROR_HANDLING: 0 NCCL_BLOCKING_WAIT: 0 TIMEOUT(ms): 60000 USE_HIGH_PRIORITY_STREAM: 0 NCCL_DEBUG: WARN ``` off: ``` NCCL_ASYNC_ERROR_HANDLING: 0 NCCL_BLOCKING_WAIT: 0 TIMEOUT(ms): 1800000 USE_HIGH_PRIORITY_STREAM: 0 NCCL_DEBUG: N/A ``` ghstack-source-id: 122751110 Test Plan: CI Reviewed By: pritamdamania87 Differential Revision: D26653699 fbshipit-source-id: 845cc1236f3838f4763c6dcf2a30d059b3d44f02
Author
Parents
Loading