From c244599ae8c9575d34f7f25942ed06a1ccd87f2a Mon Sep 17 00:00:00 2001 From: William Falcon Date: Fri, 12 Jul 2019 15:59:33 -0400 Subject: [PATCH] fixed nccl init --- pytorch_lightning/models/trainer.py | 3 --- 1 file changed, 3 deletions(-) diff --git a/pytorch_lightning/models/trainer.py b/pytorch_lightning/models/trainer.py index 193ebc01..d7b0b9c0 100644 --- a/pytorch_lightning/models/trainer.py +++ b/pytorch_lightning/models/trainer.py @@ -375,9 +375,6 @@ class Trainer(TrainerIO): :param tries: :return: """ - # hack to get nccl to stop throwing error... seems to be an nccl race condition - if self.proc_rank > 0: - sleep(10.0) root_node = os.environ['SLURM_NODELIST'].split(' ')[0] os.environ['MASTER_ADDR'] = root_node