From 4f5da45fae36adb4ee8a1c6076d100038dcf0e46 Mon Sep 17 00:00:00 2001 From: William Falcon Date: Fri, 12 Jul 2019 16:17:50 -0400 Subject: [PATCH] fixed nccl init --- pytorch_lightning/models/trainer.py | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/pytorch_lightning/models/trainer.py b/pytorch_lightning/models/trainer.py index 679661e7..6e0e9591 100644 --- a/pytorch_lightning/models/trainer.py +++ b/pytorch_lightning/models/trainer.py @@ -381,6 +381,10 @@ class Trainer(TrainerIO): port = 12910 os.environ['MASTER_PORT'] = f'{port}' + print('-'*100) + print(f'PORT: {port}') + print('-'*100) + sleep(self.proc_rank * 2) root_node = os.environ['SLURM_NODELIST'].split(' ')[0]