From 4f5eef2e7835dd86449a8e242bf56a1adda86e31 Mon Sep 17 00:00:00 2001 From: William Falcon Date: Fri, 12 Jul 2019 15:59:12 -0400 Subject: [PATCH] fixed nccl init --- pytorch_lightning/models/trainer.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/pytorch_lightning/models/trainer.py b/pytorch_lightning/models/trainer.py index 3bf2c1e888..d84ea554bf 100644 --- a/pytorch_lightning/models/trainer.py +++ b/pytorch_lightning/models/trainer.py @@ -386,7 +386,8 @@ class Trainer(TrainerIO): root_node = os.environ['SLURM_NODELIST'].split(' ')[0] os.environ['MASTER_ADDR'] = root_node os.environ['MASTER_PORT'] = f'{port}' - dist.init_process_group("nccl", rank=self.proc_rank, world_size=self.world_size) + # dist.init_process_group("nccl", rank=self.proc_rank, world_size=self.world_size) + dist.init_process_group("nccl") def __run_pretrain_routine(self, model): """