From c84700814d2fc7cb3697909d174caa5453579f47 Mon Sep 17 00:00:00 2001 From: William Falcon Date: Fri, 12 Jul 2019 16:03:17 -0400 Subject: [PATCH] fixed nccl init --- pytorch_lightning/models/trainer.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/pytorch_lightning/models/trainer.py b/pytorch_lightning/models/trainer.py index d7b0b9c0..3520f23e 100644 --- a/pytorch_lightning/models/trainer.py +++ b/pytorch_lightning/models/trainer.py @@ -379,8 +379,8 @@ 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") + dist.init_process_group("nccl", rank=self.proc_rank) + # dist.init_process_group("nccl") def __run_pretrain_routine(self, model): """