testing env init

2019-07-12 13:11:08 -04:00 · 2019-07-12 13:11:08 -04:00 · 5e033fd97a
parent 5d14b97aa6
commit 5e033fd97a
1 changed files with 4 additions and 1 deletions
--- a/pytorch_lightning/models/trainer.py
+++ b/pytorch_lightning/models/trainer.py
@ -352,7 +352,10 @@ class Trainer(TrainerIO):
        print('SLURM ROOT NODE: ', root_node)
        print('-'*100)

-        dist.init_process_group("nccl", init_method=f'env://{root_node}:12007', rank=self.proc_rank,
+        os.environ['MASTER_ADDR'] = root_node
+        os.environ['MASTER_PORT'] = '12006'
+
+        dist.init_process_group("nccl", rank=self.proc_rank,
                                world_size=self.world_size)
        # self.__init_tcp_connection(ip_file_dir)