fixed nccl init

This commit is contained in:
William Falcon
2019-07-12 16:35:20 -04:00
parent b7baa96186
commit 9f41a9e8b7
+1 -5
View File
@@ -378,11 +378,7 @@ class Trainer(TrainerIO):
port = 12910
os.environ['MASTER_PORT'] = f'{port}'
print('-'*100)
print(f'PORT: {port}')
print('-'*100)
sleep(self.proc_rank * 2)
sleep(self.proc_rank*0.5)
root_node = os.environ['SLURM_NODELIST'].split(' ')[0]
os.environ['MASTER_ADDR'] = root_node