added multi-node proc 0 ip reading

2019-07-08 09:36:27 -04:00 · 2019-07-08 09:36:27 -04:00 · 79d9adf004
parent f705f15c7a
commit 79d9adf004
1 changed files with 0 additions and 2 deletions
--- a/pytorch_lightning/models/trainer.py
+++ b/pytorch_lightning/models/trainer.py
@ -315,13 +315,11 @@ class Trainer(TrainerIO):
        print(f"GPU: {gpu_nb} - Rank: {self.proc_rank}")

        # copy model to each gpu
-        print('starting DDP')
        torch.cuda.set_device(gpu_nb)
        model.cuda(gpu_nb)
        model = LightningDistributedDataParallel(model, device_ids=[gpu_nb])

        # continue training routine
-        print('running pretrain')
        self.__run_pretrain_routine(model)

    def __get_root_node_ip(self, proc_rank, nb_gpu_nodes, ip_file_dir):