Fix hang in DDP HPC accelerators (#5157)

* Fix hang in DDP HPC accelerators

init_device was never called

* Update CHANGELOG.md

Author: ananthsub

CHANGELOG.md

@@ -54,6 +54,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 - Do not warn when the `name` key is used in the `lr_scheduler` dict ([#5057](https://github.com/PyTorchLightning/pytorch-lightning/pull/5057))
 
+- Fixed `DDPHPCAccelerator` hangs in DDP construction by calling `init_device` ([#5157](https://github.com/PyTorchLightning/pytorch-lightning/pull/5157))
 
 
 ## [1.1.0] - 2020-12-09

pytorch_lightning/accelerators/ddp_cpu_hpc_accelerator.py

@@ -43,3 +43,6 @@ def model_to_device(self, model, process_idx):
     def get_device_ids(self):
         device_ids = None
         return device_ids
+
+    def init_device(self, process_idx):
+        pass

pytorch_lightning/accelerators/ddp_hpc_accelerator.py

@@ -121,6 +121,7 @@ def ddp_train(self, process_idx, model):
         """
         # determine which process we are and world size
         self.set_world_ranks(process_idx)
+        self.init_device(process_idx)
 
         # toggle prog bar
         if (self.trainer.node_rank != 0 or process_idx != 0) and self.trainer.progress_bar_callback is not None: