ref: unify slurm and TE under backendPlugin 2/n (#4580)

(cherry picked from commit bfaf014)
Lightning-AI · Nov 11, 2020 · 8835c26 · 8835c26
1 parent e8c00bc
commit 8835c26
Show file tree

Hide file tree

Showing 5 changed files with 7 additions and 221 deletions.
diff --git a/pytorch_lightning/accelerators/__init__.py b/pytorch_lightning/accelerators/__init__.py
@@ -20,8 +20,7 @@
 from pytorch_lightning.accelerators.gpu_accelerator import GPUAccelerator
 from pytorch_lightning.accelerators.tpu_accelerator import TPUAccelerator
 from pytorch_lightning.accelerators.horovod_accelerator import HorovodAccelerator
-from pytorch_lightning.accelerators.ddp_slurm_accelerator import DDPSLURMAccelerator
-from pytorch_lightning.accelerators.ddp_torchelastic_accelerator import DDPTorchElasticAccelerator
+from pytorch_lightning.accelerators.ddp_hpc_accelerator import DDPHPCAccelerator
 from pytorch_lightning.accelerators.ddp_cpu_torchelastic_accelerator import DDPCPUTorchElasticAccelerator
 from pytorch_lightning.accelerators.ddp_cpu_slurm_accelerator import DDPCPUSLURMAccelerator
 from pytorch_lightning.accelerators.accelerator import Accelerator
diff --git a/pytorch_lightning/accelerators/accelerator_connector.py b/pytorch_lightning/accelerators/accelerator_connector.py
@@ -227,7 +227,7 @@ def select_accelerator(self):
             )
 
         elif use_slurm_ddp:
-            accelerator_backend = accelerators.DDPSLURMAccelerator(
+            accelerator_backend = accelerators.DDPHPCAccelerator(
                 self.trainer,
                 cluster_env,
                 self.trainer.plugin_connector.ddp_plugin
@@ -241,7 +241,7 @@ def select_accelerator(self):
             )
 
         elif use_torchelastic_ddp:
-            accelerator_backend = accelerators.DDPTorchElasticAccelerator(
+            accelerator_backend = accelerators.DDPHPCAccelerator(
                 self.trainer,
                 cluster_env,
                 self.trainer.plugin_connector.ddp_plugin

diff --git a/...ing/accelerators/ddp_slurm_accelerator.py → ...tning/accelerators/ddp_hpc_accelerator.py b/...ing/accelerators/ddp_slurm_accelerator.py → ...tning/accelerators/ddp_hpc_accelerator.py
@@ -26,6 +26,7 @@
 from pytorch_lightning.utilities import AMPType
 from pytorch_lightning.utilities.distributed import rank_zero_only, sync_ddp_if_available
 
+
 try:
     from hydra.utils import to_absolute_path, get_original_cwd
     from hydra.core.hydra_config import HydraConfig
@@ -35,12 +36,7 @@
     HYDRA_AVAILABLE = True
 
 
-# -------------------------------------------
-# !!!!!!!!!!!!!! NOTE !!!!!!!!!!!!!!!!!!!!!!
-# TEMP CLASS WHILE WE DECOUPLE SLURM FROM DDP
-# !!!!!!!!!!!!!! NOTE !!!!!!!!!!!!!!!!!!!!!!
-# -------------------------------------------
-class DDPSLURMAccelerator(Accelerator):
+class DDPHPCAccelerator(Accelerator):
 
     def __init__(self, trainer, cluster_environment=None, ddp_plugin=None):
         super().__init__(trainer, cluster_environment, ddp_plugin)

diff --git a/pytorch_lightning/accelerators/ddp_torchelastic_accelerator.py b/pytorch_lightning/accelerators/ddp_torchelastic_accelerator.py
diff --git a/tests/backends/test_accelerator_connector.py b/tests/backends/test_accelerator_connector.py
@@ -111,7 +111,7 @@ def test_accelerator_choice_ddp_slurm(tmpdir):
     class CB(Callback):
         def on_fit_start(self, trainer, pl_module):
             assert trainer.use_ddp
-            assert isinstance(trainer.accelerator_backend, accelerators.DDPSLURMAccelerator)
+            assert isinstance(trainer.accelerator_backend, accelerators.DDPHPCAccelerator)
             assert isinstance(trainer.accelerator_backend.cluster_environment, SLURMEnvironment)
             assert trainer.accelerator_backend.task_idx == 10
             assert trainer.accelerator_backend.cluster_environment.local_rank() == trainer.accelerator_backend.task_idx
@@ -172,7 +172,7 @@ def test_accelerator_choice_ddp_te(tmpdir):
     class CB(Callback):
         def on_fit_start(self, trainer, pl_module):
             assert trainer.use_ddp
-            assert isinstance(trainer.accelerator_backend, accelerators.DDPTorchElasticAccelerator)
+            assert isinstance(trainer.accelerator_backend, accelerators.DDPHPCAccelerator)
             assert isinstance(trainer.accelerator_backend.cluster_environment, TorchElasticEnvironment)
             assert trainer.accelerator_backend.task_idx == 10
             assert trainer.accelerator_backend.cluster_environment.local_rank() == trainer.accelerator_backend.task_idx