Added check for apex AMP and unit tests for Horovod + AMP

CHANGELOG.md

@@ -39,6 +39,8 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 - Fixed setting batch size in `LightningModule.datamodule` when using `auto_scale_batch_size` ([#3266](https://github.com/PyTorchLightning/pytorch-lightning/pull/3266))
 
+- Fixed Horovod distributed backend compatibility with native AMP ([#3404](https://github.com/PyTorchLightning/pytorch-lightning/pull/3404))
+
 ## [0.9.0] - YYYY-MM-DD
 
 ### Added

pytorch_lightning/accelerators/horovod_backend.py

@@ -72,11 +72,6 @@ def setup(self, model):
             if isinstance(scheduler, _LRScheduler):
                 scheduler.base_lrs = [lr * hvd.size() for lr in scheduler.base_lrs]
 
-        if self.trainer.amp_backend:
-            model, optimizers = model.configure_apex(amp, model, self.trainer.optimizers, self.trainer.amp_level)
-            self.trainer.optimizers = optimizers
-            self.trainer.reinit_scheduler_properties(self.trainer.optimizers, self.trainer.lr_schedulers)
-
         # Horovod: broadcast parameters & optimizer state to ensure consistent initialization
         hvd.broadcast_parameters(model.state_dict(), root_rank=0)
         for optimizer in self.trainer.optimizers:
@@ -92,6 +87,11 @@ def filter_named_parameters(model, optimizer):
             for optimizer in self.trainer.optimizers
         ]
 
+        if self.trainer.amp_backend == AMPType.APEX:
+            model, optimizers = model.configure_apex(amp, model, self.trainer.optimizers, self.trainer.amp_level)
+            self.trainer.optimizers = optimizers
+            self.trainer.reinit_scheduler_properties(self.trainer.optimizers, self.trainer.lr_schedulers)
+
         # Update logger rank info from Horovod to avoid race conditions from  different ranks
         # creating directories / writing files in the same locations.
         self.trainer.global_rank = hvd.rank()

tests/models/test_horovod.py

@@ -42,8 +42,8 @@ def _nccl_available():
 def _run_horovod(trainer_options, on_gpu=False):
     """Execute the training script across multiple workers in parallel."""
     num_processes = trainer_options.get('gpus', 2)
-    # gpus trainer argument does not apply for horovod
-    trainer_options.update(gpus=None)
+    # for Horovod, we interpret `gpus` to be set per worker
+    trainer_options.update(gpus=1 if on_gpu else None)
     tutils.reset_seed()
     cmdline = [
         'horovodrun',
@@ -110,6 +110,27 @@ def test_horovod_multi_gpu(tmpdir):
     _run_horovod(trainer_options, on_gpu=True)
 
 
+@pytest.mark.skipif(platform.system() == "Windows", reason="Horovod is not supported on Windows")
+@pytest.mark.skipif(not _nccl_available(), reason="test requires Horovod with NCCL support")
+@pytest.mark.skipif(torch.cuda.device_count() < 2, reason="test requires multi-GPU machine")
+def test_horovod_amp(tmpdir):
+    """Test Horovod with multi-GPU support."""
+    trainer_options = dict(
+        default_root_dir=str(tmpdir),
+        weights_save_path=str(tmpdir),
+        gradient_clip_val=1.0,
+        progress_bar_refresh_rate=0,
+        max_epochs=1,
+        limit_train_batches=0.4,
+        limit_val_batches=0.2,
+        gpus=2,
+        deterministic=True,
+        distributed_backend='horovod',
+        precision=16,
+    )
+    _run_horovod(trainer_options, on_gpu=True)
+
+
 @pytest.mark.skipif(platform.system() == "Windows", reason="Horovod is not supported on Windows")
 @pytest.mark.skipif(not _nccl_available(), reason="test requires Horovod with NCCL support")
 @pytest.mark.skipif(not torch.cuda.is_available(), reason="test requires GPU machine")