made ddp the default if no backend specified with multiple GPUs (#1789)

* made ddp the default if no backend specified with multiple GPUs

* fix

* spawn

Co-authored-by: Jirka <jirka.borovec@seznam.cz>

docs/source/multi_gpu.rst

@@ -132,6 +132,8 @@ Lightning allows multiple ways of training
 - Horovod (`distributed_backend='horovod'`) (multi-machine, multi-gpu, configured at runtime)
 - TPUs (`num_tpu_cores=8|x`) (tpu or TPU pod)
 
+.. note:: If you request multiple GPUs without setting a mode, ddp will be automatically used.
+
 Data Parallel (dp)
 ^^^^^^^^^^^^^^^^^^
 `DataParallel <https://pytorch.org/docs/stable/nn.html#torch.nn.DataParallel>`_ splits a batch across k GPUs. That is, if you have a batch of 32 and use dp with 2 gpus,

pytorch_lightning/trainer/distrib_data_parallel.py

@@ -203,8 +203,8 @@ def set_distributed_mode(self, distributed_backend):
             elif self.num_gpus > 1:
                 rank_zero_warn('You requested multiple GPUs but did not specify a backend, e.g.'
                                ' Trainer(distributed_backend=dp) (or ddp, ddp2).'
-                               ' Setting distributed_backend=dp for you.')
-                self.use_dp = True
+                               ' Setting distributed_backend=ddp for you.')
+                self.use_ddp = True
         elif distributed_backend == "dp":
             # do nothing if num_gpus == 0
             if self.num_gpus == 1:

tests/models/test_gpu.py

@@ -130,6 +130,7 @@ def assert_pred_same():
     trainer.fit(model)
 
 
+@pytest.mark.spawn
 @pytest.mark.skipif(torch.cuda.device_count() < 2, reason="test requires multi-GPU machine")
 def test_multi_gpu_none_backend(tmpdir):
     """Make sure when using multiple GPUs the user can't use `distributed_backend = None`."""

tests/trainer/test_trainer.py

@@ -712,7 +712,7 @@ def test_gpu_choice(tmpdir):
     ),
     pytest.param(
         dict(distributed_backend=None, gpus=2),
-        dict(use_dp=True, use_ddp=False, use_ddp2=False, num_gpus=2, on_gpu=True, single_gpu=False, num_processes=1),
+        dict(use_dp=False, use_ddp=True, use_ddp2=False, num_gpus=2, on_gpu=True, single_gpu=False, num_processes=1),
         marks=[pytest.mark.skipif(torch.cuda.device_count() < 2, reason="Multiple GPUs needed")]
     ),
     pytest.param(