diff --git a/pytorch_lightning/accelerators/ddp2_backend.py b/pytorch_lightning/accelerators/ddp2_backend.py
index 85bda4cd8deef..aa7d16a10cda9 100644
--- a/pytorch_lightning/accelerators/ddp2_backend.py
+++ b/pytorch_lightning/accelerators/ddp2_backend.py
@@ -106,13 +106,6 @@ def ddp_train(self, process_idx, mp_queue, model, is_master=False, proc_offset=0
             log.info(f'All DDP processes registered. Starting ddp with {self.trainer.world_size} processes')
             log.info('-' * 100)
 
-        # CHOOSE OPTIMIZER
-        # allow for lr schedulers as well
-        optimizers, lr_schedulers, optimizer_frequencies = self.trainer.init_optimizers(model)
-        self.trainer.optimizers = optimizers
-        self.trainer.lr_schedulers = lr_schedulers
-        self.trainer.optimizer_frequencies = optimizer_frequencies
-
         # MODEL
         # copy model to each gpu
         if self.trainer.on_gpu:
@@ -130,6 +123,13 @@ def ddp_train(self, process_idx, mp_queue, model, is_master=False, proc_offset=0
             torch.cuda.set_device(self.trainer.root_gpu)
             model.cuda(self.trainer.root_gpu)
 
+        # CHOOSE OPTIMIZER
+        # allow for lr schedulers as well
+        optimizers, lr_schedulers, optimizer_frequencies = self.trainer.init_optimizers(model)
+        self.trainer.optimizers = optimizers
+        self.trainer.lr_schedulers = lr_schedulers
+        self.trainer.optimizer_frequencies = optimizer_frequencies
+
         # set model properties before going into wrapper
         self.trainer.copy_trainer_model_properties(model)
 
diff --git a/pytorch_lightning/accelerators/ddp_backend.py b/pytorch_lightning/accelerators/ddp_backend.py
index e499feda651d9..241e480e1ba7b 100644
--- a/pytorch_lightning/accelerators/ddp_backend.py
+++ b/pytorch_lightning/accelerators/ddp_backend.py
@@ -169,13 +169,6 @@ def ddp_train(self, process_idx, mp_queue, model, is_master=False, proc_offset=0
             log.info(f'All DDP processes registered. Starting ddp with {self.trainer.world_size} processes')
             log.info('-' * 100)
 
-        # CHOOSE OPTIMIZER
-        # allow for lr schedulers as well
-        optimizers, lr_schedulers, optimizer_frequencies = self.trainer.init_optimizers(model)
-        self.trainer.optimizers = optimizers
-        self.trainer.lr_schedulers = lr_schedulers
-        self.trainer.optimizer_frequencies = optimizer_frequencies
-
         # call sync_bn before .cuda(), configure_apex and configure_ddp
         if self.trainer.sync_batchnorm:
             model = model.configure_sync_batchnorm(model)
@@ -197,6 +190,13 @@ def ddp_train(self, process_idx, mp_queue, model, is_master=False, proc_offset=0
             torch.cuda.set_device(self.trainer.root_gpu)
             model.cuda(self.trainer.root_gpu)
 
+        # CHOOSE OPTIMIZER
+        # allow for lr schedulers as well
+        optimizers, lr_schedulers, optimizer_frequencies = self.trainer.init_optimizers(model)
+        self.trainer.optimizers = optimizers
+        self.trainer.lr_schedulers = lr_schedulers
+        self.trainer.optimizer_frequencies = optimizer_frequencies
+
         # set model properties before going into wrapper
         self.trainer.copy_trainer_model_properties(model)
 
diff --git a/pytorch_lightning/accelerators/ddp_spawn_backend.py b/pytorch_lightning/accelerators/ddp_spawn_backend.py
index 9ed68f66083ad..abcc235e367ef 100644
--- a/pytorch_lightning/accelerators/ddp_spawn_backend.py
+++ b/pytorch_lightning/accelerators/ddp_spawn_backend.py
@@ -110,13 +110,6 @@ def ddp_train(self, process_idx, mp_queue, model):
             log.info(f'All DDP processes registered. Starting ddp with {self.trainer.world_size} processes')
             log.info('-' * 100)
 
-        # CHOOSE OPTIMIZER
-        # allow for lr schedulers as well
-        optimizers, lr_schedulers, optimizer_frequencies = self.trainer.init_optimizers(model)
-        self.trainer.optimizers = optimizers
-        self.trainer.lr_schedulers = lr_schedulers
-        self.trainer.optimizer_frequencies = optimizer_frequencies
-
         # call sync_bn before .cuda(), configure_apex and configure_ddp
         if self.trainer.sync_batchnorm:
             model = model.configure_sync_batchnorm(model)
@@ -129,6 +122,13 @@ def ddp_train(self, process_idx, mp_queue, model):
             torch.cuda.set_device(self.trainer.root_gpu)
             model.cuda(self.trainer.root_gpu)
 
+        # CHOOSE OPTIMIZER
+        # allow for lr schedulers as well
+        optimizers, lr_schedulers, optimizer_frequencies = self.trainer.init_optimizers(model)
+        self.trainer.optimizers = optimizers
+        self.trainer.lr_schedulers = lr_schedulers
+        self.trainer.optimizer_frequencies = optimizer_frequencies
+
         # set model properties before going into wrapper
         self.trainer.copy_trainer_model_properties(model)