Early Return Pattern "if return else return" -> "if return return" (#197)

deepspeed/pt/deepspeed_lr_schedules.py

@@ -271,9 +271,8 @@ def get_lr_from_config(config):
 
     if lr_schedule == LR_RANGE_TEST:
         return lr_params[LR_RANGE_TEST_MIN_LR], ''
-    elif lr_schedule == ONE_CYCLE:
+    if lr_schedule == ONE_CYCLE:
         return lr_params[CYCLE_MAX_LR], ''
-    else:
     # Warmup LR
     return lr_params[WARMUP_MAX_LR], ''
 
@@ -624,7 +623,6 @@ class OneCycle(object):
         """
         if self.last_batch_iteration <= self.total_size:
             return self._get_cycle_lr()
-        else:
         return self._get_decay_lr(self.last_batch_iteration - self.total_size)
 
     def step(self, batch_iteration=None):
@@ -701,7 +699,6 @@ class WarmupLR(object):
     def _get_gamma(self):
         if self.last_batch_iteration < self.warmup_num_steps:
             return self.inverse_log_warm_up * math.log(self.last_batch_iteration + 1)
-        else:
         return 1.0
 
     def _format_param(self, optimizer, param_value, param_name):
@@ -712,5 +709,4 @@ class WarmupLR(object):
                     param_name,
                     FileNotFoundError(param_value)))
             return list(param_value)
-        else:
         return [param_value] * len(optimizer.param_groups)

deepspeed/pt/loss_scaler.py

@@ -23,7 +23,6 @@ import torch
 def to_python_float(t):
     if hasattr(t, 'item'):
         return t.item()
-    else:
     return t[0]