hparams.py: adding use saved learning rate param

train.py

@@ -190,8 +190,11 @@ def train(output_directory, log_directory, checkpoint_path, warm_start, n_gpus,
         if warm_start:
             model = warm_start_model(checkpoint_path, model)
         else:
-            model, optimizer, learning_rate, iteration = load_checkpoint(
+            model, optimizer, _learning_rate, iteration = load_checkpoint(
                 checkpoint_path, model, optimizer)
+            if hparams.use_saved_learning_rate:
+                learning_rate = _learning_rate
+
             iteration += 1  # next iteration is iteration + 1
             epoch_offset = max(0, int(iteration / len(train_loader)))