diff --git a/pytorch_lightning/trainer/evaluation_loop_mixin.py b/pytorch_lightning/trainer/evaluation_loop_mixin.py
index c2ba5d16fa97a..8583805ab772d 100644
--- a/pytorch_lightning/trainer/evaluation_loop_mixin.py
+++ b/pytorch_lightning/trainer/evaluation_loop_mixin.py
@@ -138,7 +138,7 @@ def run_evaluation(self, test=False):
             self.log_metrics(log_metrics, {})
 
             # track metrics for callbacks
-            self.callback_metrics = callback_metrics
+            self.callback_metrics.update(callback_metrics)
 
             # hook
             model.on_post_performance_check()
diff --git a/pytorch_lightning/trainer/train_loop_mixin.py b/pytorch_lightning/trainer/train_loop_mixin.py
index 306416db8d7ed..68c34cf4f3793 100644
--- a/pytorch_lightning/trainer/train_loop_mixin.py
+++ b/pytorch_lightning/trainer/train_loop_mixin.py
@@ -258,7 +258,7 @@ def optimizer_closure():
         all_log_metrics = {k: v for d in all_log_metrics for k, v in d.items()}
 
         # track all metrics for callbacks
-        self.callback_metrics = {k: v for d in all_callback_metrics for k, v in d.items()}
+        self.callback_metrics.update({k: v for d in all_callback_metrics for k, v in d.items()})
 
         return 0, grad_norm_dic, all_log_metrics