diff --git a/deepspeed/runtime/fp16/fused_optimizer.py b/deepspeed/runtime/fp16/fused_optimizer.py
index 8c1d2003cb1b..5f35c1884a41 100755
--- a/deepspeed/runtime/fp16/fused_optimizer.py
+++ b/deepspeed/runtime/fp16/fused_optimizer.py
@@ -153,10 +153,10 @@ def step_fused_adam(self, closure=None):
 
         if self.overflow:
             if self.verbose:
-                logger.info("[deepspeed] OVERFLOW! Skipping step. Attempted loss "
-                            "scale: {}, reducing to {}".format(
-                                prev_scale,
-                                self.cur_scale))
+                logger.info(
+                    "[deepspeed] fp16 dynamic loss scale overflow! Skipping step. Attempted loss "
+                    "scale: {}, reducing to {}".format(prev_scale,
+                                                       self.cur_scale))
             return self.overflow
         combined_scale = self.unscale_and_clip_grads(grads_groups_flat,
                                                      norm_groups,
diff --git a/deepspeed/runtime/fp16/loss_scaler.py b/deepspeed/runtime/fp16/loss_scaler.py
index ad6f8f6227f9..954d0ea61585 100755
--- a/deepspeed/runtime/fp16/loss_scaler.py
+++ b/deepspeed/runtime/fp16/loss_scaler.py
@@ -213,7 +213,7 @@ def update_scale(self, overflow):
             optimizer.step()
         # Otherwise, don't do anything -- ie, skip iteration
         else:
-            print('OVERFLOW!')
+            print('fp16 dynamic loss scale overflow!')
 
         # Update loss scale for next iteration
         loss_scaler.update_scale(has_overflow)
diff --git a/deepspeed/runtime/fp16/unfused_optimizer.py b/deepspeed/runtime/fp16/unfused_optimizer.py
index 37edf9d5002d..c0cef6a56ba7 100755
--- a/deepspeed/runtime/fp16/unfused_optimizer.py
+++ b/deepspeed/runtime/fp16/unfused_optimizer.py
@@ -139,10 +139,10 @@ def step_fused_lamb(self, closure=None):
         self._update_scale(self.overflow)
         if self.overflow:
             if self.verbose:
-                logger.info("[deepspeed] OVERFLOW! Skipping step. Attempted loss "
-                            "scale: {}, reducing to {}".format(
-                                prev_scale,
-                                self.cur_scale))
+                logger.info(
+                    "[deepspeed] fp16 dynamic loss scale overflow! Skipping step. Attempted loss "
+                    "scale: {}, reducing to {}".format(prev_scale,
+                                                       self.cur_scale))
             return self.overflow
 
         combined_scale = self.unscale_and_clip_grads(norm_groups, apply_scale=False)
@@ -165,10 +165,10 @@ def step(self, closure=None):
         self._update_scale(self.overflow)
         if self.overflow:
             if self.verbose:
-                logger.info("[deepspeed] OVERFLOW! Skipping step. Attempted loss "
-                            "scale: {}, reducing to {}".format(
-                                prev_scale,
-                                self.cur_scale))
+                logger.info(
+                    "[deepspeed] fp16 dynamic loss scale overflow! Skipping step. Attempted loss "
+                    "scale: {}, reducing to {}".format(prev_scale,
+                                                       self.cur_scale))
             return self.overflow
 
         norm_groups = []
diff --git a/deepspeed/runtime/zero/stage1.py b/deepspeed/runtime/zero/stage1.py
index d5c7616ff87e..7cd37f904faa 100755
--- a/deepspeed/runtime/zero/stage1.py
+++ b/deepspeed/runtime/zero/stage1.py
@@ -630,10 +630,10 @@ def step(self, closure=None):
         if self.overflow:
             self.zero_grad()
             if self.verbose:
-                logger.info("[deepspeed] OVERFLOW! Skipping step. Attempted loss "
-                            "scale: {}, reducing to {}".format(
-                                prev_scale,
-                                self.loss_scale))
+                logger.info(
+                    "[deepspeed] fp16 dynamic loss scale overflow! Skipping step. Attempted loss "
+                    "scale: {}, reducing to {}".format(prev_scale,
+                                                       self.loss_scale))
             return self.overflow
 
         norm_groups = []
diff --git a/deepspeed/runtime/zero/stage2.py b/deepspeed/runtime/zero/stage2.py
index 4cf39cb5ccd6..6f3fb1cd6509 100755
--- a/deepspeed/runtime/zero/stage2.py
+++ b/deepspeed/runtime/zero/stage2.py
@@ -1355,7 +1355,7 @@ def step(self, closure=None):
             see_memory_usage('After overflow after clearing gradients')
 
             logger.info(
-                "[deepscale] OVERFLOW! Rank {} Skipping step. Attempted loss scale: {}, "
+                "[deepspeed] fp16 dynamic loss scale overflow! Rank {} Skipping step. Attempted loss scale: {}, "
                 "reducing to {}".format(dist.get_rank(),
                                         prev_scale,
                                         self.loss_scale))