diff --git a/server/text_generation_server/utils/logits_process.py b/server/text_generation_server/utils/logits_process.py
index 6d8cb71a..2decee53 100644
--- a/server/text_generation_server/utils/logits_process.py
+++ b/server/text_generation_server/utils/logits_process.py
@@ -143,6 +143,8 @@ class FrequencyPenaltyLogitsProcessor(LogitsProcessor):
         score = torch.gather(scores, 1, input_ids)
         # if score < 0 then penalty has to be multiplied to reduce the previous token probability
         score = -torch.where(score < 0, score * self.penalty, score / self.penalty)
+        # set score to 0 where input_ids is a padding token
+        score *= input_ids.ne(0)
 
         return scores.scatter_add_(1, input_ids, score)
 
@@ -168,6 +170,8 @@ class HeterogeneousFrequencyPenaltyLogitsProcessor(LogitsProcessor):
         score = -torch.where(
             score < 0, score * self.penalty_tensor, score / self.penalty_tensor
         )
+        # set score to 0 where input_ids is a padding token
+        score *= input_ids.ne(0)
 
         return scores.scatter_add_(1, input_ids, score)