Vectorize RepetitionPenaltyLogitsProcessor to improve performance (#8598)

* refactored exisiting nested loops to vectorized implementation

* replaced explicit indexing with torch.where

* modifying score for previous input_ids only

src/transformers/generation_logits_process.py

@@ -146,13 +146,13 @@ class RepetitionPenaltyLogitsProcessor(LogitsProcessor):
         self.penalty = penalty
 
     def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor) -> torch.FloatTensor:
-        for i in range(scores.shape[0]):
+        ranges = torch.arange(scores.shape[0])
-            for previous_token in set(input_ids[i].tolist()):
+        score = scores[ranges[:, None], input_ids]
-                # if score < 0 then repetition penalty has to be multiplied to reduce the previous token probability
+
-                if scores[i, previous_token] < 0:
+        # if score < 0 then repetition penalty has to be multiplied to reduce the previous token probability
-                    scores[i, previous_token] *= self.penalty
+        score = torch.where(score < 0, score * self.penalty, score / self.penalty)
-                else:
+
-                    scores[i, previous_token] /= self.penalty
+        scores[ranges[:, None], input_ids] = score
         return scores