diff --git a/src/transformers/generation/configuration_utils.py b/src/transformers/generation/configuration_utils.py
index f8702dde29..9c268f6693 100644
--- a/src/transformers/generation/configuration_utils.py
+++ b/src/transformers/generation/configuration_utils.py
@@ -131,6 +131,11 @@ class GenerationConfig(PushToHubMixin):
             the current pass after allocated time has been passed.
         stop_strings (`str or list[str]`, *optional*):
             A string or a list of strings that should terminate generation if the model outputs them.
+        entropy_threshold (`float`, *optional*):
+            Threshold for the Shannon entropy of the prediction scores. If the entropy exceeds this threshold,
+            generation will stop.
+        entropy_patience (`int`, *optional*, defaults to 1):
+            Number of consecutive steps for which the entropy must exceed the threshold before stopping.
 
         > Parameters that control the generation strategy used
 
@@ -359,6 +364,10 @@ class GenerationConfig(PushToHubMixin):
         self.max_time = kwargs.pop("max_time", None)
         self.stop_strings = kwargs.pop("stop_strings", None)
 
+        # Parameters that control the entropy-based stopping criteria
+        self.entropy_threshold = kwargs.pop("entropy_threshold", None)
+        self.entropy_patience = kwargs.pop("entropy_patience", 1)
+
         # Parameters that control the generation strategy used
         self.do_sample = kwargs.pop("do_sample", None)
         self.num_beams = kwargs.pop("num_beams", None)
diff --git a/src/transformers/generation/stopping_criteria.py b/src/transformers/generation/stopping_criteria.py
index b57136f534..79203c4c8f 100644
--- a/src/transformers/generation/stopping_criteria.py
+++ b/src/transformers/generation/stopping_criteria.py
@@ -492,6 +492,52 @@ class ConfidenceCriteria(StoppingCriteria):
         return False
 
 
+
+class EntropyStoppingCriteria(StoppingCriteria):
+    """
+    This class can be used to stop generation whenever the entropy of the prediction scores exceeds a certain
+    threshold.
+
+    Args:
+        entropy_threshold (`float`):
+            The threshold for the Shannon entropy of the prediction scores.
+        entropy_patience (`int`, *optional*, defaults to 1):
+            The number of consecutive steps for which the entropy must exceed the threshold before stopping.
+    """
+
+    def __init__(self, entropy_threshold: float, entropy_patience: int = 1):
+        self.entropy_threshold = entropy_threshold
+        self.entropy_patience = entropy_patience
+        self.patience_counter = None
+
+    @add_start_docstrings(STOPPING_CRITERIA_INPUTS_DOCSTRING)
+    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> torch.BoolTensor:
+        if scores is None or (isinstance(scores, (list, tuple)) and len(scores) == 0):
+            return torch.zeros(input_ids.shape[0], device=input_ids.device, dtype=torch.bool)
+
+        # Get the latest scores
+        if isinstance(scores, (list, tuple)):
+            latest_scores = scores[-1]
+        else:
+            latest_scores = scores
+
+        # Compute Shannon entropy
+        probs = F.softmax(latest_scores, dim=-1)
+        # Use a small epsilon to avoid log(0)
+        entropy = -torch.sum(probs * torch.log(probs + 1e-12), dim=-1)
+
+        if self.patience_counter is None or self.patience_counter.shape[0] != input_ids.shape[0]:
+            self.patience_counter = torch.zeros(input_ids.shape[0], device=input_ids.device, dtype=torch.long)
+
+        # Move patience_counter to the same device as entropy
+        self.patience_counter = self.patience_counter.to(entropy.device)
+
+        high_entropy = entropy > self.entropy_threshold
+        self.patience_counter = (self.patience_counter + 1) * high_entropy
+
+        return self.patience_counter >= self.entropy_patience
+
+
 class StoppingCriteriaList(list):
     @add_start_docstrings(STOPPING_CRITERIA_INPUTS_DOCSTRING)
     def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> torch.BoolTensor:
diff --git a/src/transformers/generation/utils.py b/src/transformers/generation/utils.py
index ffb7266a5b..def46827ff 100644
--- a/src/transformers/generation/utils.py
+++ b/src/transformers/generation/utils.py
@@ -100,6 +100,7 @@ from .logits_process import (
 )
 from .stopping_criteria import (
     ConfidenceCriteria,
+    EntropyStoppingCriteria,
     EosTokenCriteria,
     MaxLengthCriteria,
     MaxTimeCriteria,
@@ -1292,6 +1293,13 @@ class GenerationMixin(ContinuousMixin):
             criteria.append(
                 ConfidenceCriteria(assistant_confidence_threshold=generation_config.assistant_confidence_threshold)
             )
+        if generation_config.entropy_threshold is not None:
+            criteria.append(
+                EntropyStoppingCriteria(
+                    entropy_threshold=generation_config.entropy_threshold,
+                    entropy_patience=generation_config.entropy_patience,
+                )
+            )
         criteria = self._merge_criteria_processor_list(criteria, stopping_criteria)
         return criteria