tentative fix

OmniAvatar/models/wav2vec.py

@@ -65,7 +65,12 @@ class Wav2VecModel(Wav2Vec2Model):
         Returns:
             The output of the Wav2Vec model.
         """
-        self.config.output_attentions = True
+        # Set attention implementation to 'eager' when output_attentions is requested
+        if output_attentions or (output_attentions is None and getattr(self.config, 'output_attentions', False)):
+            self.config._attn_implementation = 'eager'
+            output_attentions = True
+        else:
+            output_attentions = False
 
         output_hidden_states = (
             output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
@@ -153,7 +158,12 @@ class Wav2VecModel(Wav2Vec2Model):
         Returns:
             The encoded output features.
         """
-        self.config.output_attentions = True
+        # Set attention implementation to 'eager' when output_attentions is requested
+        if output_attentions or (output_attentions is None and getattr(self.config, 'output_attentions', False)):
+            self.config._attn_implementation = 'eager'
+            output_attentions = True
+        else:
+            output_attentions = False
 
         output_hidden_states = (
             output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states