Update model.py

model.py

@@ -998,7 +998,6 @@ class WhisperEncoder(WhisperPreTrainedModel):
         # CUSTOM
         # Create conv layers dynamically based on config
         self.conv_layers = nn.ModuleList()
-        self.conv_layers = nn.ModuleList()
         for layer_config in config.conv_preprocessing_layers:
             # Create sequential module for each conv+activation pair
             conv_sequence = nn.Sequential(
@@ -1024,6 +1023,14 @@ class WhisperEncoder(WhisperPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
+    # CUSTOM
+    def get_conv_stride(self):
+        """Calculate total stride of all conv layers"""
+        total_stride = 1
+        for layer in self.conv_layers:
+            total_stride *= layer.stride[0]
+        return total_stride
+
     def _freeze_parameters(self):
         for param in self.parameters():
             param.requires_grad = False
@@ -1101,7 +1108,8 @@ class WhisperEncoder(WhisperPreTrainedModel):
                 Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
         """
 
-        expected_seq_length = self.config.max_source_positions * self.conv1.stride[0] * self.conv2.stride[0]
+        # CUSTOM
+        expected_seq_length = self.config.max_source_positions * self.get_conv_stride()
         
         # CUSTOM
         # Must be deactivated for our purpose, theoretically Whisper supports any sequence length for the encoder