Upload JointCTCAttentionEncoderDecoder

modeling_reguler.py

@@ -466,6 +466,9 @@ class JointCTCAttentionEncoderDecoder(SpeechEncoderDecoderModel):
     ) -> Union[GenerateOutput, torch.LongTensor]:
         if "encoder_outputs" in kwargs:
             self.encoder_logits = kwargs["encoder_outputs"].logits
+            self.encoder_output_lens = self.encoder._get_feat_extract_output_lengths(
+                kwargs["attention_mask"].sum(dim=1)
+            )
         output = super().generate(
             inputs,
             generation_config,