funasr/models/llm_asr_nar/model.py
@@ -214,7 +214,7 @@ def encode( self, speech: torch.Tensor, speech_lengths: torch.Tensor, **kwargs, ) -> Tuple[torch.Tensor, torch.Tensor]: ): audio_mask = kwargs.get("audio_mask", None) audio_token_lengths = audio_mask.sum(-1) if audio_mask is not None else None