funasr/models/llm_asr/model.py
@@ -496,8 +496,11 @@ batch_size, frames, _ = speech.shape with torch.cuda.amp.autocast(enabled=False): # audio encoder encoder_out, encoder_out_lens = self.audio_encoder(speech.permute(0, 2, 1), speech_lengths) encoder_out, encoder_out_lens = self.audio_encoder( speech.permute(0, 2, 1), speech_lengths ) # audio_adaptor encoder_out, encoder_out_lens = self.audio_adaptor(encoder_out, encoder_out_lens)
