update

ultravox_model.py

@@ -201,8 +201,6 @@ class UltravoxModel(transformers.LlamaPreTrainedModel):
             for i, (audio, start, length) in enumerate(
                 zip(audio_embeds, audio_token_start_idx, audio_token_len)
             ):
-                length = min(length, audio.shape[0])
-                length = min(inputs_embeds.shape[1] - start, length)
                 inputs_embeds[i, start : start + length] = audio[:length]
 
         lm_output = self.language_model.forward(