Update ultravox_model.py

ultravox_model.py

@@ -201,8 +201,9 @@ class UltravoxModel(transformers.LlamaPreTrainedModel):
             for i, (audio, start, length) in enumerate(
                 zip(audio_embeds, audio_token_start_idx, audio_token_len)
             ):
+                assert audio.shape[0] == length, f"{audio.shape=}, {start=}, {length=}, {inputs_embeds[i].shape=}"
                 inputs_embeds[i, start : start + length] = audio
-
+                
 
         lm_output = self.language_model.forward(
             inputs_embeds=inputs_embeds,