Update ultravox_model.py

ultravox_model.py

@@ -201,7 +201,8 @@ class UltravoxModel(transformers.LlamaPreTrainedModel):
             for i, (audio, start, length) in enumerate(
                 zip(audio_embeds, audio_token_start_idx, audio_token_len)
             ):
-                inputs_embeds[i, start : start + length] = audio[:length]
+                assert length <= audio.shape[0]
+                inputs_embeds[i, start : start + length].copy_(audio[:length])
                 
 
         lm_output = self.language_model.forward(