Update ultravox_processing.py

ultravox_processing.py

@@ -112,7 +112,8 @@ class UltravoxProcessor(transformers.ProcessorMixin):
         assert (
             tokenizer.eos_token is not None
         ), "The tokenizer has no EOS token. Cannot recover."
-        self.audio_replacement_token_id = tokenizer.get_vocab()[tokenizer.eos_token]
+        self.vocab = tokenizer.get_vocab()
+        self.audio_replacement = tokenizer.eos_token
         if tokenizer.pad_token_id is None:
             tokenizer.pad_token_id = tokenizer.eos_token_id
 
@@ -326,6 +327,8 @@ class UltravoxProcessor(transformers.ProcessorMixin):
             split_input_ids = tokenized_parts["input_ids"]
             input_ids: List[int] = []
 
+            audio_replacement_token_id = self.vocab[self.audio_replacement]
+
             for i, token_len in enumerate(data.get("audio_token_len", [])):
                 if not audio_is_continuation[i]:
                     placeholder_index += 1
@@ -338,7 +341,7 @@ class UltravoxProcessor(transformers.ProcessorMixin):
 
                 audio_token_start_idx.append(len(input_ids))
 
-                input_ids.extend([self.audio_replacement_token_id] * token_len)
+                input_ids.extend([audio_replacement_token_id] * token_len)
 
             # Include any tokens after the last audio.
             placeholder_index += 1