[Bugfix] Fix xgrammar failing to read a vocab_size from LlavaConfig o…

…n PixtralHF. (vllm-project#11043)
triple-Mu · Dec 10, 2024 · e35879c · e35879c
1 parent ebf7780
commit e35879c
Showing 1 changed file with 3 additions and 3 deletions.
diff --git a/vllm/model_executor/guided_decoding/xgrammar_decoding.py b/vllm/model_executor/guided_decoding/xgrammar_decoding.py
@@ -148,7 +148,7 @@ def from_guided_params(cls,
             else:
                 json_str = guided_params.json
             return cls(json_str=json_str,
-                       vocab_size=model_config.hf_config.vocab_size,
+                       vocab_size=model_config.hf_text_config.vocab_size,
                        encoded_vocab=encoded_vocab,
                        stop_token_ids=stop_token_ids,
                        backend_str=backend_str,
@@ -168,15 +168,15 @@ def from_guided_params(cls,
             else:
                 grammar_str = guided_params.grammar
             return cls(grammar_str=grammar_str,
-                       vocab_size=model_config.hf_config.vocab_size,
+                       vocab_size=model_config.hf_text_config.vocab_size,
                        encoded_vocab=encoded_vocab,
                        stop_token_ids=stop_token_ids,
                        backend_str=backend_str,
                        tokenizer_hash=tokenizer_hash,
                        max_threads=max_threads)
         elif guided_params.json_object:
             return cls(json_object=True,
-                       vocab_size=model_config.hf_config.vocab_size,
+                       vocab_size=model_config.hf_text_config.vocab_size,
                        encoded_vocab=encoded_vocab,
                        stop_token_ids=stop_token_ids,
                        backend_str=backend_str,