[Bugfix] Fix profiling.py

Signed-off-by: zh Wang <[email protected]>

Author: hhy3

examples/offline_inference/profiling.py

@@ -235,7 +235,7 @@ def get_output_len_generator() -> Generator[int, Any, Any]:
             assert isinstance(sampling_params.max_tokens, int)
 
             prompt_token_ids = torch.randint(
-                llm.llm_engine.model_config.get_vocab_size(),
+                llm.get_tokenizer().vocab_size,
                 size=(prompt_len, )).tolist()
 
             llm.llm_engine.add_request(