diff --git a/neuralmagic/benchmarks/scripts/benchmark_throughput.py b/neuralmagic/benchmarks/scripts/benchmark_throughput.py
index 0607067e3817d..f49de1ec27a3f 100644
--- a/neuralmagic/benchmarks/scripts/benchmark_throughput.py
+++ b/neuralmagic/benchmarks/scripts/benchmark_throughput.py
@@ -10,10 +10,12 @@
 import time
 from datetime import datetime
 from pathlib import Path
-from typing import List, Optional, Tuple
+from typing import List, Optional, Tuple, cast
 
 from transformers import AutoTokenizer
 
+from vllm.inputs import PromptStrictInputs
+
 from .common import (generate_synthetic_requests, num_available_gpus,
                      print_request_outputs, warmup_vllm_engine)
 from .datasets_registry import DatasetArgs, get_dataset
@@ -77,8 +79,7 @@ def run_vllm(requests: List[Tuple[str, int, int]],
         )
         # FIXME(woosuk): Do not use internal method.
         llm._add_request(
-            prompt=prompt,
-            prompt_token_ids=None,
+            inputs=cast(PromptStrictInputs, prompt),
             params=sampling_params,
         )