Run VLLM in Spark

Any improvements in performance for gpt-oss-120b?