sampling_params:
  temperature: 1.0
  top_p: 0.9
  max_tokens: 200
  n: 16
  do_sample: true
  seed: 28
  skip_special_tokens: false
vllm_params:
  gpu_memory_utilization: 0.9
  minimize_vllm_memory: false