model_name: "nvidia/Qwen3-30B-A3B-FP4"
accuracy_threshold: 0.89
num_questions: 1319
num_fewshot: 5
server_args: "--enforce-eager --max-model-len 4096"