model_name: "Qwen/Qwen3-0.6B-FP8"
accuracy_threshold: 0.375
num_questions: 1319
num_fewshot: 5
server_args: "--enforce-eager --max-model-len 4096"