| 1 | model_name: "RedHatAI/Qwen3-30B-A3B-FP8-block" |
| 2 | accuracy_threshold: 0.85 |
| 3 | num_questions: 1319 |
| 4 | num_fewshot: 5 |
| 5 | # 30B MoE + DeepGEMM/DeepEP JIT compile can exceed the 1200s default startup wait. |
| 6 | startup_max_wait_seconds: 2400 |
| 7 | server_args: "--enforce-eager --max-model-len 8192 --data-parallel-size 2 --enable-expert-parallel --all2all-backend deepep_low_latency --disable-uvicorn-access-log" |
| 8 | env: |
| 9 | VLLM_USE_DEEP_GEMM: "1" |
| 10 | VLLM_USE_DEEP_GEMM_MOE: "1" |