| 1 | model_name: "nm-testing/Qwen3-30B-A3B-FP8-block" |
| 2 | accuracy_threshold: 0.86 |
| 3 | num_questions: 1319 |
| 4 | num_fewshot: 5 |
| 5 | server_args: >- |
| 6 | --enforce-eager |
| 7 | --max-model-len 8192 |
| 8 | --moe-backend humming |
| 9 | --linear-backend humming |
| 10 | --data-parallel-size 2 |
| 11 | --enable-expert-parallel |
| 12 | --all2all-backend deepep_high_throughput |
| 13 | env: |
| 14 | VLLM_HUMMING_MOE_GEMM_TYPE: "grouped_contiguous" |