| 1 | model_name: "QuixiAI/Qwen3-30B-A3B-AWQ" |
| 2 | accuracy_threshold: 0.90 |
| 3 | num_questions: 1319 |
| 4 | num_fewshot: 5 |
| 5 | gen_prefix: " <think>\n\n</think>\n" |
| 6 | server_args: >- |
| 7 | --enforce-eager |
| 8 | --max-model-len 8192 |
| 9 | --moe-backend humming |
| 10 | --linear-backend humming |
| 11 | --dtype bfloat16 |
| 12 | env: |
| 13 | VLLM_HUMMING_INPUT_QUANT_CONFIG: '{"dtype":"int8"}' |