| 1 | model_name: "Qwen/Qwen3.5-35B-A3B" |
| 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 | --tensor-parallel-size 1 |
| 10 | --quantization humming |
| 11 | --kernel-config.enable_flashinfer_autotune=False |
| 12 | env: |
| 13 | VLLM_HUMMING_ONLINE_QUANT_CONFIG: '{"dtype":"int6","hadamard_block_size":-1}' |