| 1 | model_name: "deepseek-ai/DeepSeek-R1" |
| 2 | accuracy_threshold: 0.95 |
| 3 | num_questions: 1319 |
| 4 | num_fewshot: 5 |
| 5 | startup_max_wait_seconds: 1800 |
| 6 | env: |
| 7 | VLLM_ROCM_USE_AITER: "1" |
| 8 | VLLM_ROCM_USE_AITER_MLA: "1" |
| 9 | VLLM_ROCM_USE_AITER_MOE: "1" |
| 10 | server_args: >- |
| 11 | --enforce-eager |
| 12 | --max-model-len 4096 |
| 13 | --tensor-parallel-size 4 |
| 14 | --enable-expert-parallel |
| 15 | --attention-backend ROCM_AITER_MLA |
| 16 | --moe-backend aiter |
| 17 | --speculative-config '{"method":"mtp","num_speculative_tokens":3}' |