| 1 | # NOTE: this model was produced from Qwen/Qwen3.5-35B-A3B-FP8 |
| 2 | model_name: "amd/Qwen3.5-35B-A3B-MXFP4" |
| 3 | accuracy_threshold: 0.89 |
| 4 | tolerance: 0.03 |
| 5 | num_questions: 1319 |
| 6 | num_fewshot: 5 |
| 7 | startup_max_wait_seconds: 1800 |
| 8 | server_args: >- |
| 9 | --max-model-len 4096 |
| 10 | --tensor-parallel-size 2 |
| 11 | --moe-backend emulation |
| 12 | --gpu-memory-utilization 0.35 |