| 1 | model_name: "RedHatAI/diffusiongemma-26B-A4B-it-FP8-dynamic" |
| 2 | accuracy_threshold: 0.84 |
| 3 | num_questions: 1319 |
| 4 | num_fewshot: 5 |
| 5 | startup_max_wait_seconds: 1200 |
| 6 | use_chat_completions: true |
| 7 | # Diffusion models use a fixed internal temperature schedule and do not |
| 8 | # support per-request temperature or seed overrides. |
| 9 | temperature: 1.0 |
| 10 | seed: null |
| 11 | server_args: >- |
| 12 | --enforce-eager |
| 13 | --max-model-len 4096 |
| 14 | --tensor-parallel-size 2 |
| 15 | --attention-backend TRITON_ATTN |