| 1 | model_name: "nvidia/GLM-5.2-NVFP4" |
| 2 | accuracy_threshold: 0.90 |
| 3 | tolerance: 0.0 |
| 4 | num_questions: 1319 |
| 5 | num_fewshot: 5 |
| 6 | max_tokens: 256 |
| 7 | max_concurrency: 32 |
| 8 | request_timeout_seconds: 1800 |
| 9 | startup_max_wait_seconds: 1200 |
| 10 | server_args: >- |
| 11 | --tensor-parallel-size 4 |
| 12 | --max-model-len 8192 |
| 13 | --max-num-seqs 32 |
| 14 | --max-num-batched-tokens 4096 |
| 15 | --enforce-eager |
| 16 | --kv-cache-dtype bfloat16 |
| 17 | --attention-config '{"hisparse_config":{}}' |
| 18 | --kv-transfer-config '{"kv_connector":"HiSparseConnector","kv_role":"kv_both","kv_connector_extra_config":{"host_pool_gib":32}}' |
| 19 | env: |
| 20 | VLLM_USE_V2_MODEL_RUNNER: "1" |