# ==========================================
# ACTUAL TEST CASES
# ==========================================

test_cases:
  - name: "Qwen3-30B-A3B-W8A8-TP1"
    model: "vllm-ascend/Qwen3-30B-A3B-W8A8"
    envs:
      OMP_PROC_BIND: "false"
      OMP_NUM_THREADS: "10"
      HCCL_BUFFSIZE: "1024"
      HCCL_OP_EXPANSION_MODE: "AIV"
      PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
      SERVER_PORT: "DEFAULT_PORT"
    server_cmd:
      - "--quantization"
      - "ascend"
      - "--async-scheduling"
      - "--no-enable-prefix-caching"
      - "--tensor-parallel-size"
      - "1"
      - "--port"
      - "$SERVER_PORT"
      - "--max-model-len"
      - "5600"
      - "--max-num-batched-tokens"
      - "16384"
      - "--max-num-seqs"
      - "100"
      - "--trust-remote-code"
      - "--gpu-memory-utilization"
      - "0.9"
      - "--compilation-config"
      - '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
    benchmarks:
      perf:
        case_type: performance
        dataset_path: vllm-ascend/GSM8K-in3500-bs400
        request_conf: vllm_api_stream_chat
        dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
        num_prompts: 180
        max_out_len: 1500
        batch_size: 45
        request_rate: 0
        baseline: 1
        threshold: 0.97