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

test_cases:
  - name: "MiniMax-M2.5-w8a8"
    model: "Eco-Tech/MiniMax-M2.5-w8a8-QuaRot"
    envs:
      HCCL_BUFFSIZE: "512"
      HCCL_OP_EXPANSION_MODE: "AIV"
      PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
      OMP_NUM_THREADS: "1"
      TASK_QUEUE_ENABLE: "1"
      VLLM-ASCEND_ENABLE_FUSED_MC2: "2"
      VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
      SERVER_PORT: "DEFAULT_PORT"
    server_cmd:
      - "--no-enable-prefix-caching"
      - "--tensor-parallel-size"
      - "4"
      - "--data-parallel-size"
      - "4"
      - "--port"
      - "$SERVER_PORT"
      - "--enable-expert-parallel"
      - "--max-num-seqs"
      - "128"
      - "--max-model-len"
      - "153600"
      - "--max-num-batched-tokens"
      - "16384"
      - "--trust-remote-code"
      - "--gpu-memory-utilization"
      - "0.8"
      - "--quantization"
      - "ascend"
      - "--compilation-config"
      - '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
      - "--additional-config"
      - '{"enable_cpu_binding":true}'
      - "--async-scheduling"
    benchmarks:
      acc:
        case_type: accuracy
        dataset_path: vllm-ascend/gsm8k-lite
        request_conf: vllm_api_general_chat
        dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
        max_out_len: 32768
        batch_size: 32
        baseline: 95
        threshold: 5
      perf:
        case_type: performance
        dataset_path: vllm-ascend/GSM8K-in3500-bs2800
        request_conf: vllm_api_stream_chat
        dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
        num_prompts: 80
        max_out_len: 1500
        batch_size: 20
        request_rate: 0
        baseline: 730.0832
        threshold: 0.97