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

test_cases:
  - name: "Qwen3.5-397B-A17B-w8a8-mtp-longseq"
    model: "Eco-Tech/Qwen3.5-397B-A17B-w8a8-mtp"
    envs:
      VLLM_USE_MODELSCOPE: "true"
      PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
      HCCL_OP_EXPANSION_MODE: "AIV"
      HCCL_BUFFSIZE: "1024"
      OMP_NUM_THREADS: "1"
      TASK_QUEUE_ENABLE: "1"
      SERVER_PORT: "DEFAULT_PORT"
      VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
      VLLM_ENGINE_READY_TIMEOUT_S: "3000"
      VLLM_RPC_TIMEOUT: "6000"
    server_cmd:
      - "--tensor-parallel-size"
      - "8"
      - "--data-parallel-size"
      - "1"
      - "--prefill-context-parallel-size"
      - "2"
      - "--decode-context-parallel-size"
      - "4"
      - "--enable-expert-parallel"
      - "--port"
      - "$SERVER_PORT"
      - "--max-num-seqs"
      - "32"
      - "--quantization"
      - "ascend"
      - "--max-model-len"
      - "133120"
      - "--max-num-batched-tokens"
      - "16384"
      - "--trust-remote-code"
      - "--gpu-memory-utilization"
      - "0.9"
      - "--speculative_config"
      - '{"method": "qwen3_5_mtp", "num_speculative_tokens": 3, "enforce_eager": true}'
      - "--compilation-config"
      - '{"cudagraph_capture_sizes":[1,4,8,16,32,64], "cudagraph_mode":"FULL_DECODE_ONLY"}'
      - "--allowed-local-media-path"
      - "/"
      - "--mm-processor-cache-gb"
      - "0"
      - "--safetensors-load-strategy"
      - "lazy"
    benchmarks:
      acc_GPQA:
        case_type: accuracy
        dataset_path: vllm-ascend/gpqa
        request_conf: vllm_api_general_chat
        dataset_conf: gpqa/gpqa_gen_0_shot_cot_chat_prompt
        num_prompts: 50
        max_out_len: 32768
        batch_size: 32
        baseline: 84
        threshold: 8