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

test_cases:
  - name: "DeepSeek-V4-Flash-W8A8-A3"
    model: "Eco-Tech/DeepSeek-V4-Flash-w8a8-mtp"
    special_dependencies:
      transformers: "5.9.0"
    envs:
      OMP_PROC_BIND: "false"
      OMP_NUM_THREADS: "1"
      PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
      HCCL_BUFFSIZE: "1024"
      VLLM_ASCEND_ENABLE_FUSED_MC2: "0"
      VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
      ASCEND_LAUNCH_BLOCKING: "0"
      SERVER_PORT: "DEFAULT_PORT"
      VLLM_ENGINE_READY_TIMEOUT_S: "3000"
    server_cmd:
      - "--enable-prefix-caching"
      - "--max-model-len"
      - "1048576"
      - "--max-num-batched-tokens"
      - "10240"
      - "--gpu-memory-utilization"
      - "0.9"
      - "--max-num-seqs"
      - "64"
      - "--data-parallel-size"
      - "4"
      - "--tensor-parallel-size"
      - "4"
      - "--enable-expert-parallel"
      - "--tokenizer-mode"
      - "deepseek_v4"
      - "--tool-call-parser"
      - "deepseek_v4"
      - "--enable-auto-tool-choice"
      - "--reasoning-parser"
      - "deepseek_v4"
      - "--safetensors-load-strategy"
      - "prefetch"
      - "--quantization"
      - "ascend"
      - "--api-server-count"
      - "1"
      - "--speculative-config"
      - '{"num_speculative_tokens": 1,"method": "mtp","enforce_eager": true}'
      - "--port"
      - "$SERVER_PORT"
      - "--block-size"
      - "128"
      - "--compilation-config"
      - '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
      - "--async-scheduling"
      - "--additional-config"
      - '{"ascend_compilation_config":{"enable_npugraph_ex":true,"enable_static_kernel":false},"enable_cpu_binding":"true","enable_shared_expert_dp":true,"multistream_overlap_shared_expert":true}'
    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
        max_out_len: 65536
        batch_size: 32
        baseline: 86.36
        threshold: 5
        thinking: true
      perf:
        case_type: performance
        dataset_path: vllm-ascend/GSM8K_prefix0_in32768_bs1000_deepseek
        request_conf: vllm_api_stream_chat
        dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
        num_prompts: 64
        max_out_len: 1024
        batch_size: 16
        request_rate: 0
        baseline: 1
        threshold: 0.97