# ==========================================
# Shared Configurations
# ==========================================

_envs: &envs
  TASK_QUEUE_ENABLE: "1"
  HCCL_OP_EXPANSION_MODE: "AIV"
  VLLM_ASCEND_ENABLE_FLASHCOMM: "1"
  SERVER_PORT: "DEFAULT_PORT"

_server_cmd: &server_cmd
  - "--quantization"
  - "ascend"
  - "--no-enable-prefix-caching"
  - "--tensor-parallel-size"
  - "4"
  - "--port"
  - "$SERVER_PORT"
  - "--max-model-len"
  - "40960"
  - "--max-num-batched-tokens"
  - "40960"
  - "--block-size"
  - "128"
  - "--trust-remote-code"
  - "--reasoning-parser"
  - "qwen3"
  - "--gpu-memory-utilization"
  - "0.9"
  - "--async-scheduling"
  - "--additional-config"
  - '{"weight_prefetch_config":{"enabled":true}}'

_benchmarks: &benchmarks
  acc:
    case_type: accuracy
    dataset_path: vllm-ascend/aime2024
    request_conf: vllm_api_general_chat
    dataset_conf: aime2024/aime2024_gen_0_shot_chat_prompt
    max_out_len: 32768
    batch_size: 32
    baseline: 83.33
    threshold: 7

  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: 288
    max_out_len: 1500
    batch_size: 72
    baseline: 1
    threshold: 0.97

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

test_cases:
  - name: "Qwen3-32B-W8A8-aclgraph-a2"
    model: "vllm-ascend/Qwen3-32B-W8A8"
    envs:
      <<: *envs
    server_cmd: *server_cmd
    server_cmd_extra:
      - "--compilation-config"
      - '{"cudagraph_mode":"FULL_DECODE_ONLY","cudagraph_capture_sizes":[1,12,16,20,24,32,48,60,64,68,72,76,80]}'
    benchmarks:
      <<: *benchmarks

  - name: "Qwen3-32B-W8A8-single-a2"
    model: "vllm-ascend/Qwen3-32B-W8A8"
    envs:
      <<: *envs
    server_cmd: *server_cmd
    server_cmd_extra:
      - "--enforce-eager"
    benchmarks: