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

_envs: &envs
  HCCL_BUFFSIZE: "512"
  SERVER_PORT: "DEFAULT_PORT"
  HCCL_OP_EXPANSION_MODE: "AIV"
  OMP_PROC_BIND: "false"
  OMP_NUM_THREADS: "1"
  PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
  TASK_QUEUE_ENABLE: "1"
  VLLM_ASCEND_ENABLE_MLAPO: "1"
  VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
  VLLM_ASCEND_ENABLE_NZ: "1"

_server_cmd: &server_cmd
  - "--enable-expert-parallel"
  - "--enable-prefix-caching"
  - "--enable-chunked-prefill"
  - "--allowed-local-media-path"
  - "/"
  - "--tensor-parallel-size"
  - "4"
  - "--data-parallel-size"
  - "4"
  - "--port"
  - "$SERVER_PORT"
  - "--max-model-len"
  - "133120"
  - "--max-num-batched-tokens"
  - "8192"
  - "--max-num-seqs"
  - "16"
  - "--quantization"
  - "ascend"
  - "--trust-remote-code"
  - "--gpu-memory-utilization"
  - "0.9"
  - "--seed"
  - "42"
  - "--compilation-config"
  - '{"cudagraph_capture_sizes":[4,8,12,16,32], "cudagraph_mode":"FULL_DECODE_ONLY"}'
  - "--speculative-config"
  - '{"method":"eagle3", "model":"lightseekorg/kimi-k2.5-eagle3", "num_speculative_tokens":3}'
  - "--additional-config"
  - '{"enable_shared_expert_dp":true}'
  - "--mm-processor-cache-gb"
  - "0"
  - "--mm-encoder-tp-mode"
  - "data"

_benchmarks: &benchmarks
  acc_aime2025:
    case_type: accuracy
    dataset_path: vllm-ascend/aime2025
    request_conf: vllm_api_general_chat
    dataset_conf: aime2025/aime2025_gen_0_shot_chat_prompt
    max_out_len: 65536
    temperature: 0.0
    top_p: 1
    top_k: -1
    repetition_penalty: 1.0
    batch_size: 32
    baseline: 95
    threshold: 10
  perf:
    case_type: performance
    dataset_path: vllm-ascend/GSM8K_prefix90_in131072_bs1000_kimi
    request_conf: vllm_api_stream_chat
    dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
    num_prompts: 8
    max_out_len: 1024
    batch_size: 2
    trust_remote_code: true
    request_rate: 0
    baseline: 1
    threshold: 0.97

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

test_cases:
  - name: "Kimi-K2.5-W4A8-Case"
    model: "Eco-Tech/Kimi-K2.5-W4A8"
    envs:
      <<: *envs
    server_cmd: *server_cmd
    benchmarks:
      <<: *benchmarks