test_name: "MiniMax-PD-in32k-bs4-1"
model: "Eco-Tech/MiniMax-M2.5-w8a8-QuaRot"
num_nodes: 3
npu_per_node: 16

routing:
  type: "disaggregated_prefill"
  groups:
    prefiller: [0]
    decoder: [1, 2]

config:
  - node_index: 0
    port_start: 7100
    dp_rpc_port: 12321
    dp_size: 4
    dp_size_local: 4
    dp_rank_start: 0
    tp_size: 4
    dp_address: "${NODE_0_IP}"

  - node_index: 1
    port_start: 7100
    dp_rpc_port: 12321
    dp_size: 4
    dp_size_local: 4
    dp_rank_start: 0
    tp_size: 4
    dp_address: "${NODE_1_IP}"

  - node_index: 2
    port_start: 7100
    dp_rpc_port: 12321
    dp_size: 4
    dp_size_local: 4
    dp_rank_start: 0
    tp_size: 4
    dp_address: "${NODE_2_IP}"

env_common: &env_common
  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: "1"
  VLLM_USE_MODELSCOPE: "true"
  ASCEND_RT_VISIBLE_DEVICES: "${VISIBLE_DEVICES}"


templates:
  - node_index: 0
    envs:
      <<: *env_common
      VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
      HCCL_BUFFSIZE: "1200"
    server_cmd_template:
      - --host
      - "0.0.0.0"
      - --port
      - ${PORT}
      - --data-parallel-size
      - ${DP_SIZE}
      - --data-parallel-rank
      - ${DP_RANK}
      - --data-parallel-address
      - ${DP_ADDRESS}
      - --data-parallel-rpc-port
      - ${DP_RPC_PORT}
      - --tensor-parallel-size
      - ${TP_SIZE}
      - --enable-expert-parallel
      - --max-model-len
      - "40960"
      - --max-num-batched-tokens
      - "16384"
      - --max-num-seqs
      - "32"
      - --trust-remote-code
      - --gpu-memory-utilization
      - "0.8"
      - --quantization
      - "ascend"
      - --enforce-eager
      - --no-enable-prefix_caching
      - --speculative-config
      - '{"method":"eagle3","model":"vllm-ascend/MiniMax-M2.5-eagle-model-0318","num_speculative_tokens":3}'
      - --kv-transfer-config
      - '{"kv_connector": "MooncakeConnectorV1",
        "kv_role": "kv_producer",
        "kv_port": 30000,
        "kv_connector_extra_config": {
          "prefill": {
            "dp_size": 4,
            "tp_size": 4
          },
          "decode": {
            "dp_size": 4,
            "tp_size": 4
          }
        }}'

  - node_index: 1
    envs:
      <<: *env_common
      VLLM_ASCEND_ENABLE_FLASHCOMM1: "0"
      HCCL_BUFFSIZE: "512"
    server_cmd_template:
      - --host
      - "0.0.0.0"
      - --port
      - ${PORT}
      - --data-parallel-size
      - ${DP_SIZE}
      - --data-parallel-rank
      - ${DP_RANK}
      - --data-parallel-address
      - ${DP_ADDRESS}
      - --data-parallel-rpc-port
      - ${DP_RPC_PORT}
      - --tensor-parallel-size
      - ${TP_SIZE}
      - --enable-expert-parallel
      - --max-model-len
      - "40960"
      - --max-num-batched-tokens
      - "256"
      - --max-num-seqs
      - "16"
      - --trust-remote-code
      - --gpu-memory-utilization
      - "0.9"
      - --quantization
      - "ascend"
      - --no-enable-prefix_caching
      - --speculative-config
      - '{"method":"eagle3","model":"vllm-ascend/MiniMax-M2.5-eagle-model-0318","num_speculative_tokens":3}'
      - --compilation-config
      - '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
      - --additional-config
      - '{"recompute_scheduler_enable": true}'
      - --kv-transfer-config
      - '{"kv_connector": "MooncakeConnectorV1",
        "kv_role": "kv_consumer",
        "kv_port": 30100,
        "kv_connector_extra_config": {
          "prefill": {
            "dp_size": 4,
            "tp_size": 4
          },
          "decode": {
            "dp_size": 4,
            "tp_size": 4
          }
        }}'

  - node_index: 2
    envs:
      <<: *env_common
      VLLM_ASCEND_ENABLE_FLASHCOMM1: "0"
      HCCL_BUFFSIZE: "512"
    server_cmd_template:
      - --host
      - "0.0.0.0"
      - --port
      - ${PORT}
      - --data-parallel-size
      - ${DP_SIZE}
      - --data-parallel-rank
      - ${DP_RANK}
      - --data-parallel-address
      - ${DP_ADDRESS}
      - --data-parallel-rpc-port
      - ${DP_RPC_PORT}
      - --tensor-parallel-size
      - ${TP_SIZE}
      - --enable-expert-parallel
      - --max-model-len
      - "40960"
      - --max-num-batched-tokens
      - "256"
      - --max-num-seqs
      - "16"
      - --trust-remote-code
      - --gpu-memory-utilization
      - "0.9"
      - --quantization
      - "ascend"
      - --no-enable-prefix_caching
      - --speculative-config
      - '{"method":"eagle3","model":"vllm-ascend/MiniMax-M2.5-eagle-model-0318","num_speculative_tokens":3}'
      - --compilation-config
      - '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
      - --additional-config
      - '{"recompute_scheduler_enable": true}'
      - --kv-transfer-config
      - '{"kv_connector": "MooncakeConnectorV1",
        "kv_role": "kv_consumer",
        "kv_port": 30200,
        "kv_connector_extra_config": {
          "prefill": {
            "dp_size": 4,
            "tp_size": 4
          },
          "decode": {
            "dp_size": 4,
            "tp_size": 4
          }
        }}'

benchmarks:
  perf:
    case_type: performance
    dataset_path: vllm-ascend/GSM8K_prefix0_in32768_bs1000_Minimax
    request_conf: vllm_api_stream_chat
    dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
    num_prompts: 4
    max_out_len: 512
    batch_size: 1
    request_rate: 1
    baseline: 28.77
    threshold: 0.97