test_name: "DeepSeek-V4-Pro-w4a8-1M-PD"
model: "Eco-Tech/DeepSeek-V4-Pro-w4a8-mtp"
num_nodes: 4
npu_per_node: 16

routing:
  type: "disaggregated_prefill"
  groups:
    prefiller: [ 0, 1 ]
    decoder: [ 2, 3 ]

config:
  - node_index: 0
    port_start: 7100
    dp_rpc_port: 12321
    dp_size: 2
    dp_size_local: 1
    dp_rank_start: 0
    tp_size: 16
    dp_address: "${NODE_0_IP}"

  - node_index: 1
    port_start: 7100
    dp_rpc_port: 12321
    dp_size: 2
    dp_size_local: 1
    dp_rank_start: 1
    tp_size: 16
    dp_address: "${NODE_0_IP}"

  - node_index: 2
    port_start: 7100
    dp_rpc_port: 12321
    dp_size: 2
    dp_size_local: 1
    dp_rank_start: 0
    tp_size: 16
    dp_address: "${NODE_2_IP}"

  - node_index: 3
    port_start: 7100
    dp_rpc_port: 12321
    dp_size: 2
    dp_size_local: 1
    dp_rank_start: 1
    tp_size: 16
    dp_address: "${NODE_2_IP}"

env_common: &env_common
  VLLM_USE_MODELSCOPE: "true"
  SERVER_PORT: "${PORT}"
  ASCEND_RT_VISIBLE_DEVICES: "${VISIBLE_DEVICES}"
  VLLM_RPC_TIMEOUT: "3600000"
  VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS: "30000"
  HCCL_EXEC_TIMEOUT: "204"
  OMP_PROC_BIND: "false"
  OMP_NUM_THREADS: "10"
  PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
  HCCL_BUFFSIZE: "1024"
  TASK_QUEUE_ENABLE: "1"
  HCCL_OP_EXPANSION_MODE: "AIV"

env_prefill: &env_prefill
  <<: *env_common
  HCCL_CONNECT_TIMEOUT: "6000"

env_decode: &env_decode
  <<: *env_common
  HCCL_CONNECT_TIMEOUT: "1200"

templates:
  - node_index: 0
    envs:
      <<: *env_prefill
    server_cmd_template:
      - --host
      - "0.0.0.0"
      - --port
      - $SERVER_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
      - --seed
      - "1024"
      - --max-model-len
      - "1048576"
      - --max-num-batched-tokens
      - "2048"
      - --max-num-seqs
      - "16"
      - --no-disable-hybrid-kv-cache-manager
      - --tokenizer-mode
      - "deepseek_v4"
      - --tool-call-parser
      - "deepseek_v4"
      - --enable-auto-tool-choice
      - --reasoning-parser
      - "deepseek_v4"
      - --model-loader-extra-config
      - '{"enable_multithread_load": true, "num_threads": 128}'
      - --trust-remote-code
      - --gpu-memory-utilization
      - "0.94"
      - --quantization
      - "ascend"
      - --block-size
      - "128"
      - --enforce-eager
      - --speculative-config
      - '{"num_speculative_tokens": 1,"method": "mtp","enforce_eager": true}'
      - --additional-config
      - '{"enable_cpu_binding": true}'
      - --kv-transfer-config
      - '{"kv_connector": "MooncakeHybridConnector", "kv_role": "kv_producer", "kv_port": "30100", "engine_id": "1", "kv_connector_extra_config": {"prefill": {"dp_size": 2, "tp_size": 16}, "decode": {"dp_size": 2, "tp_size": 16}}}'

  - node_index: 1
    envs:
      <<: *env_prefill
    server_cmd_template:
      - --host
      - "0.0.0.0"
      - --port
      - $SERVER_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
      - --seed
      - "1024"
      - --max-model-len
      - "1048576"
      - --max-num-batched-tokens
      - "2048"
      - --max-num-seqs
      - "16"
      - --no-disable-hybrid-kv-cache-manager
      - --tokenizer-mode
      - "deepseek_v4"
      - --tool-call-parser
      - "deepseek_v4"
      - --enable-auto-tool-choice
      - --reasoning-parser
      - "deepseek_v4"
      - --model-loader-extra-config
      - '{"enable_multithread_load": true, "num_threads": 128}'
      - --trust-remote-code
      - --gpu-memory-utilization
      - "0.94"
      - --quantization
      - "ascend"
      - --block-size
      - "128"
      - --enforce-eager
      - --speculative-config
      - '{"num_speculative_tokens": 1,"method": "mtp","enforce_eager": true}'
      - --additional-config
      - '{"enable_cpu_binding": true}'
      - --kv-transfer-config
      - '{"kv_connector": "MooncakeHybridConnector", "kv_role": "kv_producer", "kv_port": "30100", "engine_id": "1", "kv_connector_extra_config": {"prefill": {"dp_size": 2, "tp_size": 16}, "decode": {"dp_size": 2, "tp_size": 16}}}'

  - node_index: 2
    envs:
      <<: *env_decode
    server_cmd_template:
      - --host
      - "0.0.0.0"
      - --port
      - $SERVER_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
      - --seed
      - "1024"
      - --max-model-len
      - "1048576"
      - --max-num-batched-tokens
      - "60"
      - --max-num-seqs
      - "60"
      - --async-scheduling
      - --block-size
      - "128"
      - --no-enable-prefix-caching
      - --tokenizer-mode
      - "deepseek_v4"
      - --tool-call-parser
      - "deepseek_v4"
      - --enable-auto-tool-choice
      - --reasoning-parser
      - "deepseek_v4"
      - --no-disable-hybrid-kv-cache-manager
      - --model-loader-extra-config
      - '{"enable_multithread_load": true, "num_threads": 128}'
      - --trust-remote-code
      - --speculative-config
      - '{"num_speculative_tokens": 1,"method": "mtp","enforce_eager": true}'
      - --gpu-memory-utilization
      - "0.95"
      - --quantization
      - "ascend"
      - --compilation-config
      - '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
      - --kv-transfer-config
      - '{"kv_connector": "MooncakeHybridConnector", "kv_role": "kv_consumer", "kv_port": "30800", "engine_id": "8", "kv_connector_extra_config": {"prefill": {"dp_size": 2, "tp_size": 16}, "decode": {"dp_size": 2, "tp_size": 16}}}'
      - --additional-config
      - '{"ascend_compilation_config":{"enable_npugraph_ex":true,"enable_static_kernel":false},"enable_cpu_binding":true,"multistream_overlap_shared_expert":false,"recompute_scheduler_enable":true}'

  - node_index: 3
    envs:
      <<: *env_decode
    server_cmd_template:
      - --host
      - "0.0.0.0"
      - --port
      - $SERVER_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
      - --seed
      - "1024"
      - --max-model-len
      - "1048576"
      - --max-num-batched-tokens
      - "60"
      - --max-num-seqs
      - "60"
      - --async-scheduling
      - --block-size
      - "128"
      - --no-enable-prefix-caching
      - --tokenizer-mode
      - "deepseek_v4"
      - --tool-call-parser
      - "deepseek_v4"
      - --enable-auto-tool-choice
      - --reasoning-parser
      - "deepseek_v4"
      - --no-disable-hybrid-kv-cache-manager
      - --model-loader-extra-config
      - '{"enable_multithread_load": true, "num_threads": 128}'
      - --trust-remote-code
      - --speculative-config
      - '{"num_speculative_tokens": 1,"method": "mtp","enforce_eager": true}'
      - --gpu-memory-utilization
      - "0.95"
      - --quantization
      - "ascend"
      - --compilation-config
      - '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
      - --kv-transfer-config
      - '{"kv_connector": "MooncakeHybridConnector", "kv_role": "kv_consumer", "kv_port": "30800", "engine_id": "8", "kv_connector_extra_config": {"prefill": {"dp_size": 2, "tp_size": 16}, "decode": {"dp_size": 2, "tp_size": 16}}}'
      - --additional-config
      - '{"ascend_compilation_config":{"enable_npugraph_ex":true,"enable_static_kernel":false},"enable_cpu_binding":true,"multistream_overlap_shared_expert":false,"recompute_scheduler_enable":true}'

benchmarks:
  perf_1M_1k_prefix99_warmup:
    case_type: performance
    dataset_path: vllm-ascend/GSM8K-in1M-bs4-prefix99-ds
    request_conf: vllm_api_stream_chat
    dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
    num_prompts: 1
    max_out_len: 1
    batch_size: 1
    request_rate: 0
    baseline: 0
    threshold: 0.95
  perf_1M_1k_prefix99:
    case_type: performance
    dataset_path: vllm-ascend/GSM8K-in1M-bs4-prefix99-ds
    request_conf: vllm_api_stream_chat
    dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
    num_prompts: 4
    max_out_len: 1024
    batch_size: 1
    request_rate: 0
    baseline: 11.93
    threshold: 0.95