# For disaggregated mode, set is_disaggregated: true, and set the following parameters:
# Prefiller_index: the hosts index of the node running prefiller
# Decoder_index: the hosts index of the node running decoder
# Suppose we have **4 nodes** running a 2P1D setup (2 Prefillers + 1 Decoder):
#   ┌───────────────┬───────────────┬───────────────┬───────────────┐
#   │   node0       │   node1       │   node2       │   node3       │
#   │ Prefiller #1  │ Prefiller #2  │   Decoder     │   Decoder     │
#   └───────────────┴───────────────┴───────────────┴───────────────┘
# For the prefiller nodes. the hosts should be node0 and node1
# For the decoder nodes. we only have 1 decoder node(dp+tp+ep across node2 and node3. Where node3 is running with headless mode)
# So the prefiller_host_index is [0, 1], and the decoder_host_index is [2]
test_name: "test DeepSeek-V3 disaggregated_prefill"
model: "vllm-ascend/DeepSeek-V3-W8A8"
num_nodes: 2
npu_per_node: 16
env_common:
  VLLM_USE_MODELSCOPE: true
  OMP_PROC_BIND: false
  OMP_NUM_THREADS: 1
  TASK_QUEUE_ENABLE: 1
  HCCL_BUFFSIZE: 1024
  SERVER_PORT: 8080
  NUMEXPR_MAX_THREADS: 128
  DISAGGREGATED_PREFILL_PROXY_SCRIPT: "examples/disaggregated_prefill_v1/load_balance_proxy_server_example.py"
# For None kubernetes deployment, list the IPs of all nodes used in order as follow
# cluster_hosts: [10.0.0.103, 10.0.0.139]
disaggregated_prefill:
  enabled: true
  prefiller_host_index: [0]
  decoder_host_index: [1]

deployment:
  -
    server_cmd: >
        vllm serve "vllm-ascend/DeepSeek-V3-W8A8"
        --host 0.0.0.0
        --port $SERVER_PORT
        --data-parallel-size 2
        --data-parallel-size-local 2
        --tensor-parallel-size 8
        --seed 1024
        --enforce-eager
        --enable-expert-parallel
        --max-num-seqs 16
        --max-model-len 8192
        --max-num-batched-tokens 8192
        --quantization ascend
        --trust-remote-code
        --no-enable-prefix-caching
        --gpu-memory-utilization 0.9
        --kv-transfer-config
        '{"kv_connector": "MooncakeConnectorV1",
        "kv_role": "kv_producer",
        "kv_port": "30000",
        "engine_id": "0",
        "kv_connector_extra_config": {
                  "prefill": {
                          "dp_size": 2,
                          "tp_size": 8
                  },
                  "decode": {
                          "dp_size": 2,
                          "tp_size": 8
                  }
            }
        }'

  -
    server_cmd: >
        vllm serve "vllm-ascend/DeepSeek-V3-W8A8"
        --host 0.0.0.0
        --port $SERVER_PORT
        --data-parallel-size 2
        --data-parallel-size-local 2
        --tensor-parallel-size 8
        --seed 1024
        --quantization ascend
        --max-num-seqs 16
        --max-model-len 8192
        --max-num-batched-tokens 8192
        --enable-expert-parallel
        --trust-remote-code
        --no-enable-prefix-caching
        --gpu-memory-utilization 0.9
        --kv-transfer-config
        '{"kv_connector": "MooncakeConnectorV1",
        "kv_role": "kv_consumer",
        "kv_port": "30200",
        "engine_id": "1",
        "kv_connector_extra_config": {
                  "prefill": {
                          "dp_size": 2,
                          "tp_size": 8
                  },
                  "decode": {
                          "dp_size": 2,
                          "tp_size": 8
                  }
            }
        }'
benchmarks:
  acc:
    case_type: accuracy
    dataset_path: vllm-ascend/gsm8k-lite
    request_conf: vllm_api_general_chat
    dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
    max_out_len: 4096
    batch_size: 512
    baseline: 95
    threshold: 5