xc-llm-ascend/tests/e2e/nightly/multi_node/config/DeepSeek-R1-W8A8-EPLB.yaml

test_name: "test DeepSeek-R1-W8A8 disaggregated_prefill"
model: "vllm-ascend/DeepSeek-R1-0528-W8A8"
num_nodes: 4
npu_per_node: 16
env_common:
  VLLM_USE_MODELSCOPE: true
  HCCL_BUFFSIZE: 1024
  SERVER_PORT: 8080
  OMP_PROC_BIND: false
  OMP_NUM_THREADS: 10
  PYTORCH_NPU_ALLOC_CONF: expandable_segments:True
  HCCL_DETERMINISTIC: True
  TASK_QUEUE_ENABLE: 1
  HCCL_OP_RETRY_ENABLE: "L0:0, L1:0, L2:0"
  DYNAMIC_EPLB: true

disaggregated_prefill:
  enabled: true
  prefiller_host_index: [0, 1]
  decoder_host_index: [2, 3]

deployment:
  -
    server_cmd: >
      vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
          --host 0.0.0.0
          --port $SERVER_PORT
          --data-parallel-size 2
          --data-parallel-size-local 2
          --tensor-parallel-size 8
          --enforce-eager
          --enable-expert-parallel
          --seed 1024
          --quantization ascend
          --max-num-seqs 4
          --max-model-len 36864
          --max-num-batched-tokens 16384
          --trust-remote-code
          --gpu-memory-utilization 0.9
          --speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
          --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": 32,
                            "tp_size": 1
                    }
              }
          }'
          --additional-config
          '{"enable_prefill_optimizations":true,"enable_weight_nz_layout":true,"eplb_config": {"dynamic_eplb":true,"expert_heat_collection_interval":2048,"algorithm_execution_interval":200}}'

  -
    server_cmd: >
      vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
          --host 0.0.0.0
          --port $SERVER_PORT
          --data-parallel-size 2
          --data-parallel-size-local 2
          --tensor-parallel-size 8
          --enforce-eager
          --enable-expert-parallel
          --seed 1024
          --quantization ascend
          --max-num-seqs 4
          --max-model-len 36864
          --max-num-batched-tokens 16384
          --trust-remote-code
          --gpu-memory-utilization 0.9
          --speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
          --kv-transfer-config
          '{"kv_connector": "MooncakeConnectorV1",
          "kv_role": "kv_producer",
          "kv_port": "30100",
          "engine_id": "1",
          "kv_connector_extra_config": {
                    "prefill": {
                            "dp_size": 2,
                            "tp_size": 8
                    },
                    "decode": {
                            "dp_size": 32,
                            "tp_size": 1
                    }
              }
          }'
          --additional-config
          '{"enable_prefill_optimizations":true,"enable_weight_nz_layout":true,"eplb_config": {"dynamic_eplb":true,"expert_heat_collection_interval":2048,"algorithm_execution_interval":200}}'
  -
    server_cmd: >
      vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
        --host 0.0.0.0
        --port $SERVER_PORT
        --data-parallel-size 32
        --data-parallel-size-local 16
        --data-parallel-start-rank 0
        --data-parallel-address $LOCAL_IP
        --data-parallel-rpc-port 13389
        --tensor-parallel-size 1
        --enable-expert-parallel
        --seed 1024
        --quantization ascend
        --max-num-seqs 28
        --max-model-len 36864
        --max-num-batched-tokens 256
        --trust-remote-code
        --gpu-memory-utilization 0.9
        --speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
        --kv-transfer-config
        '{"kv_connector": "MooncakeConnectorV1",
        "kv_role": "kv_consumer",
        "kv_port": "30200",
        "engine_id": "2",
        "kv_connector_extra_config": {
                  "prefill": {
                          "dp_size": 2,
                          "tp_size": 8
                  },
                  "decode": {
                          "dp_size": 32,
                          "tp_size": 1
                  }
            }
        }'
        --additional-config
        '{"multistream_overlap_shared_expert":true,"dynamic_eplb":true,"expert_heat_collection_interval":2048,"algorithm_execution_interval":200}'
  -
    server_cmd: >
      vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
        --headless
        --data-parallel-size 32
        --data-parallel-size-local 16
        --data-parallel-start-rank 16
        --data-parallel-address $MASTER_IP
        --data-parallel-rpc-port 13389
        --tensor-parallel-size 1
        --enable-expert-parallel
        --seed 1024
        --quantization ascend
        --max-num-seqs 28
        --max-model-len 36864
        --max-num-batched-tokens 256
        --trust-remote-code
        --gpu-memory-utilization 0.9
        --speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
        --kv-transfer-config
        '{"kv_connector": "MooncakeConnectorV1",
        "kv_role": "kv_consumer",
        "kv_port": "30200",
        "engine_id": "2",
        "kv_connector_extra_config": {
                  "prefill": {
                          "dp_size": 2,
                          "tp_size": 8
                  },
                  "decode": {
                          "dp_size": 32,
                          "tp_size": 1
                  }
            }
        }'
        --additional-config
        '{"multistream_overlap_shared_expert":true,"eplb_config": {"dynamic_eplb":true,"expert_heat_collection_interval":2048,"algorithm_execution_interval":200}}'
benchmarks:
  perf:
    case_type: performance
    dataset_path: vllm-ascend/GSM8K-in3500-bs2800
    request_conf: vllm_api_stream_chat
    dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
    num_prompts: 2800
    max_out_len: 1500
    batch_size: 700
    request_rate: 11.2
    baseline: 1
    threshold: 0.97
  acc:
    case_type: accuracy
    dataset_path: vllm-ascend/gsm8k
    request_conf: vllm_api_general_chat
    dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
    max_out_len: 32768
    batch_size: 512
    baseline: 95
    threshold: 5