DeepSeek-V4-Flash

DeepSeek-V4-Flash#

简介#

DeepSeek-V4在DeepSeek-V3基础上引入了多项关键升级。

流形约束超连接(mHC)增强传统残差连接；
混合注意力架构，通过Compress-4-Attention和Compress-128-Attention大幅提升长上下文效率。对于混合专家(MoE)组件，仍采用DeepSeekMoE架构，仅做少量调整。

本文档将展示模型的主要验证步骤，包括支持特性、特性配置、环境准备、单节点和多节点部署、精度及性能评估。

环境准备#

模型权重#

DeepSeek-V4-Flash-w8a8-mtp(量化版本)：需要1个Atlas 800 A3 (128G × 8)节点或1个Atlas 800 A2 (64G × 8)节点。下载模型权重

建议将模型权重下载到多节点共享目录，例如/root/.cache/

验证多节点通信(可选)#

如需部署多节点环境，需按照验证多节点通信环境验证多节点通信。

安装#

您可以直接使用官方docker镜像运行DeepSeek-V4。

A2系列

在每个节点上启动docker镜像。

export IMAGE=quay.io/ascend/vllm-ascend:v0.21.0rc1
export NAME=vllm-ascend
docker run --rm \
    --name $NAME \
    --net=host \
    --shm-size=512g \
    --device /dev/davinci0 \
    --device /dev/davinci1 \
    --device /dev/davinci2 \
    --device /dev/davinci3 \
    --device /dev/davinci4 \
    --device /dev/davinci5 \
    --device /dev/davinci6 \
    --device /dev/davinci7 \
    --device /dev/davinci_manager \
    --device /dev/devmm_svm \
    --device /dev/hisi_hdc \
    -v /usr/local/dcmi:/usr/local/dcmi \
    -v /usr/local/Ascend/driver/tools/hccn_tool:/usr/local/Ascend/driver/tools/hccn_tool \
    -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
    -v /usr/local/Ascend/driver/lib64/:/usr/local/Ascend/driver/lib64/ \
    -v /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info \
    -v /etc/ascend_install.info:/etc/ascend_install.info \
    -v /etc/hccn.conf:/etc/hccn.conf \
    -v /mnt/sfs_turbo/.cache:/root/.cache \
    -it $IMAGE bash

A3系列

在每个节点上启动docker镜像。

export IMAGE=quay.io/ascend/vllm-ascend:v0.21.0rc1-a3
export NAME=vllm-ascend
docker run --rm \
    --name $NAME \
    --net=host \
    --shm-size=512g \
    --device /dev/davinci0 \
    --device /dev/davinci1 \
    --device /dev/davinci2 \
    --device /dev/davinci3 \
    --device /dev/davinci4 \
    --device /dev/davinci5 \
    --device /dev/davinci6 \
    --device /dev/davinci7 \
    --device /dev/davinci8 \
    --device /dev/davinci9 \
    --device /dev/davinci10 \
    --device /dev/davinci11 \
    --device /dev/davinci12 \
    --device /dev/davinci13 \
    --device /dev/davinci14 \
    --device /dev/davinci15 \
    --device /dev/davinci_manager \
    --device /dev/devmm_svm \
    --device /dev/hisi_hdc \
    -v /usr/local/dcmi:/usr/local/dcmi \
    -v /usr/local/Ascend/driver/tools/hccn_tool:/usr/local/Ascend/driver/tools/hccn_tool \
    -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
    -v /usr/local/Ascend/driver/lib64/:/usr/local/Ascend/driver/lib64/ \
    -v /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info \
    -v /etc/ascend_install.info:/etc/ascend_install.info \
    -v /etc/hccn.conf:/etc/hccn.conf \
    -v /mnt/sfs_turbo/.cache:/root/.cache \
    -it $IMAGE bash

部署#

备注

在本教程中，我们假设您已将模型权重下载到/root/.cache/。您可以根据需要更改为自己的路径。

单节点部署#

DeepSeek-V4-Flash-w8a8-mtp：可部署在1个Atlas 800 A3 (128G × 8)或1个Atlas 800 A2 (64G × 8)上。

分别在每个节点上运行以下脚本。

A2系列

运行以下脚本执行在线推理。

export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD
export HCCL_BUFFSIZE=1024
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1
export TASK_QUEUE_ENABLE=1
export HCCL_OP_EXPANSION_MODE="AIV"

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
    --max_model_len 133120 \
    --max-num-batched-tokens 8192 \
    --served-model-name dsv4 \
    --gpu-memory-utilization 0.9 \
    --max-num-seqs 32 \
    --data-parallel-size 1 \
    --tensor-parallel-size 8 \
    --enable-expert-parallel \
    --tokenizer-mode deepseek_v4 \
    --tool-call-parser deepseek_v4 \
    --enable-auto-tool-choice \
    --reasoning-parser deepseek_v4 \
    --safetensors-load-strategy 'prefetch' \
    --no-enable-prefix-caching \
    --model-loader-extra-config='{"enable_multithread_load": "true", "num_threads": 128}' \
    --quantization ascend \
    --port 8900 \
    --block-size 128 \
    --speculative-config '{"num_speculative_tokens": 1,"method": "mtp","enforce_eager": true}' \
    --compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}'\
    --async-scheduling \
    --additional-config '
    {"ascend_compilation_config":{
        "enable_npugraph_ex":true,
        "enable_static_kernel":false
        },
    "enable_cpu_binding": true,
    "enable_dsa_cp": true,
    "multistream_overlap_shared_expert":true}'

A3系列

运行以下脚本执行在线推理。

export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD
export HCCL_BUFFSIZE=1024
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1
export TASK_QUEUE_ENABLE=1   
export HCCL_OP_EXPANSION_MODE="AIV"

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
    --max_model_len 1048576 \
    --max-num-batched-tokens 10240 \
    --served-model-name dsv4 \
    --gpu-memory-utilization 0.9 \
    --api-server-count 1 \
    --max-num-seqs 64 \
    --data-parallel-size 4 \
    --tensor-parallel-size 4 \
    --enable-expert-parallel \
    --tokenizer-mode deepseek_v4 \
    --tool-call-parser deepseek_v4 \
    --enable-auto-tool-choice \
    --reasoning-parser deepseek_v4 \
    --safetensors-load-strategy 'prefetch' \
    --model-loader-extra-config='{"enable_multithread_load": "true", "num_threads": 128}' \
    --quantization ascend \
    --port 8900 \
    --block-size 128 \
    --speculative-config '{"num_speculative_tokens": 1,"method": "mtp","enforce_eager": true}' \
    --compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}'\
    --async-scheduling \
    --additional-config '
    {"ascend_compilation_config":{
        "enable_npugraph_ex":true,
        "enable_static_kernel":false
        },
    "enable_cpu_binding": true,
    "multistream_overlap_shared_expert":true}'

Prefill-Decode分离式部署#

我们将展示DeepSeek-V4在Atlas 800 A3 (128G × 8)多节点环境下的部署指南，采用1P1D配置以获得更优性能。

开始之前，请

在每个节点上准备脚本launch_online_dp.py。

import argparse
import multiprocessing
import os
import subprocess
import sys

def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--dp-size",
        type=int,
        required=True,
        help="Data parallel size."
    )
    parser.add_argument(
        "--tp-size",
        type=int,
        default=1,
        help="Tensor parallel size."
    )
    parser.add_argument(
        "--dp-size-local",
        type=int,
        default=-1,
        help="Local data parallel size."
    )
    parser.add_argument(
        "--dp-rank-start",
        type=int,
        default=0,
        help="Starting rank for data parallel."
    )
    parser.add_argument(
        "--dp-address",
        type=str,
        required=True,
        help="IP address for data parallel master node."
    )
    parser.add_argument(
        "--dp-rpc-port",
        type=str,
        default=12345,
        help="Port for data parallel master node."
    )
    parser.add_argument(
        "--vllm-start-port",
        type=int,
        default=9000,
        help="Starting port for the engine."
    )
    return parser.parse_args()

args = parse_args()
dp_size = args.dp_size
tp_size = args.tp_size
dp_size_local = args.dp_size_local
if dp_size_local == -1:
    dp_size_local = dp_size
dp_rank_start = args.dp_rank_start
dp_address = args.dp_address
dp_rpc_port = args.dp_rpc_port
vllm_start_port = args.vllm_start_port

def run_command(visible_devices, dp_rank, vllm_engine_port):
    command = [
        "bash",
        "./run_dp_template.sh",
        visible_devices,
        str(vllm_engine_port),
        str(dp_size),
        str(dp_rank),
        dp_address,
        dp_rpc_port,
        str(tp_size),
    ]
    subprocess.run(command, check=True)

if __name__ == "__main__":
    template_path = "./run_dp_template.sh"
    if not os.path.exists(template_path):
        print(f"Template file {template_path} does not exist.")
        sys.exit(1)

    processes = []
    num_cards = dp_size_local * tp_size
    for i in range(dp_size_local):
        dp_rank = dp_rank_start + i
        vllm_engine_port = vllm_start_port + i
        visible_devices = ",".join(str(x) for x in range(i * tp_size, (i + 1) * tp_size))
        process = multiprocessing.Process(target=run_command,
                                        args=(visible_devices, dp_rank,
                                                vllm_engine_port))
        processes.append(process)
        process.start()

    for process in processes:
        process.join()

在每个节点上准备脚本run_dp_template.sh。

Prefill节点

nic_name="xxxx" # change to your own nic name
local_ip=xx.xx.xx.1 # change to your own ip

export HCCL_IF_IP=$local_ip
export GLOO_SOCKET_IFNAME=$nic_name
export TP_SOCKET_IFNAME=$nic_name
export HCCL_SOCKET_IFNAME=$nic_name
export VLLM_RPC_TIMEOUT=3600000
export VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=30000
export HCCL_EXEC_TIMEOUT=204
export HCCL_CONNECT_TIMEOUT=120
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=2560
export TASK_QUEUE_ENABLE=1
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1
export HCCL_OP_EXPANSION_MODE="AIV"
export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD
export ASCEND_RT_VISIBLE_DEVICES=$1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
    --host 0.0.0.0 \
    --port $2 \
    --data-parallel-size $3 \
    --data-parallel-rank $4 \
    --data-parallel-address $5 \
    --data-parallel-rpc-port $6 \
    --tensor-parallel-size $7 \
    --enable-expert-parallel \
    --seed 1024 \
    --served-model-name dsv4 \
    --max-model-len 1048576 \
    --max-num-batched-tokens 8192 \
    --max-num-seqs 16 \
    --no-disable-hybrid-kv-cache-manager \
    --model-loader-extra-config='{"enable_multithread_load": "true", "num_threads": 128}' \
    --no-enable-prefix-caching \
    --safetensors-load-strategy 'prefetch' \
    --speculative-config '{"num_speculative_tokens": 1,"method": "mtp","enforce_eager": true}' \
    --trust-remote-code \
    --block-size 128 \
    --tokenizer-mode deepseek_v4 \
    --tool-call-parser deepseek_v4 \
    --enable-auto-tool-choice \
    --reasoning-parser deepseek_v4 \
    --gpu-memory-utilization 0.9 \
    --quantization ascend \
    --enforce-eager \
    --additional-config '{"enable_cpu_binding": true, "enable_shared_expert_dp": true,  "enable_dsa_cp": true}' \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeHybridConnector",
    "kv_role": "kv_producer",
    "kv_port": "30000",
    "engine_id": "0",
    "kv_connector_extra_config": {
                "prefill": {
                        "dp_size": 4,
                        "tp_size": 4
                },
                "decode": {
                        "dp_size": 16,
                        "tp_size": 1
                }
        }
    }'

Decode节点

nic_name="xxxx" # change to your own nic name
local_ip=xx.xx.xx.2 # change to your own ip

export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD
export HCCL_OP_EXPANSION_MODE="AIV"
export TASK_QUEUE_ENABLE=1
export VLLM_RPC_TIMEOUT=3600000
export VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=30000
export HCCL_EXEC_TIMEOUT=204
export HCCL_CONNECT_TIMEOUT=1200
export HCCL_IF_IP=$local_ip
export GLOO_SOCKET_IFNAME=$nic_name
export TP_SOCKET_IFNAME=$nic_name
export HCCL_SOCKET_IFNAME=$nic_name
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=1024
export ASCEND_RT_VISIBLE_DEVICES=$1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
    --host 0.0.0.0 \
    --port $2 \
    --data-parallel-size $3 \
    --data-parallel-rank $4 \
    --data-parallel-address $5 \
    --data-parallel-rpc-port $6 \
    --tensor-parallel-size $7 \
    --enable-expert-parallel \
    --seed 1024 \
    --served-model-name dsv4 \
    --max-model-len 1048576 \
    --max-num-batched-tokens 120 \
    --max-num-seqs 60 \
    --async-scheduling \
    --block-size 128 \
    --no-disable-hybrid-kv-cache-manager \
    --no-enable-prefix-caching \
    --safetensors-load-strategy 'prefetch' \
    --trust-remote-code \
    --tokenizer-mode deepseek_v4 \
    --model-loader-extra-config='{"enable_multithread_load": "true", "num_threads": 128}' \
    --tool-call-parser deepseek_v4 \
    --enable-auto-tool-choice \
    --reasoning-parser deepseek_v4 \
    --gpu-memory-utilization 0.9 \
    --quantization ascend \
    --speculative-config '{"num_speculative_tokens": 1,"method": "mtp","enforce_eager": true}' \
    --compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeHybridConnector",
    "kv_role": "kv_consumer",
    "kv_port": "30100",
    "engine_id": "1",
    "kv_connector_extra_config": {
                "prefill": {
                        "dp_size": 4,
                        "tp_size": 4
                },
                "decode": {
                        "dp_size": 16,
                        "tp_size": 1
                }
        }
    }' \
    --additional-config '{
        "ascend_compilation_config":{
            "enable_npugraph_ex":true,
            "enable_static_kernel":false
        },
        "enable_cpu_binding":true,
        "multistream_overlap_shared_expert":true,
        "recompute_scheduler_enable":true
    }'

准备工作完成后，您可以在每个节点上使用以下命令启动服务器：

Prefill节点

# change ip to your own
python launch_online_dp.py --dp-size 4 --tp-size 4 --dp-size-local 4 --dp-rank-start 0 --dp-address xx.xx.xx.1 --dp-rpc-port 12321 --vllm-start-port 7100

Decode节点0

# change ip to your own
python launch_online_dp.py --dp-size 16 --tp-size 1 --dp-size-local 16 --dp-rank-start 0 --dp-address xx.xx.xx.2 --dp-rpc-port 12321 --vllm-start-port 7100

最后，参考Prefill-Decode分离式部署(Deepseek)部署P-D分离代理。

对于Atlas 800 A2系列机器，可以按如下方式配置部署(4*1P 1*4D)：

开始之前，请

在每个节点上准备脚本launch_online_dp.py。

    import argparse
    import multiprocessing
    import os
    import subprocess
    import sys


    def parse_args():
        parser = argparse.ArgumentParser()
        parser.add_argument("--dp-size", type=int, required=True, help="Data parallel size.")
        parser.add_argument("--tp-size", type=int, default=1, help="Tensor parallel size.")
        parser.add_argument("--dp-size-local", type=int, default=-1, help="Local data parallel size.")
        parser.add_argument("--dp-rank-start", type=int, default=0, help="Starting rank for data parallel.")
        parser.add_argument("--dp-address", type=str, required=True, help="IP address for data parallel master node.")
        parser.add_argument("--dp-rpc-port", type=str, default=12345, help="Port for data parallel master node.")
        parser.add_argument("--vllm-start-port", type=int, default=9000, help="Starting port for the engine.")
        return parser.parse_args()


    args = parse_args()
    dp_size = args.dp_size
    tp_size = args.tp_size
    dp_size_local = args.dp_size_local
    if dp_size_local == -1:
        dp_size_local = dp_size
    dp_rank_start = args.dp_rank_start
    dp_address = args.dp_address
    dp_rpc_port = args.dp_rpc_port
    vllm_start_port = args.vllm_start_port


    def run_command(visible_devices, dp_rank, vllm_engine_port):
        command = [
            "bash",
            "./run_dp_template.sh",
            visible_devices,
            str(vllm_engine_port),
            str(dp_size),
            str(dp_rank),
            dp_address,
            dp_rpc_port,
            str(tp_size),
        ]
        subprocess.run(command, check=True)


    if __name__ == "__main__":
        template_path = "./run_dp_template.sh"
        if not os.path.exists(template_path):
            print(f"Template file {template_path} does not exist.")
            sys.exit(1)

        processes = []
        num_cards = dp_size_local * tp_size
        for i in range(dp_size_local):
            dp_rank = dp_rank_start + i
            vllm_engine_port = vllm_start_port + i
            visible_devices = ",".join(str(x) for x in range(i * tp_size, (i + 1) * tp_size))
            process = multiprocessing.Process(target=run_command, args=(visible_devices, dp_rank, vllm_engine_port))
            processes.append(process)
            process.start()

        for process in processes:
            process.join()

在每个节点上准备脚本run_dp_template.sh。

Prefill节点

unset ftp_proxy
unset https_proxy
unset http_proxy
rm -rf ~/ascend/log

nic_name="xxxxxx" #eg."enp67s0f0np0"
local_ip=`hostname -I|awk -F " " '{print$1}'`

export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD
export HCCL_OP_EXPANSION_MODE="AIV"
export TASK_QUEUE_ENABLE=1
export VLLM_RPC_TIMEOUT=3600000
export VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=30000
export HCCL_EXEC_TIMEOUT=204
export HCCL_CONNECT_TIMEOUT=1200

export HCCL_IF_IP=$local_ip
export GLOO_SOCKET_IFNAME=$nic_name
export TP_SOCKET_IFNAME=$nic_name
export HCCL_SOCKET_IFNAME=$nic_name
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=1024

export ASCEND_RT_VISIBLE_DEVICES=$1
export TASK_QUEUE_ENABLE=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
    --host 0.0.0.0 \
    --port $2 \
    --data-parallel-size $3 \
    --data-parallel-rank $4 \
    --data-parallel-address $5 \
    --data-parallel-rpc-port $6 \
    --tensor-parallel-size $7 \
    --enable-expert-parallel \
    --seed 1024 \
    --served-model-name deepseek_v4 \
    --max-model-len 135000 \
    --max-num-batched-tokens 4096 \
    --max-num-seqs 16 \
    --block-size 128 \
    --enforce-eager \
    --async-scheduling \
    --no-disable-hybrid-kv-cache-manager \
    --enable-prefix-caching \
    --trust-remote-code \
    --gpu-memory-utilization 0.9 \
    --quantization ascend \
    --safetensors-load-strategy 'prefetch' \
    --model-loader-extra-config='{"enable_multithread_load": "true", "num_threads": 128}' \
    --tokenizer-mode deepseek_v4 \
    --tool-call-parser deepseek_v4 \
    --enable-auto-tool-choice \
    --reasoning-parser deepseek_v4 \
    --additional-config '{"enable_cpu_binding": true, "enable_shared_expert_dp": true}' \
    --speculative-config '{"num_speculative_tokens": 1, "method": "mtp","enforce_eager": true}' \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeHybridConnector",
    "kv_role": "kv_producer",
    "kv_port": "30000",
    "engine_id": "0",
    "kv_connector_extra_config": {
                "prefill": {
                        "dp_size": 8,
                        "tp_size": 1
                },
                "decode": {
                        "dp_size": 32,
                        "tp_size": 1
                }
        }
    }'
            

对于每个P实例，只需修改两个配置值：“kv_port”和“engine_id”。“engine_id”应从0开始依次递增，而“kv_port”(例如“30100”)必须对每个P实例唯一，如30000、30100等。

Decode节点(与另一个D节点相同)

unset ftp_proxy
unset ftp_proxy
unset https_proxy
unset http_proxy
rm -rf ~/ascend/log

nic_name="xxxxxx" #eg."enp67s0f0np0"
local_ip=`hostname -I|awk -F " " '{print$1}'`

export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD
export HCCL_OP_EXPANSION_MODE="AIV"
export TASK_QUEUE_ENABLE=1
export VLLM_RPC_TIMEOUT=3600000
export VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=30000
export HCCL_EXEC_TIMEOUT=204
export HCCL_CONNECT_TIMEOUT=1200

export HCCL_IF_IP=$local_ip
export GLOO_SOCKET_IFNAME=$nic_name
export TP_SOCKET_IFNAME=$nic_name
export HCCL_SOCKET_IFNAME=$nic_name
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=1024

export ASCEND_RT_VISIBLE_DEVICES=$1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
    --host 0.0.0.0 \
    --port $2 \
    --data-parallel-size $3 \
    --data-parallel-rank $4 \
    --data-parallel-address $5 \
    --data-parallel-rpc-port $6 \
    --tensor-parallel-size $7 \
    --enable-expert-parallel \
    --seed 1024 \
    --served-model-name deepseek_v4 \
    --max-model-len 135000 \
    --max-num-batched-tokens 60 \
    --max-num-seqs 30 \
    --async-scheduling \
    --block-size 128 \
    --no-disable-hybrid-kv-cache-manager \
    --no-enable-prefix-caching \
    --trust-remote-code \
    --gpu-memory-utilization 0.9 \
    --quantization ascend \
    --safetensors-load-strategy 'prefetch' \
    --model-loader-extra-config='{"enable_multithread_load": "true", "num_threads": 128}' \
    --tokenizer-mode deepseek_v4 \
    --tool-call-parser deepseek_v4 \
    --enable-auto-tool-choice \
    --reasoning-parser deepseek_v4 \
    --speculative-config '{"num_speculative_tokens": 1, "method": "mtp","enforce_eager": true}' \
    --compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeHybridConnector",
    "kv_role": "kv_consumer",
    "kv_port": "30400",
    "engine_id": "4",
    "kv_connector_extra_config": {
                "prefill": {
                        "dp_size": 8,
                        "tp_size": 1
                },
                "decode": {
                        "dp_size": 32,
                        "tp_size": 1
                }
        }
    }' \
    --additional-config '{
        "ascend_compilation_config":{
              "enable_npugraph_ex":true,
              "enable_static_kernel":false
        },
       "enable_cpu_binding":true,
       "multistream_overlap_shared_expert":true,
       "recompute_scheduler_enable":true
    }'

准备工作完成后，您可以在每个节点上使用以下命令启动服务器：

Prefill节点

# change ip to your own
python launch_online_dp.py --dp-size 8 --tp-size 1 --dp-size-local 8 --dp-rank-start 0 --dp-address x.x.x.x --dp-rpc-port 12321 --vllm-start-port 7100

对于每个P实例，仅--dp-address参数不同，必须配置为与其他实例在同一子网内的服务IP地址。

Decode节点

# change ip to your own
python launch_online_dp.py --dp-size 32 --tp-size 1 --dp-size-local 8 --dp-rank-start x --dp-address x.x.x.x --dp-rpc-port 12321 --vllm-start-port 7100

对于每个D实例，仅--dp-rank-start参数不同，应分别配置为0、8、16和24。每个实例的--dp-address必须设置为主D节点的IP地址，即--dp-rank-start设置为0的Decode实例的IP。

代理也通过参考Prefill-Decode分离式部署(Deepseek)实现。

对于超长序列场景，可通过调整PD(Prefill/Decode)比例和模型并行策略实现支持。例如，在1M序列场景中，可使用1*4P-1*4D比例，模型并行设置为DP4TP8模式。

功能验证#

服务器启动后，您可以使用输入提示查询模型：

curl http://<node0_ip>:<port>/v1/chat/completions \
    -H "Content-Type: application/json" \
    -d '{
        "model": "deepseek_v4",
        "messages": [
            {
                "role": "user",
                "content": "Who are you?"
            }
        ],
        "max_tokens": 256,
        "temperature": 0
    }'

精度评估#

以下是两种精度评估方法。

使用AISBench#

详情请参考使用AISBench。
执行后即可获取结果。

使用Language Model Evaluation Harness#

以gsm8k数据集作为测试数据集为例，在线模式下运行DeepSeek-V4的精度评估。

lm_eval安装请参考使用lm_eval。
运行lm_eval执行精度评估。

lm_eval \
  --model local-completions \
  --model_args model=/root/.cache/Eco-Tech/DeepSeek-V4-Flash-w8a8-mtp,base_url=http://127.0.0.1:8006/v1/completions,tokenized_requests=False,trust_remote_code=True \
  --tasks gsm8k \
  --output_path ./

执行后即可获取结果。

性能#

使用AISBench#

详情请参考使用AISBench进行性能评估。

使用vLLM Benchmark#

以DeepSeek-V4-Flash-w8a8-mtp为例运行性能评估。

更多详情请参考vllm benchmark。

vllm bench有三个子命令：

latency：基准测试单批请求的延迟。
serve：基准测试在线服务吞吐量。
throughput：基准测试离线推理吞吐量。

以serve为例，运行代码如下。

export VLLM_USE_MODELSCOPE=true
vllm bench serve --model /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp  --dataset-name random --random-input 200 --num-prompt 200 --request-rate 1 --save-result --result-dir ./