DeepSeek-V3/3.1

DeepSeek-V3/3.1#

简介#

DeepSeek-V3.1是一款混合模型，支持思维模式和非思维模式。与前一版本相比，此次升级在多个方面带来了改进。

  • 混合思维模式：通过更改chat template，一个模型支持思维模式和非思维模式。

  • 更智能的工具调用：通过后期训练优化，模型在工具使用和Agent任务方面的性能得到了显著提升。

  • 更高的思考效率：DeepSeek-V3.1-Think在回答质量上媲美DeepSeek-R1-0528，同时响应速度更快。

DeepSeek-V3.1模型首次在vllm-ascend:v0.9.1rc3中得到支持。

本文档将展示模型的主要验证步骤，包括支持的特性、特性配置、环境准备、单节点和多节点部署、准确性和性能评估。

支持的特性#

请参阅 支持的特性 获取模型的特性支持矩阵。

请参阅 特性指南 获取特性的配置方法。

环境准备#

模型权重#

  • DeepSeek-V3.1（BF16版本）：下载模型权重

  • DeepSeek-V3.1-w8a8（无mtp量化版本）：下载模型权重。

  • DeepSeek-V3.1_w8a8mix_mtp（带mix mtp量化版本）：下载模型权重。请在config.json中将torch_dtype从float16修改为bfloat16。

  • 量化 方法：msmodelslim。您可以使用这些方法对模型进行量化。

建议将模型权重下载到多节点的共享目录中，例如 /root/.cache/

验证多节点通信（可选）#

如果您想部署多节点环境，需要按照验证多节点通信环境来验证多节点通信。

安装#

您可以使用我们的官方docker镜像直接运行DeepSeek-V3.1。

根据您的机器类型选择镜像，并在您的节点上启动docker镜像，请参考使用docker进行设置。

# Update --device according to your device (Atlas A2: /dev/davinci[0-7] Atlas A3:/dev/davinci[0-15]).
# Update the vllm-ascend image according to your environment.
# Note you should download the weight to /root/.cache in advance.
# Update the vllm-ascend image
export IMAGE=m.daocloud.io/quay.io/ascend/vllm-ascend:v0.12.0rc1
export NAME=vllm-ascend

# Run the container using the defined variables
# Note: If you are running bridge network with docker, please expose available ports for multiple nodes communication in advance
docker run --rm \
--name $NAME \
--net=host \
--shm-size=1g \
--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 /root/.cache:/root/.cache \
-it $IMAGE bash

如果您想部署多节点环境，需要在每个节点上设置环境。

部署#

单节点部署#

  • 量化模型DeepSeek-V3.1_w8a8mix_mtp可以部署在1个Atlas 800 A3 (64G × 16)上。

运行以下脚本执行在线推理。

#!/bin/sh
# this obtained through ifconfig
# nic_name is the network interface name corresponding to local_ip of the current node
nic_name="xxxx"
local_ip="xxxx"

# [Optional] jemalloc
# jemalloc is for better performance, if `libjemalloc.so` is install on your machine, you can turn it on.
# export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD

# AIV
export HCCL_OP_EXPANSION_MODE="AIV"

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_ASCEND_ENABLE_MLAPO=1
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True

vllm serve /weights/DeepSeek-V3.1_w8a8mix_mtp \
--host 0.0.0.0 \
--port 8015 \
--data-parallel-size 4 \
--tensor-parallel-size 4 \
--quantization ascend \
--seed 1024 \
--served-model-name deepseek_v3 \
--enable-expert-parallel \
--max-num-seqs 16 \
--max-model-len 16384 \
--max-num-batched-tokens 4096 \
--trust-remote-code \
--no-enable-prefix-caching \
--gpu-memory-utilization 0.92 \
--speculative-config '{"num_speculative_tokens": 1, "method": "mtp"}' \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \

注意：参数解释如下。

  • 设置环境变量VLLM_ASCEND_ENABLE_MLAPO=1可以启用一个可以显著提升性能的融合算子，但它需要更多的NPU内存。因此，建议在NPU内存充足的情况下启用此选项。

  • 对于单节点部署，我们建议使用dp4tp4而不是dp2tp8。

  • --max-model-len指定最大上下文长度，即单个请求的输入和输出token的总和。对于输入长度为3.5K，输出长度为1.5K的性能测试，设置为16384已经足够，但对于精度测试，请至少设置为35000。

  • --no-enable-prefix-caching表示禁用了前缀缓存。要启用它，请删除此选项。

多节点部署#

  • DeepSeek-V3.1_w8a8mix_mtp：至少需要2个Atlas 800 A2 (64G × 8)。

分别在两个节点上运行以下脚本。

节点 0

#!/bin/sh

# this obtained through ifconfig
# nic_name is the network interface name corresponding to local_ip of the current node
nic_name="xxxx"
local_ip="xxxx"

# The value of node0_ip must be consistent with the value of local_ip set in node0 (master node)
node0_ip="xxxx"

# [Optional] jemalloc
# jemalloc is for better performance, if `libjemalloc.so` is install on your machine, you can turn it on.
# export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD

# AIV
export HCCL_OP_EXPANSION_MODE="AIV"

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 VLLM_USE_V1=1
export HCCL_BUFFSIZE=200
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export VLLM_ASCEND_ENABLE_MLAPO=1
export HCCL_INTRA_PCIE_ENABLE=1
export HCCL_INTRA_ROCE_ENABLE=0

vllm serve /weights/DeepSeek-V3.1_w8a8mix_mtp \
--host 0.0.0.0 \
--port 8004 \
--data-parallel-size 4 \
--data-parallel-size-local 2 \
--data-parallel-address $node0_ip \
--data-parallel-rpc-port 13389 \
--tensor-parallel-size 4 \
--quantization ascend \
--seed 1024 \
--served-model-name deepseek_v3 \
--enable-expert-parallel \
--max-num-seqs 20 \
--max-model-len 16384 \
--max-num-batched-tokens 4096 \
--trust-remote-code \
--no-enable-prefix-caching \
--gpu-memory-utilization 0.94 \
--speculative-config '{"num_speculative_tokens": 1, "method": "mtp"}' \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \

节点 1

#!/bin/sh

# this obtained through ifconfig
# nic_name is the network interface name corresponding to local_ip of the current node
nic_name="xxx"
local_ip="xxx"

# The value of node0_ip must be consistent with the value of local_ip set in node0 (master node)
node0_ip="xxxx"

# [Optional] jemalloc
# jemalloc is for better performance, if `libjemalloc.so` is install on your machine, you can turn it on.
# export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD

# AIV
export HCCL_OP_EXPANSION_MODE="AIV"

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 HCCL_BUFFSIZE=200
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export VLLM_ASCEND_ENABLE_MLAPO=1
export HCCL_INTRA_PCIE_ENABLE=1
export HCCL_INTRA_ROCE_ENABLE=0

vllm serve /weights/DeepSeek-V3.1_w8a8mix_mtp \
--host 0.0.0.0 \
--port 8004 \
--headless \
--data-parallel-size 4 \
--data-parallel-size-local 2 \
--data-parallel-start-rank 2 \
--data-parallel-address $node0_ip \
--data-parallel-rpc-port 13389 \
--tensor-parallel-size 4 \
--quantization ascend \
--seed 1024 \
--served-model-name deepseek_v3 \
--enable-expert-parallel \
--max-num-seqs 20 \
--max-model-len 16384 \
--max-num-batched-tokens 4096 \
--trust-remote-code \
--no-enable-prefix-caching \
--gpu-memory-utilization 0.94 \
--speculative-config '{"num_speculative_tokens": 1, "method": "mtp"}' \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \

Prefill-Decode 分离#

我们推荐使用Mooncake进行部署：Mooncake。

以Atlas 800 A3 (64G × 16)为例，我们建议部署2P1D（4个节点）而不是1P1D（2个节点），因为1P1D情况下NPU内存不足以服务高并发。

  • DeepSeek-V3.1_w8a8mix_mtp 2P1D Layerwise需要4个Atlas 800 A3 (64G × 16)。

要运行vllm-ascendPrefill-Decode Disaggregation服务，您需要在每个节点上部署launch_dp_program.py脚本和run_dp_template.sh脚本，并在预填充主节点上部署proxy.sh脚本来转发请求。

  1. 每个节点的launch_dp_program.py脚本

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()

  1. 预填充节点0的run_dp_template.sh脚本

# this obtained through ifconfig
# nic_name is the network interface name corresponding to local_ip of the current node
nic_name="xxx"
local_ip="141.xx.xx.1"

# The value of node0_ip must be consistent with the value of local_ip set in node0 (master node)
node0_ip="xxxx"

# [Optional] jemalloc
# jemalloc is for better performance, if `libjemalloc.so` is install on your machine, you can turn it on.
# export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD

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_VERSION="0.11.0"
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 VLLM_ASCEND_ENABLE_MLAPO=1
export HCCL_BUFFSIZE=256
export TASK_QUEUE_ENABLE=1
export HCCL_OP_EXPANSION_MODE="AIV"
export VLLM_USE_V1=1
export ASCEND_RT_VISIBLE_DEVICE=$1
export ASCEND_BUFFER_POOL=4:8
export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/latest/python/site-packages/mooncake:$LD_LIBRARY_PATH

vllm serve /weights/DeepSeek-V3.1_w8a8mix_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_v3 \
  --max-model-len 40000 \
  --max-num-batched-tokens 16384 \
  --max-num-seqs 8 \
  --enforce-eager \
  --trust-remote-code \
  --gpu-memory-utilization 0.9 \
  --quantization ascend \
  --no-enable-prefix-caching \
  --speculative-config '{"num_speculative_tokens": 1, "method": "mtp"}' \
  --additional-config '{"recompute_scheduler_enable":true,"enable_shared_expert_dp": true}' \
  --kv-transfer-config \
  '{"kv_connector": "MooncakeConnectorV1",
  "kv_role": "kv_producer",
  "kv_port": "30000",
  "engine_id": "0",
  "kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
  "kv_connector_extra_config": {
            "prefill": {
                    "dp_size": 2,
                    "tp_size": 8
             },
             "decode": {
                    "dp_size": 32,
                    "tp_size": 1
             }
      }
  }'

  1. 预填充节点1的run_dp_template.sh脚本

# this obtained through ifconfig
# nic_name is the network interface name corresponding to local_ip of the current node
nic_name="xxx"
local_ip="141.xx.xx.2"

# The value of node0_ip must be consistent with the value of local_ip set in node0 (master node)
node0_ip="xxxx"

# [Optional] jemalloc
# jemalloc is for better performance, if `libjemalloc.so` is install on your machine, you can turn it on.
# export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD

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_VERSION="0.11.0"
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 VLLM_ASCEND_ENABLE_MLAPO=1
export HCCL_BUFFSIZE=256
export TASK_QUEUE_ENABLE=1
export HCCL_OP_EXPANSION_MODE="AIV"
export VLLM_USE_V1=1
export ASCEND_RT_VISIBLE_DEVICE=$1
export ASCEND_BUFFER_POOL=4:8
export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/latest/python/site-packages/mooncake:$LD_LIBRARY_PATH

vllm serve /weights/DeepSeek-V3.1_w8a8mix_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_v3 \
  --max-model-len 40000 \
  --max-num-batched-tokens 16384 \
  --max-num-seqs 8 \
  --enforce-eager \
  --trust-remote-code \
  --gpu-memory-utilization 0.9 \
  --quantization ascend \
  --no-enable-prefix-caching \
  --speculative-config '{"num_speculative_tokens": 1, "method": "deepseek_mtp"}' \
  --additional-config '{"recompute_scheduler_enable":true,"enable_shared_expert_dp": true}' \
  --kv-transfer-config \
  '{"kv_connector": "MooncakeConnectorV1",
  "kv_role": "kv_producer",
  "kv_port": "30100",
  "engine_id": "1",
  "kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
  "kv_connector_extra_config": {
            "prefill": {
                    "dp_size": 2,
                    "tp_size": 8
             },
             "decode": {
                    "dp_size": 32,
                    "tp_size": 1
             }
      }
  }'

  1. 解码节点0的run_dp_template.sh脚本

# this obtained through ifconfig
# nic_name is the network interface name corresponding to local_ip of the current node
nic_name="xxx"
local_ip="141.xx.xx.3"

# The value of node0_ip must be consistent with the value of local_ip set in node0 (master node)
node0_ip="xxxx"

# [Optional] jemalloc
# jemalloc is for better performance, if `libjemalloc.so` is install on your machine, you can turn it on.
# export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD

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_VERSION="0.11.0"
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 VLLM_ASCEND_ENABLE_MLAPO=1
export HCCL_BUFFSIZE=600
export TASK_QUEUE_ENABLE=1
export HCCL_OP_EXPANSION_MODE="AIV"
export VLLM_USE_V1=1
export ASCEND_RT_VISIBLE_DEVICE=$1
export ASCEND_BUFFER_POOL=4:8
export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/latest/python/site-packages/mooncake:$LD_LIBRARY_PATH

vllm serve /weights/DeepSeek-V3.1_w8a8mix_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_v3 \
  --max-model-len 40000 \
  --max-num-batched-tokens 256 \
  --max-num-seqs 40 \
  --trust-remote-code \
  --gpu-memory-utilization 0.94 \
  --quantization ascend \
  --no-enable-prefix-caching \
  --speculative-config '{"num_speculative_tokens": 1, "method": "deepseek_mtp"}' \
  --compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
  --additional-config '{"recompute_scheduler_enable":true,"multistream_overlap_shared_expert": true,"lm_head_tensor_parallel_size":16}' \
  --kv-transfer-config \
  '{"kv_connector": "MooncakeConnectorV1",
  "kv_role": "kv_consumer",
  "kv_port": "30200",
  "engine_id": "2",
  "kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
  "kv_connector_extra_config": {
            "prefill": {
                    "dp_size": 2,
                    "tp_size": 8
             },
             "decode": {
                    "dp_size": 32,
                    "tp_size": 1
             }
      }
  }'

  1. 解码节点1的run_dp_template.sh脚本

# this obtained through ifconfig
# nic_name is the network interface name corresponding to local_ip of the current node
nic_name="xxx"
local_ip="141.xx.xx.4"

# The value of node0_ip must be consistent with the value of local_ip set in node0 (master node)
node0_ip="xxxx"

# [Optional] jemalloc
# jemalloc is for better performance, if `libjemalloc.so` is install on your machine, you can turn it on.
# export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD

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_VERSION="0.11.0"
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 VLLM_ASCEND_ENABLE_MLAPO=1
export HCCL_BUFFSIZE=600
export TASK_QUEUE_ENABLE=1
export HCCL_OP_EXPANSION_MODE="AIV"
export VLLM_USE_V1=1
export ASCEND_RT_VISIBLE_DEVICE=$1
export ASCEND_BUFFER_POOL=4:8
export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/latest/python/site-packages/mooncake:$LD_LIBRARY_PATH

vllm serve /weights/DeepSeek-V3.1_w8a8mix_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_v3 \
  --max-model-len 40000 \
  --max-num-batched-tokens 256 \
  --max-num-seqs 40 \
  --trust-remote-code \
  --gpu-memory-utilization 0.94 \
  --quantization ascend \
  --no-enable-prefix-caching \
  --speculative-config '{"num_speculative_tokens": 1, "method": "deepseek_mtp"}' \
  --compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
  --additional-config '{"recompute_scheduler_enable":true,"multistream_overlap_shared_expert": true,"lm_head_tensor_parallel_size":16}' \
  --kv-transfer-config \
  '{"kv_connector": "MooncakeConnectorV1",
  "kv_role": "kv_consumer",
  "kv_port": "30300",
  "engine_id": "3",
  "kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
  "kv_connector_extra_config": {
            "prefill": {
                    "dp_size": 2,
                    "tp_size": 8
             },
             "decode": {
                    "dp_size": 32,
                    "tp_size": 1
             }
      }
  }'

  1. 为每个节点运行服务器

# p0
python launch_dp_program.py --dp-size 2 --tp-size 8 --dp-size-local 2 --dp-rank-start 0 --dp-address 141.xx.xx.1 --dp-rpc-port 12321 --vllm-start-port 7100
# p1
python launch_dp_program.py --dp-size 2 --tp-size 8 --dp-size-local 2 --dp-rank-start 0 --dp-address 141.xx.xx.2 --dp-rpc-port 12321 --vllm-start-port 7100
# d0
python launch_dp_program.py --dp-size 32 --tp-size 1 --dp-size-local 16 --dp-rank-start 0 --dp-address 141.xx.xx.3 --dp-rpc-port 12321 --vllm-start-port 7100
# d1
python launch_dp_program.py --dp-size 32 --tp-size 1 --dp-size-local 16 --dp-rank-start 16 --dp-address 141.xx.xx.3 --dp-rpc-port 12321 --vllm-start-port 7100

  1. 预填充主节点的proxy.sh脚本

python load_balance_proxy_server_example.py \
  --port 1999 \
  --host 141.xx.xx.1 \
  --prefiller-hosts \
    141.xx.xx.1 \
    141.xx.xx.1 \
    141.xx.xx.2 \
    141.xx.xx.2 \
  --prefiller-ports \
    7100 7101 7100 7101 \
  --decoder-hosts \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.3 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
    141.xx.xx.4 \
  --decoder-ports \
    7100 7101 7102 7103 7104 7105 7106 7107 7108 7109 7110 7111 7112 7113 7114 7115 \
    7100 7101 7102 7103 7104 7105 7106 7107 7108 7109 7110 7111 7112 7113 7114 7115 \

  1. 运行代理

在与预填充服务实例相同的节点上运行代理服务器。您可以在仓库的examples中找到代理程序：load_balance_proxy_layerwise_server_example.py 或 load_balance_proxy_server_example.py。

cd vllm-ascend/examples/disaggregated_prefill_v1/
bash proxy.sh

功能验证#

服务器启动后，您可以用输入提示查询模型

curl http://<node0_ip>:<port>/v1/completions \
    -H "Content-Type: application/json" \
    -d '{
        "model": "deepseek_v3",
        "prompt": "The future of AI is",
        "max_tokens": 50,
        "temperature": 0
    }'

精度评估#

这里有两种准确率评估方法。

使用 AISBench#

  1. 详情请参阅使用 AISBench。

  2. 执行后，您将获得结果。以下是vllm-ascend:0.11.0rc1中DeepSeek-V3.1_w8a8mix_mtp的结果，仅供参考。

数据集

version

指标

模式

vllm-api-general-chat

备注

ceval

-

accuracy

gen

90.94

1个Atlas 800 A3 (64G × 16)

gsm8k

-

accuracy

gen

96.28

1个Atlas 800 A3 (64G × 16)

使用语言模型评估框架#

尚未测试。

性能#

使用 AISBench#

详情请参阅使用 AISBench 进行性能评估。

使用 vLLM Benchmark#

以DeepSeek-V3.1_w8a8mix_mtp的性能评估为例。

更多详情请参阅 vllm benchmark。

有三个 vllm bench 子命令

  • latency：对单批请求的延迟进行基准测试。

  • serve：对在线服务吞吐量进行基准测试。

  • throughput：对离线推理吞吐量进行基准测试。

以 serve 为例。运行代码如下。

vllm bench serve --model vllm-ascend/DeepSeek-V3.1_w8a8mix_mtp  --dataset-name random --random-input 1024 --num-prompt 200 --request-rate 1 --save-result --result-dir ./

大约几分钟后，您就可以得到性能评估结果。

目录