跳转至

GLM-5/GLM-5.1

1 引言

本文档同时适用于 GLM-5 和 GLM-5.1。除非另有说明，本文档中所有关于 GLM-5 的描述、配置和部署流程同样适用于 GLM-5.1。为简洁起见，下文统一使用 GLM-5 指代 GLM-5 和 GLM-5.1。

GLM-5 采用混合专家（MoE）架构，面向复杂的系统工程和长周期智能体任务。

GLM-5 模型首次在 vllm-ascend:v0.17.0rc1 中得到支持，所有 v0.17.0rc1 及更高版本 均可稳定运行。如需使用最新特性（如 PD 分离、MTP），建议使用最新的候选发布版或正式版。transformers 版本需升级至 5.2.0 或更高版本。

本文档将展示该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、精度及性能评估。

2 支持的特性

请参考支持的特性获取模型支持的特性矩阵。

请参考特性指南获取特性的配置方法。

3 前提条件

3.1 模型权重

建议将模型权重下载到多节点的共享目录中，例如 /root/.cache/

3.2 验证多节点通信（可选）

如果需要多节点部署，请按照验证多节点通信环境指南进行通信验证。

4 安装

4.1 Docker 镜像安装

您可以直接使用我们的官方 Docker 镜像来运行 GLM-5/5.1。

在每个节点上启动 Docker 镜像。

export IMAGE=quay.io/ascend/vllm-ascend:|vllm_ascend_version|-a3
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/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 /root/.cache:/root/.cache \
-it $IMAGE bash

在每个节点上启动 Docker 镜像。

export IMAGE=quay.io/ascend/vllm-ascend:|vllm_ascend_version|
docker run --rm \
    --name vllm-ascend \
    --shm-size=1g \
    --net=host \
    --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

如果要部署多节点环境，需要在每个节点上进行环境设置。

要验证环境是否安装成功，请参考安装。

4.2 源码安装

此外，如果您不想使用上述 Docker 镜像，也可以从源码构建所有内容：

  • 从源码安装 vllm-ascend，请参考安装。

如果要部署多节点环境，需要在每个节点上进行环境设置。

5 在线服务部署

5.1 单节点在线部署

  • 量化模型 glm-5-w4a8 和 glm-5.1-w4a8 可部署在 1 台 Atlas 800 A3（64G × 16）上。

运行以下脚本执行在线推理。

常见问题提示：如果遇到问题，请参考常见问题解答。

# The version of transformers needs to be upgraded to 5.2.0.
# pip install transformers==5.2.0 --upgrade

export HCCL_OP_EXPANSION_MODE="AIV"
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=1
export HCCL_BUFFSIZE=200
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export VLLM_ASCEND_BALANCE_SCHEDULING=1
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w4a8 \
--host 0.0.0.0 \
--port 8077 \
--data-parallel-size 1 \
--tensor-parallel-size 16 \
--enable-expert-parallel \
--seed 1024 \
--served-model-name glm-5 \
--max-num-seqs 8 \
--max-model-len 200000 \
--max-num-batched-tokens 4096 \
--trust-remote-code \
--gpu-memory-utilization 0.95 \
--quantization ascend \
--enable-chunked-prefill \
--enable-prefix-caching \
--additional-config '{"multistream_overlap_shared_expert": true}' \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
--speculative-config '{"num_speculative_tokens": 3, "method": "deepseek_mtp", "enforce_eager": true}'
  • 量化模型 glm-5-w8a8 和 glm-5.1-w8a8 可部署在 1 台 Atlas 800 A3（64G × 16）上。

运行以下脚本执行在线推理。

export HCCL_OP_EXPANSION_MODE="AIV"
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=1
export HCCL_BUFFSIZE=200
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export VLLM_ASCEND_BALANCE_SCHEDULING=1
export VLLM_ASCEND_ENABLE_MLAPO=1
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w8a8 \
--host 0.0.0.0 \
--port 8077 \
--data-parallel-size 1 \
--tensor-parallel-size 16 \
--enable-expert-parallel \
--seed 1024 \
--served-model-name glm-5 \
--max-num-seqs 8 \
--max-model-len 40960 \
--max-num-batched-tokens 4096 \
--trust-remote-code \
--gpu-memory-utilization 0.95 \
--quantization ascend \
--enable-chunked-prefill \
--enable-prefix-caching \
--additional-config '{"multistream_overlap_shared_expert": true}' \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
--speculative-config '{"num_speculative_tokens": 3, "method": "deepseek_mtp", "enforce_eager": true}'
  • 量化模型 glm-5-w4a8 可部署在 1 台 Atlas 800 A2（64G × 8）上。

运行以下脚本执行在线推理。

常见问题提示：如果遇到问题，请参考常见问题解答。

export HCCL_OP_EXPANSION_MODE="AIV"
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=1
export HCCL_BUFFSIZE=200
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export VLLM_ASCEND_BALANCE_SCHEDULING=1
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w4a8 \
--host 0.0.0.0 \
--port 8077 \
--data-parallel-size 1 \
--tensor-parallel-size 8 \
--enable-expert-parallel \
--seed 1024 \
--served-model-name glm-5 \
--max-num-seqs 2 \
--max-model-len 32768 \
--max-num-batched-tokens 4096 \
--trust-remote-code \
--gpu-memory-utilization 0.95 \
--quantization ascend \
--enable-chunked-prefill \
--enable-prefix-caching \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
--additional-config '{"multistream_overlap_shared_expert": true}' \
--speculative-config '{"num_speculative_tokens": 3, "method": "deepseek_mtp", "enforce_eager": true}'

注意： 参数说明如下：

  • 对于单节点部署，在低延迟场景下，建议使用 dp1tp16 并关闭专家并行。

5.2 多节点部署

如果要部署多节点环境，需要按照验证多节点通信环境进行多节点通信验证。

常见问题提示：如果遇到问题，请参考常见问题解答。

  • glm-5-bf16 和 glm-5.1-bf16：至少需要 2 台 Atlas 800 A3（64G × 16）。

分别在两个节点上运行以下脚本。

节点 0

# 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"

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=1
export HCCL_BUFFSIZE=200
export VLLM_ASCEND_BALANCE_SCHEDULING=1
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-bf16 \
--host 0.0.0.0 \
--port 8077 \
--data-parallel-size 2 \
--data-parallel-size-local 1 \
--data-parallel-address $node0_ip \
--data-parallel-rpc-port 12890 \
--tensor-parallel-size 16 \
--seed 1024 \
--served-model-name glm-5 \
--enable-expert-parallel \
--max-num-seqs 16 \
--max-model-len 8192 \
--max-num-batched-tokens 4096 \
--trust-remote-code \
--gpu-memory-utilization 0.95 \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
--speculative-config '{"num_speculative_tokens": 3, "method": "deepseek_mtp", "enforce_eager": true}'

节点 1

# 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"

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=1
export HCCL_BUFFSIZE=200
export VLLM_ASCEND_BALANCE_SCHEDULING=1
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-bf16 \
--host 0.0.0.0 \
--port 8077 \
--headless \
--data-parallel-size 2 \
--data-parallel-size-local 1 \
--data-parallel-start-rank 1 \
--data-parallel-address $node0_ip \
--data-parallel-rpc-port 12890 \
--tensor-parallel-size 16 \
--seed 1024 \
--served-model-name glm-5 \
--enable-expert-parallel \
--max-num-seqs 16 \
--max-model-len 8192 \
--max-num-batched-tokens 4096 \
--trust-remote-code \
--gpu-memory-utilization 0.95 \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
--speculative-config '{"num_speculative_tokens": 3, "method": "deepseek_mtp", "enforce_eager": true}'

分别在两个节点上运行以下脚本。

节点 0

# 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="xxx"

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=1
export HCCL_BUFFSIZE=200
export VLLM_ASCEND_BALANCE_SCHEDULING=1
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w4a8 \
--host 0.0.0.0 \
--port 8077 \
--data-parallel-size 2 \
--data-parallel-size-local 1 \
--data-parallel-address $node0_ip \
--data-parallel-rpc-port 13389 \
--tensor-parallel-size 8 \
--quantization ascend \
--seed 1024 \
--served-model-name glm-5 \
--enable-expert-parallel \
--max-num-seqs 2 \
--max-model-len 131072 \
--max-num-batched-tokens 4096 \
--trust-remote-code \
--gpu-memory-utilization 0.95 \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
--additional-config '{"multistream_overlap_shared_expert": true}' \
--speculative-config '{"num_speculative_tokens": 3, "method": "deepseek_mtp", "enforce_eager": true}'

节点 1

# 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="xxx"

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=1
export HCCL_BUFFSIZE=200
export VLLM_ASCEND_BALANCE_SCHEDULING=1
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w4a8 \
--host 0.0.0.0 \
--port 8077 \
--headless \
--data-parallel-size 2 \
--data-parallel-size-local 1 \
--data-parallel-start-rank 1 \
--data-parallel-address $node0_ip \
--data-parallel-rpc-port 13389 \
--tensor-parallel-size 8 \
--quantization ascend \
--seed 1024 \
--served-model-name glm-5 \
--enable-expert-parallel \
--max-num-seqs 2 \
--max-model-len 131072 \
--max-num-batched-tokens 4096 \
--trust-remote-code \
--gpu-memory-utilization 0.95 \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
--additional-config '{"multistream_overlap_shared_expert": true}' \
--speculative-config '{"num_speculative_tokens": 3, "method": "deepseek_mtp", "enforce_eager": true}'

python adjust_weight.py "path_of_bf16_weight"

# adjust_weight.py
from safetensors.torch import safe_open, save_file
import torch
import json
import os
import sys

target_keys = ["model.embed_tokens.weight", "lm_head.weight"]

def get_tensor_info(file_path):
   with safe_open(file_path, framework="pt", device="cpu") as f:
         tensor_names = f.keys()
         tensor_dict = {}
         for name in tensor_names:
            tensor = f.get_tensor(name)
            tensor_dict[name] = tensor
         return tensor_dict


if __name__ == "__main__":
   directory_path = sys.argv[1]
   json_name = "model.safetensors.index.json"
   json_path = os.path.join(directory_path, json_name)
   with open(json_path, 'r', encoding='utf-8') as f:
         json_data = json.load(f)
   weight_map = json_data.get('weight_map', {})
   file_list = []
   for key in target_keys:
         safetensor_file = weight_map.get(key)
         file_list.append(directory_path + safetensor_file)

   new_dict = {}
   for file_path in file_list:
         tensor_dict = get_tensor_info(file_path)
         for key in target_keys:
            if key in tensor_dict:
               if key == "model.embed_tokens.weight":
                     new_key = "model.layers.78.embed_tokens.weight"
               elif key == "lm_head.weight":
                     new_key = "model.layers.78.shared_head.head.weight"
               new_dict[new_key] = tensor_dict[key]

   new_file_name = os.path.join(directory_path, "mtp-others.safetensors")
   new_keys = ["model.layers.78.embed_tokens.weight", "model.layers.78.shared_head.head.weight"]
   save_file(tensors=new_dict, filename=new_file_name)
   for key in new_keys:
         json_data["weight_map"][key] = "mtp-others.safetensors"
   with open(json_path, 'w', encoding='utf-8') as f:
         json.dump(json_data, f, indent=2)
  • glm-5-w8a8：需要 2 台 Atlas 800 A3（64G × 16）。

分别在两个节点上运行以下脚本。

节点 0

# 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"

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=1
export HCCL_BUFFSIZE=200
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export VLLM_ASCEND_BALANCE_SCHEDULING=1
export VLLM_ASCEND_ENABLE_MLAPO=1
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w8a8 \
--host 0.0.0.0 \
--port 8077 \
--data-parallel-size 2 \
--data-parallel-size-local 1 \
--data-parallel-address $node0_ip \
--data-parallel-rpc-port 12890 \
--tensor-parallel-size 16 \
--seed 1024 \
--served-model-name glm-5 \
--enable-expert-parallel \
--max-num-seqs 16 \
--max-model-len 200000 \
--max-num-batched-tokens 4096 \
--trust-remote-code \
--gpu-memory-utilization 0.95 \
--quantization ascend \
--enable-chunked-prefill \
--enable-prefix-caching \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
--additional-config '{"multistream_overlap_shared_expert": true}' \
--speculative-config '{"num_speculative_tokens": 3, "method": "deepseek_mtp", "enforce_eager": true}'

节点 1

# 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"

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=1
export HCCL_BUFFSIZE=200
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export VLLM_ASCEND_BALANCE_SCHEDULING=1
export VLLM_ASCEND_ENABLE_MLAPO=1
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w8a8 \
--host 0.0.0.0 \
--port 8077 \
--headless \
--data-parallel-size 2 \
--data-parallel-size-local 1 \
--data-parallel-start-rank 1 \
--data-parallel-address $node0_ip \
--data-parallel-rpc-port 12890 \
--tensor-parallel-size 16 \
--seed 1024 \
--served-model-name glm-5 \
--enable-expert-parallel \
--max-num-seqs 16 \
--max-model-len 200000 \
--max-num-batched-tokens 4096 \
--trust-remote-code \
--gpu-memory-utilization 0.95 \
--quantization ascend \
--enable-chunked-prefill \
--enable-prefix-caching \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
--additional-config '{"multistream_overlap_shared_expert": true}' \
--speculative-config '{"num_speculative_tokens": 3, "method": "deepseek_mtp", "enforce_eager": true}'

5.3 预填充-解码分离

我们将在多节点环境中展示 GLM-5 的部署指南，采用 1P1D 配置以获得更佳性能。预填充-解码分离 指的是将预填充阶段和解码阶段分散到不同节点上，以提升吞吐量和降低延迟。

开始之前，请

  1. 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()

  2. 在每个节点上准备脚本 run_dp_template.sh。

    1. 预填充节点 0

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

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=1
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=256

export ASCEND_AGGREGATE_ENABLE=1
export ASCEND_TRANSPORT_PRINT=1
export ACL_OP_INIT_MODE=1
export ASCEND_A3_ENABLE=1
# Timeout (in seconds) for automatically releasing the prefiller’s KV cache for a particular request.
export VLLM_MOONCAKE_ABORT_REQUEST_TIMEOUT=480

export ASCEND_RT_VISIBLE_DEVICES=$1
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1

export VLLM_ASCEND_ENABLE_FUSED_MC2=1
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w8a8 \
    --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 \
    --speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp", "enforce_eager": true}' \
    --profiler-config \
    '{"profiler": "torch",
    "torch_profiler_dir": "./vllm_profile",
    "torch_profiler_with_stack": false}' \
    --seed 1024 \
    --served-model-name glm-5 \
    --max-model-len 131072 \
    --additional-config '{"multistream_overlap_shared_expert": true, "enable_dsa_cp": true}' \
    --max-num-batched-tokens 4096 \
    --trust-remote-code \
    --max-num-seqs 64 \
    --enable-chunked-prefill \
    --quantization ascend \
    --gpu-memory-utilization 0.95 \
    --enforce-eager \
    --enable-auto-tool-choice \
    --tool-call-parser glm47 \
    --reasoning-parser glm45 \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeConnectorV1",
    "kv_role": "kv_producer",
    "kv_port": "30000",
    "kv_connector_extra_config": {
                "use_ascend_direct": true,
                "prefill": {
                        "dp_size": 2,
                        "tp_size": 16
                },
                "decode": {
                        "dp_size": 16,
                        "tp_size": 4
                }
        }
    }'

    2. 预填充节点 1

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

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=1
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=256

export ASCEND_AGGREGATE_ENABLE=1
export ASCEND_TRANSPORT_PRINT=1
export ACL_OP_INIT_MODE=1
export ASCEND_A3_ENABLE=1
# Timeout (in seconds) for automatically releasing the prefiller’s KV cache for a particular request.
export VLLM_MOONCAKE_ABORT_REQUEST_TIMEOUT=480

export ASCEND_RT_VISIBLE_DEVICES=$1
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True

export VLLM_ASCEND_ENABLE_FLASHCOMM1=1

export VLLM_ASCEND_ENABLE_FUSED_MC2=1
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w8a8 \
    --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 \
    --speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp", "enforce_eager": true}' \
    --profiler-config \
    '{"profiler": "torch",
    "torch_profiler_dir": "./vllm_profile",
    "torch_profiler_with_stack": false}' \
    --seed 1024 \
    --served-model-name glm-5 \
    --max-model-len 131072 \
    --additional-config '{"multistream_overlap_shared_expert": true, "enable_dsa_cp": true}' \
    --max-num-batched-tokens 4096 \
    --trust-remote-code \
    --max-num-seqs 64 \
    --enable-chunked-prefill \
    --gpu-memory-utilization 0.95 \
    --quantization ascend \
    --enforce-eager \
    --enable-auto-tool-choice \
    --tool-call-parser glm47 \
    --reasoning-parser glm45 \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeConnectorV1",
    "kv_role": "kv_producer",
    "kv_port": "30000",
    "kv_connector_extra_config": {
                "use_ascend_direct": true,
                "prefill": {
                        "dp_size": 2,
                        "tp_size": 16
                },
                "decode": {
                        "dp_size": 16,
                        "tp_size": 4
                }
        }
    }'

    3. 解码节点 0

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

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

#Mooncake
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=1

export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=256


export ASCEND_AGGREGATE_ENABLE=1
export ASCEND_TRANSPORT_PRINT=1
export ACL_OP_INIT_MODE=1
export ASCEND_A3_ENABLE=1
# Timeout (in seconds) for automatically releasing the prefiller’s KV cache for a particular request.
export VLLM_MOONCAKE_ABORT_REQUEST_TIMEOUT=480

export TASK_QUEUE_ENABLE=1

export ASCEND_RT_VISIBLE_DEVICES=$1

export VLLM_ASCEND_ENABLE_FUSED_MC2=1
export VLLM_ASCEND_ENABLE_MLAPO=1
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w8a8 \
    --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 \
    --speculative-config '{"num_speculative_tokens": 3,  "method":"deepseek_mtp", "enforce_eager": true}' \
    --profiler-config \
    '{"profiler": "torch",
    "torch_profiler_dir": "./vllm_profile",
    "torch_profiler_with_stack": false}' \
    --seed 1024 \
    --served-model-name glm-5 \
    --max-model-len 200000 \
    --max-num-batched-tokens 32 \
    --compilation-config '{"cudagraph_mode":"FULL_DECODE_ONLY"}' \
    --additional-config '{"multistream_overlap_shared_expert": true, "recompute_scheduler_enable": true}' \
    --trust-remote-code \
    --max-num-seqs 8 \
    --gpu-memory-utilization 0.92 \
    --quantization ascend \
    --enable-auto-tool-choice \
    --tool-call-parser glm47 \
    --reasoning-parser glm45 \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeConnectorV1",
    "kv_role": "kv_consumer",
    "kv_port": "30100",
    "kv_connector_extra_config": {
                "use_ascend_direct": true,
                "prefill": {
                        "dp_size": 2,
                        "tp_size": 16
                },
                "decode": {
                        "dp_size": 16,
                        "tp_size": 4
                }
        }
    }'

    4. 解码节点 1

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

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

#Mooncake
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=1

export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=256

export ASCEND_AGGREGATE_ENABLE=1
export ASCEND_TRANSPORT_PRINT=1
export ACL_OP_INIT_MODE=1
export ASCEND_A3_ENABLE=1
# Timeout (in seconds) for automatically releasing the prefiller’s KV cache for a particular request.
export VLLM_MOONCAKE_ABORT_REQUEST_TIMEOUT=480

export TASK_QUEUE_ENABLE=1

export ASCEND_RT_VISIBLE_DEVICES=$1

export VLLM_ASCEND_ENABLE_FUSED_MC2=1
export VLLM_ASCEND_ENABLE_MLAPO=1
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w8a8 \
    --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 \
    --speculative-config '{"num_speculative_tokens": 3,  "method":"deepseek_mtp", "enforce_eager": true}' \
    --profiler-config \
    '{"profiler": "torch",
    "torch_profiler_dir": "./vllm_profile",
    "torch_profiler_with_stack": false}' \
    --seed 1024 \
    --served-model-name glm-5 \
    --max-model-len 200000 \
    --max-num-batched-tokens 32 \
    --compilation-config '{"cudagraph_mode":"FULL_DECODE_ONLY"}' \
    --additional-config '{"multistream_overlap_shared_expert": true, "recompute_scheduler_enable": true}' \
    --trust-remote-code \
    --max-num-seqs 8 \
    --gpu-memory-utilization 0.92 \
    --quantization ascend \
    --enable-auto-tool-choice \
    --tool-call-parser glm47 \
    --reasoning-parser glm45 \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeConnectorV1",
    "kv_role": "kv_consumer",
    "kv_port": "30100",
    "kv_connector_extra_config": {
                "use_ascend_direct": true,
                "prefill": {
                        "dp_size": 2,
                        "tp_size": 16
                },
                "decode": {
                        "dp_size": 16,
                        "tp_size": 4
                }
        }
    }'

    5. 解码节点 2

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

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

#Mooncake
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=1

export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=256

export ASCEND_AGGREGATE_ENABLE=1
export ASCEND_TRANSPORT_PRINT=1
export ACL_OP_INIT_MODE=1
export ASCEND_A3_ENABLE=1
# Timeout (in seconds) for automatically releasing the prefiller’s KV cache for a particular request.
export VLLM_MOONCAKE_ABORT_REQUEST_TIMEOUT=480

export TASK_QUEUE_ENABLE=1

export ASCEND_RT_VISIBLE_DEVICES=$1

export VLLM_ASCEND_ENABLE_FUSED_MC2=1
export VLLM_ASCEND_ENABLE_MLAPO=1
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w8a8 \
    --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 \
    --speculative-config '{"num_speculative_tokens": 3,  "method":"deepseek_mtp", "enforce_eager": true}' \
    --profiler-config \
    '{"profiler": "torch",
    "torch_profiler_dir": "./vllm_profile",
    "torch_profiler_with_stack": false}' \
    --seed 1024 \
    --served-model-name glm-5 \
    --max-model-len 200000 \
    --max-num-batched-tokens 32 \
    --compilation-config '{"cudagraph_mode":"FULL_DECODE_ONLY"}' \
    --additional-config '{"multistream_overlap_shared_expert": true, "recompute_scheduler_enable": true}' \
    --trust-remote-code \
    --max-num-seqs 8 \
    --gpu-memory-utilization 0.92 \
    --quantization ascend \
    --enable-auto-tool-choice \
    --tool-call-parser glm47 \
    --reasoning-parser glm45 \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeConnectorV1",
    "kv_role": "kv_consumer",
    "kv_port": "30100",
    "kv_connector_extra_config": {
                "use_ascend_direct": true,
                "prefill": {
                        "dp_size": 2,
                        "tp_size": 16
                },
                "decode": {
                        "dp_size": 16,
                        "tp_size": 4
                }
        }
    }'

    6. 解码节点 3

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

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

#Mooncake
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=1

export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=256

export ASCEND_AGGREGATE_ENABLE=1
export ASCEND_TRANSPORT_PRINT=1
export ACL_OP_INIT_MODE=1
export ASCEND_A3_ENABLE=1
# Timeout (in seconds) for automatically releasing the prefiller’s KV cache for a particular request.
export VLLM_MOONCAKE_ABORT_REQUEST_TIMEOUT=480

export TASK_QUEUE_ENABLE=1

export ASCEND_RT_VISIBLE_DEVICES=$1

export VLLM_ASCEND_ENABLE_FUSED_MC2=1
export VLLM_ASCEND_ENABLE_MLAPO=1
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/GLM5-w8a8 \
    --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 \
    --speculative-config '{"num_speculative_tokens": 3,  "method":"deepseek_mtp", "enforce_eager": true}' \
    --profiler-config \
    '{"profiler": "torch",
    "torch_profiler_dir": "./vllm_profile",
    "torch_profiler_with_stack": false}' \
    --seed 1024 \
    --served-model-name glm-5 \
    --max-model-len 200000 \
    --max-num-batched-tokens 32 \
    --compilation-config '{"cudagraph_mode":"FULL_DECODE_ONLY"}' \
    --additional-config '{"multistream_overlap_shared_expert": true, "recompute_scheduler_enable": true}' \
    --trust-remote-code \
    --max-num-seqs 8 \
    --gpu-memory-utilization 0.92 \
    --quantization ascend \
    --enable-auto-tool-choice \
    --tool-call-parser glm47 \
    --reasoning-parser glm45 \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeConnectorV1",
    "kv_role": "kv_consumer",
    "kv_port": "30100",
    "kv_connector_extra_config": {
                "use_ascend_direct": true,
                "prefill": {
                        "dp_size": 2,
                        "tp_size": 16
                },
                "decode": {
                        "dp_size": 16,
                        "tp_size": 4
                }
        }
    }'

准备工作完成后，可以在每个节点上使用以下命令启动服务器：

  1. 预填充节点 0

    # change ip to your own
python launch_online_dp.py --dp-size 2 --tp-size 16 --dp-size-local 1 --dp-rank-start 0 --dp-address $node_p0_ip --dp-rpc-port 10521 --vllm-start-port 6700

  2. 预填充节点 1

    # change ip to your own
python launch_online_dp.py --dp-size 2 --tp-size 16 --dp-size-local 1 --dp-rank-start 1 --dp-address $node_p0_ip --dp-rpc-port 10521 --vllm-start-port 6700

  3. 解码节点 0

    # change ip to your own
python launch_online_dp.py --dp-size 16 --tp-size 4 --dp-size-local 4 --dp-rank-start 0 --dp-address $node_d0_ip --dp-rpc-port 10523 --vllm-start-port 6721

  4. 解码节点 1

    # change ip to your own
python launch_online_dp.py --dp-size 16 --tp-size 4 --dp-size-local 4 --dp-rank-start 4 --dp-address $node_d0_ip --dp-rpc-port 10523 --vllm-start-port 6721

  5. 解码节点 2

    # change ip to your own
python launch_online_dp.py --dp-size 16 --tp-size 4 --dp-size-local 4 --dp-rank-start 8 --dp-address $node_d0_ip --dp-rpc-port 10523 --vllm-start-port 6721

  6. 解码节点 3

    # change ip to your own
python launch_online_dp.py --dp-size 16 --tp-size 4 --dp-size-local 4 --dp-rank-start 12 --dp-address $node_d0_ip --dp-rpc-port 10523 --vllm-start-port 6721

5.4 请求转发

要设置请求转发，请在任意机器上运行以下脚本。您可以在仓库的示例中获取代理程序：load_balance_proxy_server_example.py

unset http_proxy
unset https_proxy

python load_balance_proxy_server_example.py \
    --port 8000 \
    --host 0.0.0.0 \
    --prefiller-hosts \
       $node_p0_ip \
       $node_p1_ip \
    --prefiller-ports \
       6700 \
       6700 \
    --decoder-hosts \
      $node_d0_ip \
      $node_d0_ip \
      $node_d0_ip \
      $node_d0_ip \
      $node_d1_ip \
      $node_d1_ip \
      $node_d1_ip \
      $node_d1_ip \
      $node_d2_ip \
      $node_d2_ip \
      $node_d2_ip \
      $node_d2_ip \
      $node_d3_ip \
      $node_d3_ip \
      $node_d3_ip \
      $node_d3_ip \
    --decoder-ports \
      6721 6722 6723 6724 \
      6721 6722 6723 6724 \
      6721 6722 6723 6724 \
      6721 6722 6723 6724

注意：

以下是一些用于优化的配置：

  • VLLM_ASCEND_ENABLE_FLASHCOMM1：启用 FlashComm 优化，以减少预填充节点上的通信和计算开销。启用 FlashComm 后，layer_sharding 列表不能包含 o_proj 作为元素。
  • VLLM_ASCEND_ENABLE_FUSED_MC2：启用 dispatch_ffn_combine 融合算子。
  • VLLM_ASCEND_ENABLE_MLAPO：启用融合算子 MlaPreprocessOperation。

有关上述环境变量的进一步说明和限制，请参考以下 Python 文件：envs.py

6 功能验证

服务器启动后，您可以使用输入提示来查询模型：

curl http://<node0_ip>:<port>/v1/completions \
    -H "Content-Type: application/json" \
    -d '{
        "model": "glm-5",
        "prompt": "The future of AI is",
        "max_completion_tokens": 15,
        "temperature": 0
    }'

预期结果：

{"id": "chatcmlib-bc44ad093dec79a2", "object": "chat.completion", "created": "1770903266", "model": "glm-5", "choices": [{ "index": 0, "message": {"role": "assistant", "content": "The future of AI is not one thing, but a convergence of several powerful trends.", "annotations": "null", "audio": "null", "function_call": "null", "tool_calls": [], "reasoning": "null"}, "logprobs": "null", "finish_reason": "length", "stop_reason": "null", "token_ids": null}], "service_tier": "null", "system fingerprint": "null", "usage": {"prompt_tokens": 5, "total_tokens": 20, "completion_tokens": 15, "prompt_tokens_details": null}, "prompt_logprobs": "null", "prompt_token_ids": "null", "kv_transfer_params": null}

7 精度评估

7.1 使用 AISBench

  1. 有关详细信息，请参阅 使用 AISBench。

  2. 执行后，您可以获取结果。

8 性能评估

8.1 使用 AISBench

有关详细信息，请参阅 使用 AISBench 进行性能评估。

8.2 使用 vLLM Benchmark

有关更多详细信息，请参阅 vllm benchmark。

9 性能调优

9.1 推荐配置

注意：以下配置在特定测试环境中经过验证，仅供参考。最佳配置取决于最大输入/输出长度、前缀缓存命中率、精度要求和部署机器比例等因素。建议参考第 9.2 节根据实际情况进行调优。

表 1：场景概览

场景 部署模式 *NPU总数 权重版本 关键考量
高吞吐量 1P1D部署 32 (A3) GLM5-w8a8/GLM5.1-w8a8 在P节点上使用dp4 tp8，在D节点上使用dp8 dp4，以平衡延迟和吞吐量
低延迟 1P1D部署 32 (A3) GLM5-w8a8/GLM5.1-w8a8 在P和D节点上均使用dp4 tp8，以降低延迟

*Total NPUs 表示所有节点上使用的 NPU 总数。

表 2：详细节点配置

场景 配置 NPU数量 TP DP 最大序列数 最大批处理Token数 最大模型长度 MTP推测数量
High Throughput (A3) 1P1D deployment 32 P:8 D:4 P:4 D:8 P:64 D:128 P:4096 D:32 P:133120 D:150000 3
Low Latency (A3) 1P1D deployment 32 4 8 P:64 D:128 P:4096 D:32 P:133120 D:150000 3

10 常见问题解答

  • 常见问题提示：如果遇到问题，请参考 常见问题解答。

  • 问：如何解决 ValueError: Tokenizer class TokenizersBackend does not exist or is not currently imported？

答：请将 transformers 的版本更新到 5.2.0

  • 问：如何为 GLM-5 启用函数调用？

答：请在 vLLM 启动命令中添加以下配置

--tool-call-parser glm47 \
--reasoning-parser glm45 \
--enable-auto-tool-choice \