MetaX-Tech Developer Forum 论坛首页
  • 沐曦开发者
search
Sign in

e411

  • Members
  • Joined 2026年5月8日
  • message 帖子
  • forum 主题
  • favorite 关注者
  • favorite_border Follows
  • person_outline 详细信息

e411 has posted 15 messages.

  • See post chevron_right
    e411
    Members
    6月30日发布的vllm 0.21.0是否支持deepseek v4部署?如果支持,正确的启动命令是? 解决中 2026年7月2日 10:52

    你们现在有比较新的dsv4的验证镜像吗,我这边真的很想部署一个dsv4来用(

  • See post chevron_right
    e411
    Members
    6月30日发布的vllm 0.21.0是否支持deepseek v4部署?如果支持,正确的启动命令是? 解决中 2026年7月2日 10:42

    注意到vllm镜像里的tilelang没有maca适配代码

  • See post chevron_right
    e411
    Members
    6月30日发布的vllm 0.21.0是否支持deepseek v4部署?如果支持,正确的启动命令是? 解决中 2026年7月2日 10:40

    我进一步排查后,发现其中的tilelang有问题,我对比了贵公司官网发布的tilelang安装包与vllm镜像内的tilelang,发现vllm镜像内tilelang是上游版本,而不是贵公司适配过的tilelang maca版本,其中不含任何maca代码,只有cuda,vllm调用到了cuda,导致maca工具链编译报错,请联系贵公司研发重新适配打包这个vllm的镜像,谢谢!

  • See post chevron_right
    e411
    Members
    6月30日发布的vllm 0.21.0是否支持deepseek v4部署?如果支持,正确的启动命令是? 解决中 2026年7月2日 10:35

    没有变化,日志仍然是之前一样。

  • See post chevron_right
    e411
    Members
    6月30日发布的vllm 0.21.0是否支持deepseek v4部署?如果支持,正确的启动命令是? 解决中 2026年7月1日 17:18

    source /etc/profile.d/conda.sh 是为了加载vllm所在的虚拟环境,因为docker run -d拉起来的bash是非交互的,因此不会自动加载conda环境,为了正确加载后续vllm命令,必须加载一下conda.sh。

  • See post chevron_right
    e411
    Members
    6月30日发布的vllm 0.21.0是否支持deepseek v4部署?如果支持,正确的启动命令是? 解决中 2026年7月1日 16:52
    [Wed Jul  1 22:58:53 2026] AMD-Vi: Disabling interrupt remapping
    [Wed Jul  1 22:58:54 2026] ACPI: Using IOAPIC for interrupt routing
    [Wed Jul  1 22:58:54 2026] ACPI: PCI: Interrupt link LNKA configured for IRQ 0
    [Wed Jul  1 22:58:54 2026] ACPI: PCI: Interrupt link LNKB configured for IRQ 0
    [Wed Jul  1 22:58:54 2026] ACPI: PCI: Interrupt link LNKC configured for IRQ 0
    [Wed Jul  1 22:58:54 2026] ACPI: PCI: Interrupt link LNKD configured for IRQ 0
    [Wed Jul  1 22:58:54 2026] ACPI: PCI: Interrupt link LNKE configured for IRQ 0
    [Wed Jul  1 22:58:54 2026] ACPI: PCI: Interrupt link LNKF configured for IRQ 0
    [Wed Jul  1 22:58:54 2026] ACPI: PCI: Interrupt link LNKG configured for IRQ 0
    [Wed Jul  1 22:58:54 2026] ACPI: PCI: Interrupt link LNKH configured for IRQ 0
    [Wed Jul  1 22:58:57 2026] RAS: Correctable Errors collector initialized.
    [Wed Jul  1 22:59:12 2026] hrtimer: interrupt took 30958313 ns
    

    我认为这和卡没有任何关系,是vllm metax内部的一些软件错误

  • See post chevron_right
    e411
    Members
    6月30日发布的vllm 0.21.0是否支持deepseek v4部署?如果支持,正确的启动命令是? 解决中 2026年7月1日 16:02

    一、软硬件信息
    1.服务器厂家: 超云
    2.沐曦GPU型号: C500
    3.操作系统内核版本: 6.8.0-124-generic
    4.是否开启CPU虚拟化:是
    5.mx-smi回显:

    mx-smi  version: 2.3.4
    
    =================== MetaX System Management Interface Log ===================
    Timestamp                                         : Wed Jul  1 23:56:36 2026
    
    Attached GPUs                                     : 8
    +---------------------------------------------------------------------------------+
    | MX-SMI 2.3.4                       Kernel Mode Driver Version: 3.9.10           |
    | MACA Version: 3.7.0.38             BIOS Version: 1.35.2.0                       |
    |------------------+-----------------+---------------------+----------------------|
    | Board       Name | GPU   Persist-M | Bus-id              | GPU-Util      sGPU-M |
    | Pwr:Usage/Cap    | Temp       Perf | Memory-Usage        | GPU-State            |
    |==================+=================+=====================+======================|
    | 0     MetaX C500 | 0           Off | 0000:04:00.0        | 0%          Disabled |
    | 38W / 350W       | 40C          P0 | 860/65536 MiB       | Available            |
    +------------------+-----------------+---------------------+----------------------+
    | 1     MetaX C500 | 1           Off | 0000:05:00.0        | 0%          Disabled |
    | 39W / 350W       | 41C          P0 | 860/65536 MiB       | Available            |
    +------------------+-----------------+---------------------+----------------------+
    | 2     MetaX C500 | 2           Off | 0000:06:00.0        | 0%          Disabled |
    | 42W / 350W       | 43C          P0 | 860/65536 MiB       | Available            |
    +------------------+-----------------+---------------------+----------------------+
    | 3     MetaX C500 | 3           Off | 0000:07:00.0        | 0%          Disabled |
    | 38W / 350W       | 41C          P0 | 860/65536 MiB       | Available            |
    +------------------+-----------------+---------------------+----------------------+
    | 4     MetaX C500 | 4           Off | 0000:0b:00.0        | 0%          Disabled |
    | 40W / 350W       | 42C          P0 | 860/65536 MiB       | Available            |
    +------------------+-----------------+---------------------+----------------------+
    | 5     MetaX C500 | 5           Off | 0000:0c:00.0        | 0%          Disabled |
    | 41W / 350W       | 43C          P0 | 860/65536 MiB       | Available            |
    +------------------+-----------------+---------------------+----------------------+
    | 6     MetaX C500 | 6           Off | 0000:0d:00.0        | 0%          Disabled |
    | 42W / 350W       | 41C          P0 | 860/65536 MiB       | Available            |
    +------------------+-----------------+---------------------+----------------------+
    | 7     MetaX C500 | 7           Off | 0000:0e:00.0        | 0%          Disabled |
    | 39W / 350W       | 40C          P0 | 860/65536 MiB       | Available            |
    +------------------+-----------------+---------------------+----------------------+
    
    +---------------------------------------------------------------------------------+
    | Process:                                                                        |
    |  GPU                    PID         Process Name                 GPU Memory     |
    |                                                                  Usage(MiB)     |
    |=================================================================================|
    |  no process found                                                               |
    +---------------------------------------------------------------------------------+
    
    End of Log
    

    6.docker info回显:

    Client: Docker Engine - Community
     Version:    29.3.0
     Context:    default
     Debug Mode: false
     Plugins:
      buildx: Docker Buildx (Docker Inc.)
        Version:  v0.31.1
        Path:     /usr/libexec/docker/cli-plugins/docker-buildx
      compose: Docker Compose (Docker Inc.)
        Version:  v5.1.0
        Path:     /usr/libexec/docker/cli-plugins/docker-compose
      model: Docker Model Runner (Docker Inc.)
        Version:  v1.1.8
        Path:     /usr/libexec/docker/cli-plugins/docker-model
    
    Server:
     Containers: 12
      Running: 4
      Paused: 0
      Stopped: 8
     Images: 15
     Server Version: 29.3.0
     Storage Driver: overlayfs
      driver-type: io.containerd.snapshotter.v1
     Logging Driver: json-file
     Cgroup Driver: systemd
     Cgroup Version: 2
     Plugins:
      Volume: local
      Network: bridge host ipvlan macvlan null overlay
      Log: awslogs fluentd gcplogs gelf journald json-file local splunk syslog
     CDI spec directories:
      /etc/cdi
      /var/run/cdi
     Swarm: inactive
     Runtimes: io.containerd.runc.v2 runc
     Default Runtime: runc
     Init Binary: docker-init
     containerd version: dea7da592f5d1d2b7755e3a161be07f43fad8f75
     runc version: v1.3.4-0-gd6d73eb8
     init version: de40ad0
     Security Options:
      apparmor
      seccomp
       Profile: builtin
      cgroupns
     Kernel Version: 6.8.0-124-generic
     Operating System: Ubuntu-Server 24.04.4 LTS (Noble Numbat)
     OSType: linux
     Architecture: x86_64
     CPUs: 192
     Total Memory: 503.5GiB
     Name: server2
     ID: 404bb473-c286-41a5-a18e-b94481c6f73e
     Docker Root Dir: /var/lib/docker
     Debug Mode: false
     Experimental: false
     Insecure Registries:
      ::1/128
      127.0.0.0/8
     Live Restore Enabled: false
     Firewall Backend: iptables
    

    7.镜像版本:cr.metax-tech.com/public-ai-release/maca/vllm-metax:0.21.0-maca.ai3.7.1.106-torch2.8-py312-ubuntu22.04-amd64
    8.启动容器命令:

    docker run -d \
        --device=/dev/dri \
        --device=/dev/mxcd \
        --device=/dev/infiniband \
        --group-add video \
        --security-opt seccomp=unconfined \
        --security-opt apparmor=unconfined \
        --shm-size 100gb \
        --ulimit memlock=-1 \
        --privileged=true \
        --network host \
        -v /mnt/modelscope:/llm_models \
        --name vllm-deepseek-v4-flash \
        cr.metax-tech.com/public-ai-release/maca/vllm-metax:0.21.0-maca.ai3.7.1.106-torch2.8-py312-ubuntu22.04-amd64 bash -c "source /etc/profile.d/conda.sh && source /llm_models/run-deepseek-v4-entry.sh"
    

    9.容器内执行命令(/llm_models/run-deepseek-v4-entry.sh的内容):

    #!/bin/bash
    export MACA_SMALL_PAGESIZE_ENABLE=1
    export MACA_VLLM_ENABLE_MCTLASS_FUSED_MOE=0
    export MACA_VLLM_ENABLE_MCTLASS_PYTHON_API=1
    
    vllm serve /llm_models/DeepSeek-V4-Flash-FlexSMQ-AWQ-W8A8 --served-model-name DeepSeek-V4-Flash --trust-remote-code \
    --kv-cache-dtype bfloat16 --block-size 256 --port 8001 --max-model-len 1048576 --gpu-memory-utilization 0.9 \
    --tokenizer-mode deepseek_v4 --tool-call-parser deepseek_v4 \
    --enable-auto-tool-choice --reasoning-parser deepseek_v4 --default-chat-template-kwargs '{"enable_thinking": true}' \
    --compilation-config '{"cudagraph_mode":"FULL_AND_PIECEWISE", "custom_ops":["all"]}' \
    --max-num-seq 32 -tp 8 --speculative_config '{"method": "mtp", "num_speculative_tokens": 1}'
    

    二、问题现象
    请描述详细的问题现象日志。若日志过长,请上传附件(txt格式)。

  • See post chevron_right
    e411
    Members
    6月30日发布的vllm 0.21.0是否支持deepseek v4部署?如果支持,正确的启动命令是? 解决中 2026年7月1日 15:53

    硬件 C500 * 8

    我使用的命令是:

    #!/bin/bash
    export MACA_SMALL_PAGESIZE_ENABLE=1
    export MACA_VLLM_ENABLE_MCTLASS_FUSED_MOE=0
    export MACA_VLLM_ENABLE_MCTLASS_PYTHON_API=1
    
    vllm serve /llm_models/DeepSeek-V4-Flash-FlexSMQ-AWQ-W8A8 --served-model-name DeepSeek-V4-Flash --trust-remote-code \
    --kv-cache-dtype bfloat16 --block-size 256 --port 8001 --max-model-len 1048576 --gpu-memory-utilization 0.9 \
    --tokenizer-mode deepseek_v4 --tool-call-parser deepseek_v4 \
    --enable-auto-tool-choice --reasoning-parser deepseek_v4 --default-chat-template-kwargs '{"enable_thinking": true}' \
    --compilation-config '{"cudagraph_mode":"FULL_AND_PIECEWISE", "custom_ops":["all"]}' \
    --max-num-seq 32 -tp 8 --speculative_config '{"method": "mtp", "num_speculative_tokens": 1}'
    

    然后在加载完模型之后产生巨量的tilelang的编译报错,然后失败退出

  • See post chevron_right
    e411
    Members
    metax-driver-3.8.0.10-deb-x86_64.run 最新驱动存在bug,DKMS无法正确编译 已解决 2026年7月1日 15:18
    sudo sh metax-driver-3.8.0.10-deb-x86_64.run
    [sudo] password for cast503:
    Verifying archive integrity...  100%   MD5 checksums are OK. All good.
    Uncompressing MetaX driver installer  100%
    uninstalling metax-linux-dkms
    (Reading database ... 142019 files and directories currently installed.)
    Removing metax-linux-dkms (3.8.30) ...
    Module metax-linux-3.8.30 for kernel 6.8.0-110-generic (x86_64).
    Before uninstall, this module version was ACTIVE on this kernel.
    
    metax.ko.zst:
     - Uninstallation
       - Deleting from: /lib/modules/6.8.0-110-generic/updates/dkms/
     - Original module
       - No original module was found for this module on this kernel.
       - Use the dkms install command to reinstall any previous module version.
    
    Running the post_remove script:
    depmod....
    Module metax-linux-3.8.30 for kernel 6.8.0-124-generic (x86_64).
    Before uninstall, this module version was ACTIVE on this kernel.
    
    metax.ko.zst:
     - Uninstallation
       - Deleting from: /lib/modules/6.8.0-124-generic/updates/dkms/
     - Original module
       - No original module was found for this module on this kernel.
       - Use the dkms install command to reinstall any previous module version.
    
    Running the post_remove script:
    depmod....
    Deleting module metax-linux-3.8.30 completely from the DKMS tree.
    uninstalling mxsmt
    (Reading database ... 141807 files and directories currently installed.)
    Removing mxsmt (3.7.2.30) ...
    mxsmt package uninstalled successfully
    Purging configuration files for mxsmt (3.7.2.30) ...
    mxsmt package uninstalled successfully
    dpkg: warning: while removing mxsmt, directory '/opt/mxdriver' not empty so not removed
    uninstalling mxfw
    (Reading database ... 141795 files and directories currently installed.)
    Removing mxfw (3.7.2) ...
    mxfw package uninstalled successfully
    Purging configuration files for mxfw (3.7.2) ...
    mxfw package uninstalled successfully
    
    deb/mxfw_*.deb
    deb/mxsmt_*.deb
    deb/metax-linux[-_]*.deb
    installing packages under /opt/mxdriver
    installing deb/mxfw_3.8.0-1.all.deb
    Selecting previously unselected package mxfw.
    (Reading database ... 141709 files and directories currently installed.)
    Preparing to unpack deb/mxfw_3.8.0-1.all.deb ...
    Unpacking mxfw (3.8.0) ...
    Setting up mxfw (3.8.0) ...
    mxfw package installed successfully
    installing deb/mxsmt_3.8.0.10.amd64.deb
    Selecting previously unselected package mxsmt.
    (Reading database ... 141795 files and directories currently installed.)
    Preparing to unpack deb/mxsmt_3.8.0.10.amd64.deb ...
    Unpacking mxsmt (3.8.0.10) ...
    Setting up mxsmt (3.8.0.10) ...
    mxsmt package installed successfully
    Processing triggers for libc-bin (2.39-0ubuntu8.7) ...
    installing deb/metax-linux_3.9.10-1_amd64.deb
    Selecting previously unselected package metax-linux.
    (Reading database ... 141807 files and directories currently installed.)
    Preparing to unpack .../metax-linux_3.9.10-1_amd64.deb ...
    Unpacking metax-linux (3.9.10) ...
    Setting up metax-linux (3.9.10) ...
    
    Notice: The current kernel version (6.8.0-124-generic) is not supported.
    Please upgrade your kernel to one of the following versions:
    5.10.0-0.deb10.28-amd64
    5.10.0-29-amd64
    5.10.0-30-amd64
    5.10.200-4-velinux1u1-amd64
    5.15.0-100-generic
    5.15.0-105-generic
    5.15.0-112-generic
    5.15.0-119-generic
    5.15.0-124-generic
    5.15.0-130-generic
    5.15.0-138-generic
    5.15.0-25-generic
    5.15.0-46-generic
    5.15.0-58-generic
    5.15.0-60-generic
    5.15.0-72-generic
    5.15.0-78-generic
    5.15.0-88-generic
    5.15.119-rt65-home-ear1.1
    5.19.0-46-generic
    5.19.17
    5.4.0-131-generic
    5.4.0-144-generic
    5.4.0-148-generic
    5.4.0-176-generic
    5.4.0-26-generic
    5.4.0-42-generic
    5.4.18-142-generic
    6.1.141-1-generic
    6.1.152-1-generic
    6.12.34-1-generic
    6.12.47-1-generic
    6.14.0-1003-oem
    6.14.0-27-generic
    6.8.0-31-generic
    6.8.0-49-generic
    6.8.0-53-generic
    And then reinstall the metax-linux package.
    
    dpkg: error processing package metax-linux (--install):
     installed metax-linux package post-installation script subprocess returned error exit status 1
    Errors were encountered while processing:
     metax-linux
    uninstalling metax-linux
    (Reading database ... 141924 files and directories currently installed.)
    Removing metax-linux (3.9.10) ...
    metax-linux package uninstalled successfully
    Purging configuration files for metax-linux (3.9.10) ...
    metax-linux package uninstalled successfully
    Reading package lists... Done
    Building dependency tree... Done
    Reading state information... Done
    dkms is already the newest version (3.0.11-1ubuntu13).
    autoconf is already the newest version (2.71-3).
    automake is already the newest version (1:1.16.5-1.3ubuntu1).
    The following packages were automatically installed and are no longer required:
      libdebuginfod-common libdebuginfod1t64 libfwupd2 libgusb2 libmspack0t64 libxmlsec1t64 libxmlsec1t64-openssl
      libxslt1.1 zerofree
    Use 'sudo apt autoremove' to remove them.
    0 upgraded, 0 newly installed, 0 to remove and 0 not upgraded.
    Selecting previously unselected package metax-linux-dkms.
    (Reading database ... 141807 files and directories currently installed.)
    Preparing to unpack .../metax-linux-dkms_3.9.10-1_amd64.deb ...
    Unpacking metax-linux-dkms (3.9.10) ...
    Setting up metax-linux-dkms (3.9.10) ...
    Loading new metax-linux-3.9.10 DKMS files...
    Building for 6.8.0-124-generic
    Building for architecture x86_64
    Building initial module for 6.8.0-124-generic
    autoreconf: export WARNINGS=
    autoreconf: Entering directory '.'
    autoreconf: configure.ac: not using Gettext
    autoreconf: running: aclocal --force
    autoreconf: configure.ac: tracing
    autoreconf: configure.ac: not using Libtool
    autoreconf: configure.ac: not using Intltool
    autoreconf: configure.ac: not using Gtkdoc
    autoreconf: running: /usr/bin/autoconf --force
    configure.ac:23: warning: The macro `AC_HELP_STRING' is obsolete.
    configure.ac:23: You should run autoupdate.
    ./lib/autoconf/general.m4:204: AC_HELP_STRING is expanded from...
    ./lib/autoconf/general.m4:1534: AC_ARG_ENABLE is expanded from...
    m4/config.m4:14: AC_METAX_CONFIG is expanded from...
    configure.ac:23: the top level
    autoreconf: running: /usr/bin/autoheader --force
    autoreconf: configure.ac: not using Automake
    autoreconf: 'config/install-sh' is updated
    autoreconf: Leaving directory '.'
    ERROR (dkms apport): unable to determine source package for metax-linux-dkms
    Error! Bad return status for module build on kernel: 6.8.0-124-generic (x86_64)
    Consult /var/lib/dkms/metax-linux/3.9.10/build/make.log for more information.
    dpkg: error processing package metax-linux-dkms (--install):
     installed metax-linux-dkms package post-installation script subprocess returned error exit status 10
    Errors were encountered while processing:
     metax-linux-dkms
    ERROR: failed to install deb/metax-linux_3.9.10-1_amd64.deb
    

    错误原因:metax-linux-3.9.10/common/mx_devinfo_c500.h 中代码有问题:

    #define MX_DEV_PCI_C500_LIST    \
    MX_DEV_PCI_C500(0x9999, 0x4001, 0x4019, c500, MXC500, mx_devinfo_c500)     \
    MX_DEV_PCI_C500(0x9999, 0x4000, 0x4018, c550, MXC550, mx_devinfo_c550)     \
    MX_DEV_PCI_C500(0x9999, 0x4083, 0x409b, n260, MXN260, mx_devinfo_n260)     \
    MX_DEV_PCI_C500(0x9999, 0x4040, 0x4058, c500x, MXC500X, mx_devinfo_c500x)  \
    MX_DEV_PCI_C500(0x9999, 0x4020, 0x4038, c588, MXC588, mx_devinfo_c588)
    

    正确的代码应该是:

    #define MX_DEV_PCI_C500_LIST    \
    MX_DEV_PCI_C500(0x9999, 0x4001, 0x4019, c500, "MXC500", mx_devinfo_c500)     \
    MX_DEV_PCI_C500(0x9999, 0x4000, 0x4018, c550, "MXC550", mx_devinfo_c550)     \
    MX_DEV_PCI_C500(0x9999, 0x4083, 0x409b, n260, "MXN260", mx_devinfo_n260)     \
    MX_DEV_PCI_C500(0x9999, 0x4040, 0x4058, c500x, "MXC500X", mx_devinfo_c500x)  \
    MX_DEV_PCI_C500(0x9999, 0x4020, 0x4038, c588, "MXC588", mx_devinfo_c588)
    

    也就是说倒数第二个参数本应是字符串,双引号都丢失了

  • See post chevron_right
    e411
    Members
    C550的gpu使用率为何不能突破百分之80? 已解决 2026年6月26日 09:59

    据我了解,并发多了之后,华为的机器也未必就很快,体验也就一般般。910b的算力和带宽都和沐曦差不多的

  • See post chevron_right
    e411
    Members
    C550的gpu使用率为何不能突破百分之80? 已解决 2026年6月26日 09:58

    至于npu的占用,这个应该是不同的厂家配置的占用率的算法不一样,没法直接比较,尤其是华为是npu,沐曦是gpu,架构差距很大,两个指标没法直接比较

  • See post chevron_right
    e411
    Members
    C550的gpu使用率为何不能突破百分之80? 已解决 2026年6月26日 09:57

    用sglang部署,开MTP,开speculative V2。我这里4*C500用sglang部署qwen3.6 27b,可以达到单请求80token/s左右,8请求总吞吐应该有300token/s左右,1个请求能跑个40左右,也能接受。vllm-metax的qwen的MTP是坏的,打开推理结果会乱码

  • See post chevron_right
    e411
    Members
    C550的gpu使用率为何不能突破百分之80? 已解决 2026年6月26日 09:49

    大模型推理的任务特性就决定了他的gpu占用率不会特别高。qwen3.6 27b已经是密度比较大的了,这占用还比较高,如果是那些moe模型更难跑高。通讯、数据等等占了大头

  • See post chevron_right
    e411
    Members
    DeepSeek-V4 已解决 2026年6月4日 09:59

    我运行是没跑起来,提示算子重复注册

  • See post chevron_right
    e411
    Members
    DeepSeek-V4-Flash 量化部署 已解决 2026年5月8日 20:45

    请问下 modelscope.cn/models/metax-tech/DeepSeek-V4-Flash-FlexSMQ-AWQ-W8A8 这个量化模型在8卡沐曦C500上是否能够部署,如果要部署应使用哪个镜像,使用什么参数启动?

  • 沐曦开发者论坛
powered by misago