{"id":2657,"date":"2026-08-31T18:00:00","date_gmt":"2026-08-31T10:00:00","guid":{"rendered":"https:\/\/www.o6s.net\/?p=2657"},"modified":"2026-09-01T16:06:42","modified_gmt":"2026-09-01T08:06:42","slug":"%e6%9e%81%e9%80%9f%e4%b8%8a%e6%89%8b%ef%bc%81openeuler-24-03-%e6%90%ad%e5%bb%ba-llama-cpp-%e7%8e%af%e5%a2%83%e5%8f%8a-benchmark-%e6%80%a7%e8%83%bd%e5%ae%9e%e6%b5%8b","status":"publish","type":"post","link":"https:\/\/www.o6s.net\/index.php\/2026\/08\/31\/%e6%9e%81%e9%80%9f%e4%b8%8a%e6%89%8b%ef%bc%81openeuler-24-03-%e6%90%ad%e5%bb%ba-llama-cpp-%e7%8e%af%e5%a2%83%e5%8f%8a-benchmark-%e6%80%a7%e8%83%bd%e5%ae%9e%e6%b5%8b\/","title":{"rendered":"\u6781\u901f\u4e0a\u624b\uff01openEuler 24.03 \u642d\u5efa llama.cpp \u73af\u5883\u53ca Benchmark \u6027\u80fd\u5b9e\u6d4b"},"content":{"rendered":"
<\/section>\n
\n
\n
\n
\n

\u76ee\u524d\uff0cOpenAtom openEuler\uff08\u7b80\u79f0\u201copenEuler\u201d\u6216\u201c\u5f00\u6e90\u6b27\u62c9\u201d\uff0924.03 \u5df2<\/span><\/span>\u539f\u751f\u652f\u6301 llama.cpp<\/span><\/strong>\u3002\u7ee7\u4e0a\u4e00\u7bc7\u300a\u624b\u628a\u624b\u5b9e\u6d4b\uff01openEuler 24.03 \u90e8\u7f72 SGLang\uff0cBenchmark \u6570\u636e\u51fa\u7089<\/a>\u300b\u4e4b\u540e\uff0c\u672c\u6587\u5c06\u7ee7\u7eed\u56f4\u7ed5 AI \u63a8\u7406\u6846\u67b6\u5c55\u5f00\u5b9e\u6d4b\uff0c\u5e26\u4f60<\/span><\/span>\u5728 openEuler 24.03 \u73af\u5883\u4e0b\u4ece\u96f6\u642d\u5efa llama.cpp\u3002<\/span><\/strong><\/p>\n


<\/span><\/p>\n

\u672c\u6587\u4ece\u6700\u5e72\u51c0\u7684 openEuler \u548c CANN \u57fa\u7840\u955c\u50cf\u5f00\u59cb\uff0c\u8be6\u7ec6\u4ecb\u7ecd llama.cpp \u7684\u73af\u5883\u51c6\u5907\u3001\u6e90\u7801\u7f16\u8bd1\u53ca\u63a8\u7406\u670d\u52a1\u542f\u52a8\uff0c\u5e76\u901a\u8fc7\u4e00\u8f6e Random Benchmark \u5b9e\u6d4b\u5176\u63a8\u7406\u6027\u80fd\u3002\u7a76\u7adf\u5728 openEuler \u4e0a\u90e8\u7f72 llama.cpp \u7684\u8fc7\u7a0b\u662f\u5426\u987a\u7545\uff0c\u5b9e\u9645\u6027\u80fd\u8868\u73b0\u5982\u4f55\uff1f\u4e0b\u9762\u5c31\u4e00\u8d77\u52a8\u624b\u8bd5\u8bd5\u3002<\/span><\/span><\/p>\n

\n


<\/span><\/h2>\n
\n
\n
\n
\n

<\/span><\/section>\n

<\/span><\/section>\n<\/section>\n
\n
\n
\u00a0<\/span>\u4e00\u3001\u73af\u5883\u51c6\u5907<\/span><\/span><\/strong><\/section>\n<\/section>\n
\n

<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n

<\/span><\/section>\n<\/section>\n<\/section>\n


<\/span><\/h2>\n

\u672c\u6b21\u5b9e\u6d4b\u7684<\/span>\u5bbf\u4e3b\u673a\u73af\u5883<\/span><\/strong>\u4e3a\uff1a<\/span><\/span><\/p>\n\n\n\n\n\n\n\n\n
\n
\u9879\u76ee<\/span><\/span><\/section>\n<\/td>\n
\n
\u7248\u672c\u4fe1\u606f<\/span><\/span><\/section>\n<\/td>\n<\/tr>\n
\n
OS<\/span><\/span><\/section>\n<\/td>\n
\n
openEuler release 24.03 \u00a0 (LTS-SP4)<\/span><\/span><\/section>\n<\/td>\n<\/tr>\n
\n
\u5185\u6838<\/span><\/span><\/section>\n<\/td>\n
\n
6.6.0<\/span><\/span><\/section>\n<\/td>\n<\/tr>\n
\n
Docker<\/span><\/span><\/section>\n<\/td>\n
\n
18.09.0<\/span><\/span><\/section>\n<\/td>\n<\/tr>\n
\n
NPU \u9a71\u52a8<\/span><\/span><\/section>\n<\/td>\n
\n
26.0.rc1<\/span><\/span><\/section>\n<\/td>\n<\/tr>\n
\n
NPU \u56fa\u4ef6<\/span><\/span><\/section>\n<\/td>\n
\n
7.7.0.10.220<\/span><\/span><\/section>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

<\/span><\/section>\n

\u258f1.\u786e\u4fdd\u5bbf\u4e3b\u673a\u4f7f\u7528 openEuler \u56fd\u5185\u52a0\u901f yum \u6e90<\/span><\/span><\/strong><\/p>\n


<\/span><\/p>\n

\ud83d\udda5\ufe0f<\/span>\u00a0\u5bbf\u4e3b\u673a\u6267\u884c<\/span><\/strong><\/span><\/p>\n


<\/span><\/p>\n

\u8fd0\u884c\u00a0<\/span><\/span>cat \/etc\/yum.repos.d\/openEuler.repo<\/span><\/span><\/code>\u00a0\u68c0\u67e5\uff0c\u5982\u679c\u6587\u4ef6\u4e0d\u5b58\u5728\u6216\u8005\u4f7f\u7528\u7684\u6e90\u4e0d\u662f openEuler \u56fd\u5185\u955c\u50cf\uff0c\u53ef\u4ee5\u6309\u5982\u4e0b\u4fee\u6539\uff0c\u63d0\u5347\u00a0<\/span><\/span>yum install<\/span><\/span><\/code>\u00a0\u4e0b\u8f7d\u901f\u5ea6\u3002<\/span><\/span><\/p>\n


<\/span><\/p>\n

\u4ee5\u4e0b\u662f openEuler 24.03 LTS SP4 \u955c\u50cf\u6e90\u914d\u7f6e\uff1a<\/span><\/span><\/p>\n

\n
\n
\n
\n
\n
    \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • <\/ul>\n
    #generic-repos is licensed under the Mulan PSL v2.<\/span><\/span><\/code>#Youcan use this software according to the terms and conditions of the Mulan PSL v2.<\/span><\/span><\/code>#You may obtain a copy of Mulan PSL v2 at:<\/span><\/span><\/code># \u00a0 \u00a0http:\/\/license.coscl.org.cn\/MulanPSL2<\/span><\/span><\/code>#THIS SOFTWARE IS PROVIDED ON AN \"AS IS\" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR<\/span><\/span><\/code>#IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY OR FIT FOR A PARTICULAR<\/span><\/span><\/code>#PURPOSE.<\/span><\/span><\/code>#See the Mulan PSL v2 for more details.<\/span><\/span><\/code>
    <\/span><\/code>[OS]<\/span><\/span><\/code>name<\/span>=OS<\/span><\/code>baseurl<\/span>=https:\/\/mirrors.huaweicloud.com\/openeuler\/openEuler-24.03<\/span>-LTS-SP4\/OS\/$basearch<\/span>\/<\/span><\/code>#metalink=https:\/\/mirrors.openeuler.org\/metalink?repo=$releasever\/OS&arch=$basearch<\/span><\/span><\/code>metadata_expire<\/span>=1<\/span>h<\/span><\/code>enabled<\/span>=1<\/span><\/span><\/code>gpgcheck<\/span>=1<\/span><\/span><\/code>gpgkey<\/span>=http:\/\/repo.openeuler.org\/openEuler-24.03<\/span>-LTS-SP4\/OS\/$basearch<\/span>\/RPM-GPG-KEY-openEuler<\/span><\/code>
    <\/span><\/code>[everything]<\/span><\/span><\/code>name<\/span>=everything<\/span><\/code>baseurl<\/span>=https:\/\/mirrors.huaweicloud.com\/openeuler\/openEuler-24.03<\/span>-LTS-SP4\/everything\/$basearch<\/span>\/<\/span><\/code>#metalink=https:\/\/mirrors.openeuler.org\/metalink?repo=$releasever\/everything&arch=$basearch<\/span><\/span><\/code>metadata_expire<\/span>=1<\/span>h<\/span><\/code>enabled<\/span>=1<\/span><\/span><\/code>gpgcheck<\/span>=1<\/span><\/span><\/code>gpgkey<\/span>=http:\/\/repo.openeuler.org\/openEuler-24.03<\/span>-LTS-SP4\/everything\/$basearch<\/span>\/RPM-GPG-KEY-openEuler<\/span><\/code>
    <\/span><\/code>[EPOL]<\/span><\/span><\/code>name<\/span>=EPOL<\/span><\/code>baseurl<\/span>=https:\/\/mirrors.huaweicloud.com\/openeuler\/openEuler-24.03<\/span>-LTS-SP4\/EPOL\/main\/$basearch<\/span>\/<\/span><\/code>#metalink=https:\/\/mirrors.openeuler.org\/metalink?repo=$releasever\/EPOL\/main&arch=$basearch<\/span><\/span><\/code>metadata_expire<\/span>=1<\/span>h<\/span><\/code>enabled<\/span>=1<\/span><\/span><\/code>gpgcheck<\/span>=1<\/span><\/span><\/code>gpgkey<\/span>=http:\/\/repo.openeuler.org\/openEuler-24.03<\/span>-LTS-SP4\/OS\/$basearch<\/span>\/RPM-GPG-KEY-openEuler<\/span><\/code>
    <\/span><\/code>[debuginfo]<\/span><\/span><\/code>name<\/span>=debuginfo<\/span><\/code>baseurl<\/span>=https:\/\/mirrors.huaweicloud.com\/openeuler\/openEuler-24.03<\/span>-LTS-SP4\/debuginfo\/$basearch<\/span>\/<\/span><\/code>#metalink=https:\/\/mirrors.openeuler.org\/metalink?repo=$releasever\/debuginfo&arch=$basearch<\/span><\/span><\/code>metadata_expire<\/span>=1<\/span>h<\/span><\/code>enabled<\/span>=1<\/span><\/span><\/code>gpgcheck<\/span>=1<\/span><\/span><\/code>gpgkey<\/span>=http:\/\/repo.openeuler.org\/openEuler-24.03<\/span>-LTS-SP4\/debuginfo\/$basearch<\/span>\/RPM-GPG-KEY-openEuler<\/span><\/code>
    <\/span><\/code>[source]<\/span><\/span><\/code>name<\/span>=source<\/span><\/code>baseurl<\/span>=https:\/\/mirrors.huaweicloud.com\/openeuler\/openEuler-24.03<\/span>-LTS-SP4\/source\/<\/span><\/code>#metalink=https:\/\/mirrors.openeuler.org\/metalink?repo=$releasever&arch=source<\/span><\/span><\/code>metadata_expire<\/span>=1<\/span>h<\/span><\/code>enabled<\/span>=1<\/span><\/span><\/code>gpgcheck<\/span>=1<\/span><\/span><\/code>gpgkey<\/span>=http:\/\/repo.openeuler.org\/openEuler-24.03<\/span>-LTS-SP4\/source\/RPM-GPG-KEY-openEuler<\/span><\/code>
    <\/span><\/code>[update]<\/span><\/span><\/code>name<\/span>=update<\/span><\/code>baseurl<\/span>=https:\/\/mirrors.huaweicloud.com\/openeuler\/openEuler-24.03<\/span>-LTS-SP4\/update\/$basearch<\/span>\/<\/span><\/code>#metalink=https:\/\/mirrors.openeuler.org\/metalink?repo=$releasever\/update&arch=$basearch<\/span><\/span><\/code>metadata_expire<\/span>=1<\/span>h<\/span><\/code>enabled<\/span>=1<\/span><\/span><\/code>gpgcheck<\/span>=1<\/span><\/span><\/code>gpgkey<\/span>=http:\/\/repo.openeuler.org\/openEuler-24.03<\/span>-LTS-SP4\/OS\/$basearch<\/span>\/RPM-GPG-KEY-openEuler<\/span><\/code>
    <\/span><\/code>[update-source]<\/span><\/span><\/code>name<\/span>=update-source<\/span><\/code>baseurl<\/span>=https:\/\/mirrors.huaweicloud.com\/openeuler\/openEuler-24.03<\/span>-LTS-SP4\/update\/source\/<\/span><\/code>#metalink=https:\/\/mirrors.openeuler.org\/metalink?repo=$releasever&arch=source<\/span><\/span><\/code>metadata_expire<\/span>=1<\/span>h<\/span><\/code>enabled<\/span>=1<\/span><\/span><\/code>gpgcheck<\/span>=1<\/span><\/span><\/code>gpgkey<\/span>=http:\/\/repo.openeuler.org\/openEuler-24.03<\/span>-LTS-SP4\/source\/RPM-GPG-KEY-openEuler<\/span><\/code><\/pre>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

    <\/span><\/p>\n
    \u258f2. \u5b89\u88c5 Docker<\/span><\/span><\/strong><\/span><\/h3>\n

    <\/span><\/p>\n
    \ud83d\udda5\ufe0f\u00a0<\/span>\u5bbf\u4e3b\u673a\u6267\u884c<\/span><\/strong><\/span><\/p>\n

    <\/span><\/p>\n
    openEuler 24.03 \u8f6f\u4ef6\u4ed3\u91cc\u6709 Docker\uff0c\u76f4\u63a5\u5b89\u88c5\uff1a<\/span><\/span><\/p>\n
    \n
      \n
    • \n
    • \n
    • \n
    • \n
    • \n
    • \n
    • \n
    • <\/ul>\n
      # \u5b89\u88c5 Docker<\/span><\/span><\/code>sudo<\/span>\u00a0dnf install -y docker<\/span><\/code>
      <\/span><\/code># \u542f\u52a8 Docker \u5e76\u8bbe\u7f6e\u5f00\u673a\u81ea\u542f<\/span><\/span><\/code>sudo<\/span>\u00a0systemctl\u00a0enable<\/span>\u00a0--now docker<\/span><\/code>
      <\/span><\/code># \u9a8c\u8bc1\u5b89\u88c5\u662f\u5426\u6210\u529f<\/span><\/span><\/code>docker --version<\/span><\/code><\/pre>\n<\/section>\n

      <\/span><\/p>\n
      \u258f3. \u914d\u7f6e\u514d sudo \u8fd0\u884c\uff08\u53ef\u9009\uff09<\/span><\/span><\/strong><\/span><\/h3>\n

      <\/span><\/p>\n
      \ud83d\udda5\ufe0f<\/span>\u00a0\u5bbf\u4e3b\u673a\u6267\u884c<\/span><\/strong><\/span><\/p>\n

      <\/span><\/p>\n
      \u4e3a\u4e86\u65b9\u4fbf\u65e5\u5e38\u64cd\u4f5c\uff0c\u5efa\u8bae\u5c06\u5f53\u524d\u7528\u6237\u52a0\u5165\u00a0<\/span><\/span>docker<\/span><\/span><\/code>\u00a0\u7ec4\uff0c\u8fd9\u6837\u8fd0\u884c\u547d\u4ee4\u65f6\u65e0\u9700\u6bcf\u6b21\u90fd\u8f93\u5165\u00a0<\/span><\/span>sudo<\/span><\/code>\uff1a<\/span><\/span><\/p>\n
      \n
        \n
      • \n
      • \n
      • \n
      • \n
      • <\/ul>\n
        # \u5c06\u5f53\u524d\u7528\u6237\u52a0\u5165 docker \u7ec4<\/span><\/span><\/code>sudo<\/span>\u00a0usermod -aG docker\u00a0$USER<\/span><\/span><\/code>
        <\/span><\/code># \u5237\u65b0\u7ec4\u6743\u9650\u4f7f\u914d\u7f6e\u7acb\u5373\u751f\u6548<\/span><\/span><\/code>newgrp docker<\/span><\/code><\/pre>\n<\/section>\n

        <\/span><\/p>\n
        \u258f4. \u62c9\u53d6 openEuler 24.03\u4e0e CANN \u57fa\u7840\u955c\u50cf<\/span><\/span><\/strong><\/p>\n

        <\/span><\/p>\n
        \ud83d\udda5\ufe0f\u00a0<\/span>\u5bbf\u4e3b\u673a\u6267\u884c<\/span><\/strong><\/span><\/p>\n

        <\/span><\/p>\n
        \u672c\u6587\u4f7f\u7528\u57fa\u4e8e openEuler 24.03 SP4 \u4e0e Python 3.11 \u7684\u534e\u4e3a\u5b98\u65b9 CANN 9.0.0 \u5bb9\u5668\u955c\u50cf\uff1a<\/span><\/span><\/p>\n
        \n
          \n
        • <\/ul>\n
          docker<\/span>\u00a0pull swr.cn-south-1<\/span>.myhuaweicloud.com\/ascendhub\/cann:9<\/span>.0<\/span>.0<\/span>-910<\/span>b-openeuler24.03<\/span>-py3.11<\/span>-devel<\/span><\/code><\/pre>\n<\/section>\n

          <\/span><\/p>\n
          \u258f5. \u786c\u4ef6\u4e0e\u9a71\u52a8<\/span><\/span><\/strong><\/p>\n

          <\/span><\/p>\n
          \ud83d\udda5\ufe0f\u00a0<\/span>\u5bbf\u4e3b\u673a\u6267\u884c<\/span><\/strong><\/span><\/p>\n

          <\/span><\/p>\n
          \u786e\u4fdd\u5bbf\u4e3b\u673a\u5df2\u6b63\u786e\u5b89\u88c5\u6607\u817e\u9a71\u52a8\uff0cNPU \u8bbe\u5907\u53ef\u89c1\uff1a<\/span><\/span><\/p>\n
          \n
            \n
          • \n
          • \n
          • \n
          • \n
          • \n
          • <\/ul>\n
            #<\/span>\u00a0\u67e5\u770b NPU \u8bbe\u5907<\/span><\/code>ls \/dev\/davinci*<\/span><\/code>#<\/span>\u00a0\u9884\u671f\u8f93\u51fa\uff1a\/dev\/davinci0 \u00a0\/dev\/davinci2 \u00a0\/dev\/davinci_manager ...<\/span><\/code>
            <\/span><\/code># \u67e5\u770b NPU \u72b6\u6001<\/span><\/code>npu-smi info<\/span><\/code><\/pre>\n<\/section>\n

            <\/span><\/p>\n
            \u8fd0\u884c\u00a0<\/span><\/span>npu-smi info<\/span><\/code>\u00a0\u786e\u8ba4 NPU \u8bbe\u5907\u72b6\u6001\u6b63\u5e38\uff1a<\/span><\/span><\/p>\n

            <\/span><\/span><\/p>\n
            <\/section>\n

            <\/span><\/span><\/p>\n
            \u258f6. \u6a21\u578b\u6743\u91cd\u51c6\u5907<\/span><\/span><\/strong><\/span><\/h3>\n

            <\/span><\/p>\n
            \ud83d\udda5\ufe0f\u00a0<\/span>\u5bbf\u4e3b\u673a\u6267\u884c<\/span><\/strong><\/span><\/p>\n

            <\/span><\/p>\n
            \u5c06\u6a21\u578b\u6743\u91cd\u653e\u5230\u5bbf\u4e3b\u673a\u7edf\u4e00\u76ee\u5f55\uff0c\u540e\u7eed\u5bb9\u5668\u6302\u8f7d\u4f7f\u7528\uff1a<\/span><\/span><\/p>\n
            \n
              \n
            • \n
            • \n
            • <\/ul>\n
              # \u4ee5 Qwen3-0.6B \u4e3a\u4f8b<\/span><\/span><\/code>ls<\/span>\u00a0\/home\/models\/Qwen3-0<\/span>.6<\/span>B\/<\/span><\/code># \u9884\u671f\u5305\u542b\uff1aconfig.json \u00a0tokenizer.json \u00a0model*.safetensors ...<\/span><\/span><\/code><\/pre>\n<\/section>\n

              <\/span><\/p>\n
              \u258f7. \u56fd\u5185\u955c\u50cf\u52a0\u901f<\/span><\/span><\/strong><\/p>\n

              <\/span><\/p>\n
              \u5728\u5bb9\u5668\u5185\u64cd\u4f5c\u524d\uff0c\u5148\u89c4\u5212\u597d\u955c\u50cf\u914d\u7f6e\u3002\u4ee5\u4e0b\u955c\u50cf\u7b56\u7565\u5c06\u5728\u540e\u7eed\u5404\u6846\u67b6\u6b65\u9aa4\u4e2d\u9010\u4e00\u5e94\u7528\uff1a<\/span><\/span><\/p>\n
              \n
              <\/section>\n
              \n

              <\/span><\/p>\n<\/section>\n
              \n
              \n
              \n
              \n

              <\/span><\/section>\n

              <\/span><\/section>\n<\/section>\n
              \n
              \n
              \u4e8c\u3001llama.cpp \u63a8\u7406\u73af\u5883\u642d\u5efa<\/span><\/strong><\/span><\/section>\n<\/section>\n
              \n

              <\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n

              <\/span><\/section>\n<\/section>\n<\/section>\n

              <\/span><\/h2>\n
              \u672c\u6b21 llama.cpp \u7f16\u8bd1\u73af\u5883\u4f4d\u4e8e\u5bb9\u5668\u5185\u90e8\uff0c\u5bb9\u5668\u955c\u50cf\u57fa\u4e8e openEuler 24.03 LTS SP4 \u6784\u5efa\uff0c\u5df2\u7ecf\u914d\u7f6e\u597d\u6607\u817e\u9a71\u52a8\u548c CANN\uff0c\u65e0\u9700\u505a\u4efb\u4f55\u914d\u7f6e\u3002<\/span><\/span><\/p>\n

              <\/span><\/p>\n
              llama.cpp \u662f\u7eaf C++ \u5b9e\u73b0\uff0c\u901a\u8fc7 GGML_CANN \u540e\u7aef\u9a71\u52a8\u6607\u817e NPU\uff0c\u5b83\u9700\u8981\u7f16\u8bd1\u6e90\u7801\uff0c\u4e14\u6a21\u578b\u5fc5\u987b\u662f GGUF \u683c\u5f0f\u3002<\/span><\/span><\/p>\n

              <\/span><\/span><\/p>\n
              \u7248\u672c\u914d\u5957\uff1a<\/span><\/span><\/strong><\/p>\n\n\n\n\n\n\n
              \n
              \u7ec4\u4ef6<\/span><\/span><\/section>\n<\/td>\n
              		\n
              \u7248\u672c<\/span><\/span><\/section>\n<\/td>\n<\/tr>\n
              \n
              llama.cpp<\/span><\/span><\/section>\n<\/td>\n
              		\n
              master \u5206\u652f\uff08\u6700\u65b0\uff09<\/span><\/span><\/section>\n<\/td>\n<\/tr>\n
              \n
              CANN\uff08\u5bb9\u5668\u5185\uff09<\/span><\/span><\/section>\n<\/td>\n
              		\n
              9.0.0<\/span><\/span><\/section>\n<\/td>\n<\/tr>\n
              \n
              GGUF \u91cf\u5316<\/span><\/span><\/section>\n<\/td>\n
              		\n
              f16\uff08\u4e5f\u652f\u6301 q8_0 \/ q4_0\uff09<\/span><\/span><\/section>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

              <\/span><\/p>\n
              \u258fStep 1\uff1a\u542f\u52a8\u5bb9\u5668<\/span><\/strong><\/span><\/h3>\n

              <\/span><\/p>\n
              \ud83d\udda5\ufe0f\u00a0<\/span>\u5bbf\u4e3b\u673a\u6267\u884c<\/span><\/strong><\/span><\/p>\n

              <\/span><\/p>\n
              \u542f\u52a8\u5bb9\u5668\u5e76\u6302\u8f7d NPU \u8bbe\u5907\u3001\u6a21\u578b\u76ee\u5f55\u7b49\uff1a<\/span><\/span><\/p>\n
              \n
                \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • \n
              • <\/ul>\n
                docker<\/span>\u00a0run -d\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 --name llama-cpp-ascend-910<\/span>b\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 --device=\/dev\/davinci2\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 --device=\/dev\/davinci_manager\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 --device=\/dev\/devmm_svm\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 --device=\/dev\/hisi_hdc\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 -v \/usr\/local\/dcmi:\/usr\/local\/dcmi:ro\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 -v \/usr\/local\/bin\/npu-smi:\/usr\/local\/bin\/npu-smi:ro\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 -v \/usr\/local\/Ascend\/driver:\/usr\/local\/Ascend\/driver:ro\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 -v \/etc\/ascend_install.info:\/etc\/ascend_install.info:ro\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 -v \/home\/models:\/models\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 -v $(pwd)\/llamacpp-workspace:\/workspace\/home\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 -v $(pwd)\/pip-cache:\/root\/.cache\/pip\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 -w \/workspace\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 -p\u00a08001<\/span>:8001<\/span>\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 --shm-size=16<\/span>g\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 --ulimit memlock=-1<\/span>\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 --ulimit stack=67108864<\/span>\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 -e OMP_NUM_THREADS=8<\/span>\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 swr.cn-south-1<\/span>.myhuaweicloud.com\/ascendhub\/cann:9<\/span>.0<\/span>.0<\/span>-910<\/span>b-openeuler24.03<\/span>-py3.11<\/span>-devel\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 sleep infinity<\/span><\/code><\/pre>\n<\/section>\n
                \n
                \n
                \u989d\u5916\u6302\u8f7d\u00a0<\/span><\/span><\/span>llamacpp-workspace\u00a0<\/span><\/span><\/code>\u7528\u4e8e\u6301\u4e45\u5316\u6e90\u7801\u548c\u7f16\u8bd1\u4ea7\u7269\uff0c\u4e8c\u6b21\u90e8\u7f72\u53ef\u8df3\u8fc7\u7f16\u8bd1\u3002<\/span><\/span><\/span><\/p>\n
                -v \/home\/models:\/models<\/span><\/span><\/code>\u00a0\u6302\u8f7d\u5bbf\u4e3b\u673a\u7684\u6a21\u578b\u6743\u91cd\u8def\u5f84\uff0c\u540e\u7eed\u542f\u52a8\u63a8\u7406\u670d\u52a1\u7528\u5f97\u5230\uff0c\u7528\u6237\u53ef\u81ea\u884c\u914d\u7f6e\u8fd9\u4e2a\u8def\u5f84\u3002<\/span><\/span><\/span><\/p>\n<\/section>\n<\/section>\n

                <\/span><\/p>\n
                \u8fdb\u5165\u5bb9\u5668<\/span><\/span><\/strong><\/h4>\n

                <\/span><\/p>\n
                \ud83d\udce6<\/span>\u00a0\u5bb9\u5668\u5185\u6267\u884c<\/span><\/strong><\/span><\/p>\n
                \n
                  \n
                • <\/ul>\n
                  docker<\/span>\u00a0exec -it llama-cpp-ascend-910<\/span>b bash -l<\/span><\/code><\/pre>\n<\/section>\n

                  <\/span><\/p>\n
                  \u258fStep 2\uff1a\u7f16\u8bd1 llama.cpp<\/span><\/strong><\/span><\/h3>\n

                  <\/span><\/p>\n
                  \ud83d\udce6\u00a0<\/span>\u5bb9\u5668\u5185\u6267\u884c<\/span><\/strong><\/span><\/p>\n

                  <\/span><\/strong><\/span><\/p>\n
                  \u5148 source CANN \u73af\u5883\u53d8\u91cf\uff1a<\/span><\/span><\/p>\n
                  \n
                    \n
                  • \n
                  • <\/ul>\n
                    # \u5bb9\u5668\u5185\u6267\u884c\u2014\u2014\u5148 source CANN \u73af\u5883\u53d8\u91cf<\/span><\/span><\/code>source<\/span>\u00a0\/usr\/local\/Ascend\/ascend-toolkit\/set_env.sh<\/span><\/code><\/pre>\n<\/section>\n

                    <\/span><\/p>\n<\/section>\n
                    2.1 \u5b89\u88c5\u7f16\u8bd1\u5de5\u5177\u94fe<\/span><\/p>\n
                    \u66ff\u6362 yum \u6e90\u5e76\u5b89\u88c5\u7f16\u8bd1\u4f9d\u8d56\uff1a<\/span><\/span><\/p>\n
                    \n
                      \n
                    • \n
                    • \n
                    • \n
                    • \n
                    • \n
                    • \n
                    • \n
                    • \n
                    • <\/ul>\n
                      # \u66ff\u6362 openEuler \u56fd\u5185\u52a0\u901f yum \u6e90<\/span><\/span><\/code>sed -i -e\u00a0's|https:\/\/repo.openeuler.org\/|https:\/\/mirrors.huaweicloud.com\/openeuler\/|'<\/span>\u00a0<\/span><\/code>\u00a0 \u00a0 \u00a0 \u00a0-e\u00a0'\/^metalink=\/s\/^\/#\/'<\/span>\u00a0\/etc\/yum.repos.d\/openEuler.repo<\/span><\/code>
                      <\/span><\/code># \u5b89\u88c5\u7f16\u8bd1\u4f9d\u8d56<\/span><\/span><\/code>command<\/span>\u00a0-v cmake \u00a0>\/dev\/null 2>&1 || dnf install -y cmake<\/span><\/code>command<\/span>\u00a0-v git \u00a0 \u00a0>\/dev\/null 2>&1 || dnf install -y git<\/span><\/code>command<\/span>\u00a0-v gcc \u00a0 \u00a0>\/dev\/null 2>&1 || dnf install -y gcc gcc-c++ make<\/span><\/code>command<\/span>\u00a0-v ccache >\/dev\/null 2>&1 || dnf install -y ccache<\/span><\/code><\/pre>\n<\/section>\n
                      ccache<\/span><\/code>\u00a0\u53ef\u4ee5\u7f13\u5b58\u7f16\u8bd1\u4e2d\u95f4\u6587\u4ef6\uff0c\u589e\u91cf\u7f16\u8bd1\u65f6\u5927\u5e45\u52a0\u901f\u3002<\/span><\/p>\n

                      <\/span><\/p>\n
                      2.2 \u514b\u9686\u6e90\u7801<\/span><\/span><\/strong><\/h4>\n
                      \u901a\u8fc7 GitHub \u955c\u50cf\u4ee3\u7406\u52a0\u901f\u62c9\u53d6\u6e90\u7801\uff1a<\/span><\/span><\/p>\n
                      \n
                        \n
                      • \n
                      • \n
                      • \n
                      • <\/ul>\n
                        git\u00a0clone<\/span>\u00a0--depth 1 <\/span><\/code>\u00a0 \u00a0 https:\/\/ghfast.top\/https:\/\/github.com\/ggml-org\/llama.cpp.git <\/span><\/code>\u00a0 \u00a0 \/workspace\/home\/llama.cpp<\/span><\/code>cd<\/span>\u00a0\/workspace\/home\/llama.cpp<\/span><\/code><\/pre>\n<\/section>\n

                        <\/span><\/p>\n
                        2.3 \u7f16\u8bd1\uff08CANN \u540e\u7aef\uff09<\/span><\/span><\/strong><\/p>\n
                        \u4f7f\u7528 CMake \u914d\u7f6e\u5e76\u7f16\u8bd1 llama.cpp\uff0c\u542f\u7528 CANN \u540e\u7aef\uff1a<\/span><\/span><\/p>\n
                        \n
                          \n
                        • \n
                        • \n
                        • <\/ul>\n
                          cmake<\/span>\u00a0-B build -DGGML_CANN=on<\/span>\u00a0-DCMAKE_BUILD_TYPE=release\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 \u00a0 -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache<\/span><\/code>cmake<\/span>\u00a0--build build --config release -j $(nproc)<\/span><\/code><\/pre>\n<\/section>\n
                          \u7f16\u8bd1\u65f6\u95f4\u53d6\u51b3\u4e8e\u673a\u5668\u6027\u80fd\uff0c\u901a\u5e38\u5728\u51e0\u5206\u949f\u5230\u5341\u51e0\u5206\u949f\u3002<\/span><\/span><\/p>\n

                          <\/span><\/p>\n
                          2.4 \u9a8c\u8bc1\u7f16\u8bd1\u4ea7\u7269<\/span><\/span><\/strong><\/h4>\n
                          \u786e\u8ba4\u7f16\u8bd1\u4ea7\u7269\u5df2\u6b63\u786e\u751f\u6210\uff1a<\/span><\/span><\/p>\n
                          \n
                            \n
                          • \n
                          • <\/ul>\n
                            ls<\/span>\u00a0-lh build\/bin\/llama-server<\/span><\/code>build\/bin\/llama-server --version<\/span><\/code><\/pre>\n<\/section>\n

                            <\/span><\/p>\n
                            \u7f16\u8bd1\u5b8c\u6210\u540e\uff0c\u786e\u8ba4<\/span><\/span>\u00a0llama-server\u00a0<\/span><\/code>\u5df2\u6b63\u786e\u751f\u6210\u5e76\u67e5\u770b\u7248\u672c\u4fe1\u606f\uff1a<\/span><\/span><\/p>\n
                            <\/section>\n

                            <\/span><\/span><\/p>\n
                            <\/span><\/section>\n
                            \u258fStep 3\uff1a\u51c6\u5907 GGUF \u6a21\u578b<\/span><\/strong><\/p>\n

                            <\/span><\/p>\n
                            llama.cpp \u53ea\u652f\u6301 GGUF \u683c\u5f0f\u3002\u5982\u679c\u4f60\u7684\u6a21\u578b\u76ee\u5f55\u4e2d\u5df2\u6709<\/span>\u00a0<\/span>.gguf<\/span><\/span>\u00a0\u6587\u4ef6\uff0c\u53ef\u4ee5\u76f4\u63a5\u4f7f\u7528\uff1b\u5426\u5219\u9700\u8981\u4ece HuggingFace safetensors \u683c\u5f0f\u8f6c\u6362\u3002<\/span><\/span><\/p>\n

                            <\/span><\/p>\n
                            \u56fd\u5185\u7528\u6237\u63a8\u8350\u4ece ModelScope \u76f4\u63a5\u4e0b\u8f7d GGUF \u683c\u5f0f\u7684\u6743\u91cd\uff0c\u901f\u5ea6\u8f83\u5feb\uff0c\u672c\u6587\u4f7f\u7528\u7684\u662f ModelScope \u4e0a\u7684\u00a0<\/span><\/span>unsloth\/Qwen3.5-9B-GGUF<\/span><\/code>\uff0c\u4e0d\u9700\u8981\u8f6c\u6362\u3002<\/span><\/span><\/p>\n

                            <\/span><\/p>\n
                            \u60c5\u51b5 A\uff1a\u5df2\u6709 GGUF \u6587\u4ef6<\/span><\/span><\/strong><\/p>\n
                            \n
                              \n
                            • \n
                            • <\/ul>\n
                              ls<\/span>\u00a0\/models\/Qwen3-0<\/span>.6<\/span>B\/*.gguf<\/span><\/code># \u76f4\u63a5\u4f7f\u7528\u5df2\u6709\u7684 gguf \u6587\u4ef6<\/span><\/span><\/code><\/pre>\n<\/section>\n

                              <\/span><\/p>\n
                              \u60c5\u51b5 B\uff1a\u9700\u8981\u8f6c\u6362<\/span><\/span><\/strong><\/p>\n
                              \n
                                \n
                              • \n
                              • \n
                              • \n
                              • \n
                              • <\/ul>\n
                                pip<\/span>\u00a0install numpy sentencepiece transformers\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 -i https:\/\/mirrors.huaweicloud.com\/repository\/pypi\/simple<\/span><\/code>python3<\/span>\u00a0\/workspace\/home\/llama.cpp\/convert_hf_to_gguf.py \/models\/Qwen3-0<\/span>.6<\/span>B\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 --outfile \/models\/Qwen3-0<\/span>.6<\/span>B\/Qwen3-0<\/span>.6<\/span>B-f16.gguf\u00a0<\/span><\/span><\/code>\u00a0 \u00a0 --outtype f16<\/span><\/code><\/pre>\n<\/section>\n
                                \n
                                \n
                                \u6ce8\u610f\uff1a<\/span><\/span><\/span><\/strong>CANN9.0.0 \u540e\u7aef\u76ee\u524d\u652f\u6301 f16 \/ q8_0 \/ q4_0 \u91cf\u5316\u7c7b\u578b\uff0c\u5176\u4ed6\u91cf\u5316\u683c\u5f0f\u9700\u8981\u81ea\u884c\u6d4b\u8bd5\uff0c\u4e0d\u4e00\u5b9a\u80fd\u7528\u3002<\/span><\/span><\/span><\/em><\/span><\/p>\n<\/section>\n<\/section>\n

                                <\/span><\/p>\n
                                \u258fStep 4\uff1a\u542f\u52a8\u63a8\u7406\u670d\u52a1<\/span><\/strong><\/span><\/h3>\n

                                <\/span><\/p>\n
                                \ud83d\udce6<\/span>\u00a0\u5bb9\u5668\u5185\u6267\u884c<\/span><\/strong><\/span><\/p>\n

                                <\/span><\/p>\n
                                \u542f\u52a8 llama.cpp \u63a8\u7406\u670d\u52a1\uff0c\u540e\u53f0\u8fd0\u884c\u5e76\u5c06\u65e5\u5fd7\u5199\u5165\u6587\u4ef6\uff1a<\/span><\/span><\/p>\n
                                \n
                                  \n
                                • \n
                                • \n
                                • \n
                                • \n
                                • \n
                                • \n
                                • \n
                                • \n
                                • \n
                                • \n
                                • <\/ul>\n
                                  source<\/span>\u00a0\/usr\/local\/Ascend\/ascend-toolkit\/set_env.sh<\/span><\/span><\/code>
                                  <\/span><\/code>\/workspace\/home\/llama.cpp<\/span>\/build\/bin\/llama-server <\/span><\/code>\u00a0 \u00a0\u00a0--model<\/span>\u00a0\/models\/unsloth\/Qwen3.5<\/span>-9<\/span>B-GGUF\/Qwen3.5<\/span>-9<\/span>B-Q8_0.gguf<\/span>\u00a0<\/span><\/code>\u00a0 \u00a0\u00a0--host<\/span>\u00a00.0<\/span>.0.0<\/span>\u00a0<\/span><\/code>\u00a0 \u00a0\u00a0--port<\/span>\u00a08001<\/span>\u00a0<\/span><\/code>\u00a0 \u00a0\u00a0--n-gpu-layers<\/span>\u00a099<\/span>\u00a0<\/span><\/code>\u00a0 \u00a0\u00a0--split-mode<\/span>\u00a0none<\/span>\u00a0<\/span><\/code>\u00a0 \u00a0\u00a0--ctx-size<\/span>\u00a08192<\/span>\u00a0<\/span><\/code>\u00a0 \u00a0\u00a0--alias<\/span>\u00a0Qwen3.5<\/span>-9<\/span>B <\/span><\/code>\u00a0 \u00a0 > \/workspace\/logs\/llama_server.log<\/span>\u00a02<\/span>>&1<\/span>\u00a0&<\/span><\/code><\/pre>\n<\/section>\n
                                  \n
                                  \n
                                  \n
                                  \n
                                  \n
                                  >>\u53c2\u6570\u8bf4\u660e\uff1a<\/span><\/strong><\/p>\n
                                  \n
                                  \n
                                  \n
                                    \n
                                  • \n
                                    --n-gpu-layers 99<\/span><\/code>\uff1a\u5c06\u5168\u90e8\u5c42\u653e\u5230 NPU \u4e0a\uff0899 \u8868\u793a\u201c\u5168\u90e8\u201d\uff09<\/span><\/p>\n<\/li>\n
                                  • \n
                                    --split-mode layer<\/span><\/code>\uff1a\u591a\u5361\u65f6\u6309\u5c42\u5207\u5206\uff1b\u5355\u5361\u573a\u666f\u53ef\u6539\u4e3a\u00a0<\/span>none<\/span><\/p>\n<\/li>\n
                                  • \n
                                    --ctx-size 8192<\/span><\/code>\uff1a\u4e0a\u4e0b\u6587\u7a97\u53e3\u5927\u5c0f<\/span><\/p>\n<\/li>\n<\/ul>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

                                    <\/span><\/p>\n
                                    \u258fStep 5\uff1a\u9a8c\u8bc1\u670d\u52a1<\/span><\/strong><\/span><\/h3>\n

                                    <\/span><\/p>\n
                                    5.1 \u7b49\u5f85\u5c31\u7eea<\/span><\/span><\/strong><\/p>\n
                                    \n
                                      \n
                                    • \n
                                    • \n
                                    • \n
                                    • \n
                                    • \n
                                    • <\/ul>\n
                                      while<\/span>\u00a0! curl -fsS http:\/\/127.0.0.1:8001\/health 2>\/dev\/null;\u00a0do<\/span><\/span><\/code>\u00a0 \u00a0\u00a0echo<\/span>\u00a0\"\u7b49\u5f85\u670d\u52a1\u5c31\u7eea...\"<\/span><\/span><\/code>\u00a0 \u00a0\u00a0sleep<\/span>\u00a05<\/span><\/code>done<\/span><\/span><\/code>echo<\/span>\u00a0\"\u670d\u52a1\u5df2\u5c31\u7eea\uff01\"<\/span><\/span><\/code>curl -s http:\/\/127.0.0.1:8001\/v1\/models | python -m json.tool<\/span><\/code><\/pre>\n<\/section>\n

                                      <\/span><\/p>\n
                                      5.2 curl \u63a8\u7406\u9a8c\u8bc1<\/span><\/span><\/strong><\/p>\n

                                      <\/span><\/p>\n
                                      \u53d1\u9001\u4e00\u6761\u6d4b\u8bd5\u8bf7\u6c42\uff0c\u9a8c\u8bc1\u63a8\u7406\u670d\u52a1\u662f\u5426\u6b63\u5e38\u54cd\u5e94\uff1a<\/span><\/span><\/p>\n
                                      \n
                                        \n
                                      • \n
                                      • \n
                                      • \n
                                      • \n
                                      • \n
                                      • \n
                                      • \n
                                      • \n
                                      • <\/ul>\n
                                        curl -sS http:\/\/127.0.0.1:8001\/v1\/chat\/completions <\/span><\/span><\/code>\u00a0 \u00a0 -H \"Content-Type: application\/json\" <\/span><\/code>\u00a0 \u00a0 -d '{<\/span><\/code>\u00a0 \u00a0 \u00a0 \u00a0\u00a0\"model\"<\/span>:\u00a0\"Qwen3-0.6B\"<\/span>,<\/span><\/code>\u00a0 \u00a0 \u00a0 \u00a0\u00a0\"messages\"<\/span>: [{\"role\"<\/span>:\u00a0\"user\"<\/span>,\u00a0\"content\"<\/span>:\u00a0\"\u8bf7\u7528\u4e00\u53e5\u8bdd\u4ecb\u7ecd\u534e\u4e3a\u6607\u817e\u3002\"<\/span>}],<\/span><\/code>\u00a0 \u00a0 \u00a0 \u00a0\u00a0\"max_tokens\"<\/span>:\u00a0256<\/span>,<\/span><\/code>\u00a0 \u00a0 \u00a0 \u00a0\u00a0\"temperature\"<\/span>:\u00a00<\/span>,<\/span><\/code>\u00a0 \u00a0 \u00a0 \u00a0\u00a0\"chat_template_kwargs\"<\/span>: {\"enable_thinking\"<\/span>:\u00a0false<\/span>}<\/span><\/code>\u00a0 \u00a0 }'<\/span><\/code><\/pre>\n<\/section>\n

                                        <\/span><\/p>\n
                                        \u63a8\u7406\u9a8c\u8bc1\u8fd4\u56de\u7ed3\u679c\u5982\u4e0b\uff1a<\/span><\/span><\/p>\n
                                        <\/section>\n

                                        <\/span><\/p>\n
                                        \u258fStep 6\uff1a\u8fd0\u884c Benchmark<\/span><\/strong><\/span><\/h3>\n

                                        <\/span><\/p>\n
                                        \ud83d\udce6\u00a0\u5bb9\u5668\u5185\u6267\u884c<\/span><\/span><\/span><\/p>\n

                                        <\/span><\/span><\/p>\n
                                        llama.cpp \u4f7f\u7528\u5185\u7f6e\u7684\u00a0<\/span>llama-bench<\/span><\/span>\u00a0\u5de5\u5177\uff0c\u8f93\u51fa pp\uff08prefill \u541e\u5410\uff09\u548c tg\uff08decode \u901f\u5ea6\uff09\uff0c\u7528\u4e8e\u6027\u80fd\u6d4b\u8bd5\uff1a<\/span><\/span><\/p>\n
                                        \n
                                          \n
                                        • \n
                                        • \n
                                        • \n
                                        • \n
                                        • \n
                                        • \n
                                        • \n
                                        • \n
                                        • \n
                                        • <\/ul>\n
                                          source\u00a0\/usr\/<\/span>local\/Ascend\/<\/span>ascend-<\/span>toolkit\/<\/span>set_env.sh<\/span><\/code>
                                          <\/span><\/code>\/workspace\/<\/span>home\/llama.cpp\/<\/span>build\/bin\/<\/span>llama-<\/span>bench <\/span><\/code>\u00a0 \u00a0\u00a0-<\/span>m\u00a0\/models\/<\/span>unsloth\/Qwen3.5-9B-GGUF\/<\/span>Qwen3<\/span>.5<\/span>-<\/span>9B-<\/span>Q8_0<\/span>.gguf <\/span><\/code>\u00a0 \u00a0\u00a0-<\/span>n\u00a0512<\/span>\u00a0<\/span><\/code>\u00a0 \u00a0\u00a0-<\/span>p\u00a02048<\/span>\u00a0<\/span><\/code>\u00a0 \u00a0\u00a0-<\/span>r\u00a03<\/span>\u00a0<\/span><\/code>\u00a0 \u00a0\u00a0-<\/span>ngl\u00a099<\/span>\u00a0<\/span><\/code>\u00a0 \u00a0\u00a0-<\/span>sm none <\/span><\/code>\u00a0 \u00a0\u00a0|<\/span>\u00a0tee\u00a0\/workspace\/<\/span>logs\/<\/span>llama_cpp_benchmark.txt<\/span><\/code><\/pre>\n<\/section>\n
                                          \n
                                          \n
                                          \n
                                          \n
                                          \n
                                          >>\u53c2\u6570\u8bf4\u660e\uff1a<\/span><\/strong><\/p>\n
                                          \n
                                          \n
                                          \n