{"id":41476,"date":"2025-08-20T23:47:12","date_gmt":"2025-08-20T15:47:12","guid":{"rendered":"https:\/\/www.kdjingpai.com\/?p=41476"},"modified":"2025-08-20T23:47:33","modified_gmt":"2025-08-20T15:47:33","slug":"vllm-cli","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/pt\/vllm-cli\/","title":{"rendered":"vLLM CLI\uff1a\u4f7f\u7528 vLLM \u90e8\u7f72\u5927\u8bed\u8a00\u6a21\u578b\u7684\u547d\u4ee4\u884c\u5de5\u5177"},"content":{"rendered":"<p>vllm-cli \u662f\u4e00\u4e2a\u4e3a <a href=\"https:\/\/www.kdjingpai.com\/ja\/vllm\/\">vLLM<\/a> \u8bbe\u8ba1\u7684\u547d\u4ee4\u884c\u754c\u9762\u5de5\u5177\uff0c\u5b83\u8ba9\u90e8\u7f72\u548c\u7ba1\u7406\u5927\u8bed\u8a00\u6a21\u578b\u53d8\u5f97\u66f4\u52a0\u7b80\u5355\u3002\u8fd9\u4e2a\u5de5\u5177\u540c\u65f6\u63d0\u4f9b\u4e86\u4ea4\u4e92\u5f0f\u83dc\u5355\u754c\u9762\u548c\u4f20\u7edf\u7684\u547d\u4ee4\u884c\u6a21\u5f0f\u3002\u7528\u6237\u53ef\u4ee5\u901a\u8fc7\u5b83\u7ba1\u7406\u672c\u5730\u548c\u8fdc\u7a0b\u6a21\u578b\u3001\u4f7f\u7528\u9884\u8bbe\u6216\u81ea\u5b9a\u4e49\u7684\u914d\u7f6e\u65b9\u6848\u3001\u5e76\u5b9e\u65f6\u76d1\u63a7\u6a21\u578b\u670d\u52a1\u5668\u7684\u8fd0\u884c\u72b6\u6001\u3002\u5bf9\u4e8e\u9700\u8981\u5728\u672c\u5730\u5feb\u901f\u6d4b\u8bd5\u4e0d\u540c\u6a21\u578b\uff0c\u6216\u8005\u5c06\u6a21\u578b\u670d\u52a1\u96c6\u6210\u5230\u81ea\u52a8\u5316\u811a\u672c\u4e2d\u7684\u5f00\u53d1\u8005\u6765\u8bf4\uff0cvllm-cli \u63d0\u4f9b\u4e86\u4e00\u4e2a\u9ad8\u6548\u4e14\u6613\u4e8e\u64cd\u4f5c\u7684\u89e3\u51b3\u65b9\u6848\u3002\u5b83\u8fd8\u5185\u7f6e\u4e86\u7cfb\u7edf\u4fe1\u606f\u68c0\u67e5\u548c\u65e5\u5fd7\u67e5\u770b\u529f\u80fd\uff0c\u5e2e\u52a9\u7528\u6237\u5728\u9047\u5230\u95ee\u9898\u65f6\u5feb\u901f\u5b9a\u4f4d\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-41487\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/08\/76f8ccf34b74014-scaled.png\" alt=\"\" width=\"2560\" height=\"1119\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/08\/76f8ccf34b74014-scaled.png 2560w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/08\/76f8ccf34b74014-1536x672.png 1536w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/08\/76f8ccf34b74014-2048x895.png 2048w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/08\/76f8ccf34b74014-18x8.png 18w\" sizes=\"auto, (max-width: 2560px) 100vw, 2560px\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li><strong>\u4ea4\u4e92\u6a21\u5f0f<\/strong>: \u63d0\u4f9b\u4e00\u4e2a\u529f\u80fd\u4e30\u5bcc\u7684\u7ec8\u7aef\u754c\u9762\uff0c\u7528\u6237\u53ef\u4ee5\u901a\u8fc7\u83dc\u5355\u5bfc\u822a\u6765\u64cd\u4f5c\uff0c\u964d\u4f4e\u4e86\u4f7f\u7528\u95e8\u69db\u3002<\/li>\n<li><strong>\u547d\u4ee4\u884c\u6a21\u5f0f<\/strong>: \u652f\u6301\u76f4\u63a5\u7684\u547d\u4ee4\u884c\u6307\u4ee4\uff0c\u65b9\u4fbf\u96c6\u6210\u5230\u81ea\u52a8\u5316\u811a\u672c\u548c\u5de5\u4f5c\u6d41\u4e2d\u3002<\/li>\n<li><strong>\u6a21\u578b\u7ba1\u7406<\/strong>: \u80fd\u591f\u81ea\u52a8\u53d1\u73b0\u5e76\u7ba1\u7406\u5b58\u50a8\u5728\u672c\u5730\u7684\u6a21\u578b\u6587\u4ef6\u3002<\/li>\n<li><strong>\u8fdc\u7a0b\u6a21\u578b\u652f\u6301<\/strong>: \u65e0\u9700\u9884\u5148\u4e0b\u8f7d\uff0c\u53ef\u4ee5\u76f4\u63a5\u4ece HuggingFace Hub \u52a0\u8f7d\u5e76\u8fd0\u884c\u6a21\u578b\u3002<\/li>\n<li><strong>\u914d\u7f6e\u65b9\u6848<\/strong>: \u5185\u7f6e\u4e86\u591a\u79cd\u9488\u5bf9\u4e0d\u540c\u573a\u666f\uff08\u5982\u9ad8\u541e\u5410\u91cf\u3001\u4f4e\u5185\u5b58\uff09\u4f18\u5316\u7684\u914d\u7f6e\uff0c\u540c\u65f6\u4e5f\u652f\u6301\u7528\u6237\u81ea\u5b9a\u4e49\u914d\u7f6e\u3002<\/li>\n<li><strong>\u670d\u52a1\u5668\u76d1\u63a7<\/strong>: \u53ef\u4ee5\u5b9e\u65f6\u67e5\u770b vLLM \u670d\u52a1\u5668\u7684\u72b6\u6001\uff0c\u5305\u62ec GPU \u4f7f\u7528\u7387\u548c\u65e5\u5fd7\u4fe1\u606f\u3002<\/li>\n<li><strong>\u7cfb\u7edf\u4fe1\u606f<\/strong>: \u68c0\u67e5\u5e76\u663e\u793a GPU\u3001\u5185\u5b58\u548c CUDA \u7684\u517c\u5bb9\u6027\u60c5\u51b5\u3002<\/li>\n<li><strong>\u65e5\u5fd7\u67e5\u770b\u5668<\/strong>: \u5f53\u670d\u52a1\u5668\u542f\u52a8\u5931\u8d25\u65f6\uff0c\u53ef\u4ee5\u65b9\u4fbf\u5730\u67e5\u770b\u5b8c\u6574\u7684\u65e5\u5fd7\u6587\u4ef6\u4ee5\u6392\u67e5\u9519\u8bef\u3002<\/li>\n<li><strong>LoRA \u652f\u6301<\/strong>: \u5141\u8bb8\u5728\u52a0\u8f7d\u57fa\u7840\u6a21\u578b\u7684\u540c\u65f6\uff0c\u6302\u8f7d\u4e00\u4e2a\u6216\u591a\u4e2a LoRA \u9002\u914d\u5668\u3002<\/li>\n<\/ul>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<p>vllm-cli \u65e8\u5728\u7b80\u5316\u4f7f\u7528 vLLM \u90e8\u7f72\u5927\u8bed\u8a00\u6a21\u578b\u7684\u6d41\u7a0b\u3002\u4e0b\u9762\u662f\u8be6\u7ec6\u7684\u5b89\u88c5\u548c\u4f7f\u7528\u6b65\u9aa4\uff0c\u5e2e\u52a9\u4f60\u5feb\u901f\u4e0a\u624b\u3002<\/p>\n<h3><strong>1. \u5b89\u88c5<\/strong><\/h3>\n<p><strong>\u5148\u51b3\u6761\u4ef6<\/strong><br \/>\n\u5728\u5b89\u88c5\u4e4b\u524d\uff0c\u8bf7\u786e\u4fdd\u4f60\u7684\u7cfb\u7edf\u6ee1\u8db3\u4ee5\u4e0b\u6761\u4ef6\uff1a<\/p>\n<ul>\n<li>Python 3.11 \u6216\u66f4\u9ad8\u7248\u672c\u3002<\/li>\n<li>\u4e00\u5757\u652f\u6301 CUDA \u7684 NVIDIA GPU\u3002<\/li>\n<li>\u5df2\u7ecf\u5b89\u88c5\u4e86 vLLM \u6838\u5fc3\u5305\u3002<\/li>\n<\/ul>\n<p><strong>\u4ece PyPI \u5b89\u88c5<\/strong><br \/>\n\u6700\u7b80\u5355\u7684\u5b89\u88c5\u65b9\u5f0f\u662f\u901a\u8fc7 pip \u4ece PyPI \u5b98\u65b9\u4ed3\u5e93\u5b89\u88c5\uff1a<\/p>\n<pre><code>pip install vllm-cli\r\n<\/code><\/pre>\n<p><strong>\u4ece\u6e90\u7801\u7f16\u8bd1\u5b89\u88c5<\/strong><br \/>\n\u5982\u679c\u4f60\u60f3\u4f53\u9a8c\u6700\u65b0\u7684\u529f\u80fd\uff0c\u53ef\u4ee5\u9009\u62e9\u4ece GitHub \u6e90\u7801\u7f16\u8bd1\u5b89\u88c5\u3002<br \/>\n\u9996\u5148\uff0c\u514b\u9686\u9879\u76ee\u4ed3\u5e93\u5230\u672c\u5730\uff1a<\/p>\n<pre><code>git clone https:\/\/github.com\/Chen-zexi\/vllm-cli.git\r\ncd vllm-cli\r\n<\/code><\/pre>\n<p>\u7136\u540e\uff0c\u5b89\u88c5\u6240\u9700\u7684\u4f9d\u8d56\u5e93\u3002\u5efa\u8bae\u5728\u4e00\u4e2a\u5e72\u51c0\u7684\u865a\u62df\u73af\u5883\u91cc\u6267\u884c\u8fd9\u4e9b\u64cd\u4f5c\u3002<\/p>\n<pre><code># \u5b89\u88c5\u4f9d\u8d56\r\npip install -r requirements.txt\r\npip install hf-model-tool\r\n# \u4ee5\u5f00\u53d1\u8005\u6a21\u5f0f\u5b89\u88c5\r\npip install -e .\r\n<\/code><\/pre>\n<h3><strong>2. \u4f7f\u7528\u65b9\u6cd5<\/strong><\/h3>\n<p>vllm-cli \u63d0\u4f9b\u4e86\u4e24\u79cd\u64cd\u4f5c\u6a21\u5f0f\uff1a\u4ea4\u4e92\u5f0f\u754c\u9762\u548c\u547d\u4ee4\u884c\u6307\u4ee4\u3002<\/p>\n<h4><strong>\u4ea4\u4e92\u6a21\u5f0f<\/strong><\/h4>\n<p>\u8fd9\u662f\u6700\u9002\u5408\u521d\u5b66\u8005\u4e0a\u624b\u7684\u65b9\u5f0f\u3002\u76f4\u63a5\u5728\u7ec8\u7aef\u8f93\u5165\u4ee5\u4e0b\u547d\u4ee4\u542f\u52a8\uff1a<\/p>\n<pre><code>vllm-cli\r\n<\/code><\/pre>\n<p>\u542f\u52a8\u540e\uff0c\u4f60\u4f1a\u770b\u5230\u4e00\u4e2a\u6b22\u8fce\u754c\u9762\uff0c\u5176\u4e2d\u5305\u542b\u83dc\u5355\u9a71\u52a8\u7684\u9009\u9879\uff0c\u5f15\u5bfc\u4f60\u5b8c\u6210\u6a21\u578b\u9009\u62e9\u3001\u914d\u7f6e\u548c\u670d\u52a1\u542f\u52a8\u7b49\u6240\u6709\u6b65\u9aa4\u3002<\/p>\n<ul>\n<li><strong>\u6a21\u578b\u9009\u62e9<\/strong>: \u754c\u9762\u4f1a\u5217\u51fa\u672c\u5730\u53d1\u73b0\u7684\u6a21\u578b\u548c HuggingFace Hub \u4e0a\u7684\u8fdc\u7a0b\u6a21\u578b\u3002\u4f60\u53ef\u4ee5\u76f4\u63a5\u9009\u62e9\u4e00\u4e2a\u8fdb\u884c\u90e8\u7f72\u3002<\/li>\n<li><strong>\u5feb\u901f\u542f\u52a8<\/strong>: \u5982\u679c\u4f60\u4e4b\u524d\u5df2\u7ecf\u6210\u529f\u8fd0\u884c\u8fc7\u4e00\u6b21\uff0c\u6b64\u529f\u80fd\u4f1a\u81ea\u52a8\u52a0\u8f7d\u4e0a\u6b21\u7684\u914d\u7f6e\uff0c\u5b9e\u73b0\u4e00\u952e\u542f\u52a8\u3002<\/li>\n<li><strong>\u81ea\u5b9a\u4e49\u914d\u7f6e<\/strong>: \u8fdb\u5165\u9ad8\u7ea7\u914d\u7f6e\u83dc\u5355\uff0c\u4f60\u53ef\u4ee5\u8c03\u6574\u5305\u62ec\u91cf\u5316\u65b9\u5f0f\u3001\u5f20\u91cf\u5e76\u884c\u5927\u5c0f\u7b49\u5728\u5185\u7684\u6570\u5341\u79cd vLLM \u53c2\u6570\u3002<\/li>\n<li><strong>\u670d\u52a1\u5668\u76d1\u63a7<\/strong>: \u670d\u52a1\u542f\u52a8\u540e\uff0c\u53ef\u4ee5\u5728\u76d1\u63a7\u754c\u9762\u770b\u5230\u5b9e\u65f6\u7684 GPU \u5229\u7528\u7387\u3001\u670d\u52a1\u5668\u72b6\u6001\u548c\u65e5\u5fd7\u6d41\u3002<\/li>\n<\/ul>\n<h4><strong>\u547d\u4ee4\u884c\u6a21\u5f0f<\/strong><\/h4>\n<p>\u547d\u4ee4\u884c\u6a21\u5f0f\u9002\u5408\u81ea\u52a8\u5316\u811a\u672c\u548c\u9ad8\u7ea7\u7528\u6237\u3002\u6838\u5fc3\u547d\u4ee4\u662f\u00a0<code>serve<\/code>\u3002<\/p>\n<p><strong>\u57fa\u672c\u7528\u6cd5<\/strong><br \/>\n\u4f7f\u7528\u9ed8\u8ba4\u914d\u7f6e\u542f\u52a8\u4e00\u4e2a\u6a21\u578b\u670d\u52a1\uff1a<\/p>\n<pre><code>vllm-cli serve &lt;MODEL_NAME&gt;\r\n<\/code><\/pre>\n<p>\u5176\u4e2d\u00a0<code>&lt;MODEL_NAME&gt;<\/code>\u00a0\u662f\u6a21\u578b\u7684\u540d\u79f0\uff0c\u4f8b\u5982\u00a0<code>Qwen\/Qwen2-1.5B-Instruct<\/code>\u3002<\/p>\n<p><strong>\u4f7f\u7528\u9884\u8bbe\u914d\u7f6e<\/strong><br \/>\n\u4f60\u53ef\u4ee5\u4f7f\u7528\u00a0<code>--profile<\/code>\u00a0\u53c2\u6570\u6765\u6307\u5b9a\u4e00\u4e2a\u5185\u7f6e\u7684\u4f18\u5316\u914d\u7f6e\u3002\u4f8b\u5982\uff0c\u4f7f\u7528\u00a0<code>high_throughput<\/code>\u00a0\u914d\u7f6e\u6765\u83b7\u5f97\u6700\u5927\u6027\u80fd\uff1a<\/p>\n<pre><code>vllm-cli serve &lt;MODEL_NAME&gt; --profile high_throughput\r\n```\u5185\u7f6e\u7684\u914d\u7f6e\u65b9\u6848\u5305\u62ec\uff1a\r\n- `standard`: \u667a\u80fd\u9ed8\u8ba4\u503c\u7684\u6700\u5c0f\u5316\u914d\u7f6e\u3002\r\n- `moe_optimized`: \u4e3a MoE\uff08\u6df7\u5408\u4e13\u5bb6\uff09\u6a21\u578b\u4f18\u5316\u3002\r\n- `high_throughput`: \u8ffd\u6c42\u6700\u5927\u8bf7\u6c42\u541e\u5410\u91cf\u7684\u6027\u80fd\u914d\u7f6e\u3002\r\n- `low_memory`: \u9002\u7528\u4e8e\u5185\u5b58\u53d7\u9650\u73af\u5883\u7684\u914d\u7f6e\uff0c\u4f8b\u5982\u542f\u7528 FP8 \u91cf\u5316\u3002\r\n**\u4f20\u9012\u81ea\u5b9a\u4e49\u53c2\u6570**\r\n\u4f60\u4e5f\u53ef\u4ee5\u76f4\u63a5\u5728\u547d\u4ee4\u884c\u4e2d\u4f20\u9012\u4efb\u610f vLLM \u652f\u6301\u7684\u53c2\u6570\u3002\u4f8b\u5982\uff0c\u540c\u65f6\u6307\u5b9a AWQ \u91cf\u5316\u548c\u5f20\u91cf\u5e76\u884c\u6570\u4e3a2\uff1a\r\n```bash\r\nvllm-cli serve &lt;MODEL_NAME&gt; --quantization awq --tensor-parallel-size 2\r\n<\/code><\/pre>\n<p><strong>\u5176\u4ed6\u5e38\u7528\u547d\u4ee4<\/strong><\/p>\n<ul>\n<li><strong>\u5217\u51fa\u53ef\u7528\u6a21\u578b<\/strong>:\n<pre><code>vllm-cli models\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u663e\u793a\u7cfb\u7edf\u4fe1\u606f<\/strong>:\n<pre><code>vllm-cli info\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u68c0\u67e5\u6b63\u5728\u8fd0\u884c\u7684\u670d\u52a1<\/strong>:\n<pre><code>vllm-cli status\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u505c\u6b62\u4e00\u4e2a\u670d\u52a1<\/strong>\u00a0(\u9700\u8981\u6307\u5b9a\u7aef\u53e3\u53f7):\n<pre><code>vllm-cli stop --port 8000\r\n<\/code><\/pre>\n<\/li>\n<\/ul>\n<h3><strong>3. \u914d\u7f6e\u6587\u4ef6<\/strong><\/h3>\n<p>vllm-cli \u7684\u914d\u7f6e\u6587\u4ef6\u5b58\u50a8\u5728\u7528\u6237\u76ee\u5f55\u4e0b\u7684\u00a0<code>~\/.config\/vllm-cli\/<\/code>\u00a0\u4e2d\u3002<\/p>\n<ul>\n<li><code>config.yaml<\/code>: \u4e3b\u914d\u7f6e\u6587\u4ef6\u3002<\/li>\n<li><code>user_profiles.json<\/code>: \u7528\u6237\u81ea\u5b9a\u4e49\u7684\u914d\u7f6e\u65b9\u6848\u3002<\/li>\n<li><code>cache.json<\/code>: \u7528\u4e8e\u7f13\u5b58\u6a21\u578b\u5217\u8868\u548c\u7cfb\u7edf\u4fe1\u606f\uff0c\u4ee5\u63d0\u9ad8\u6027\u80fd\u3002<\/li>\n<\/ul>\n<p>\u5f53\u9047\u5230\u6a21\u578b\u52a0\u8f7d\u5931\u8d25\u7b49\u95ee\u9898\u65f6\uff0c\u5de5\u5177\u4f1a\u63d0\u4f9b\u9009\u9879\u8ba9\u4f60\u76f4\u63a5\u67e5\u770b\u65e5\u5fd7\uff0c\u8fd9\u5bf9\u4e8e\u8c03\u8bd5\u975e\u5e38\u6709\u7528\u3002<\/p>\n<h2>\u5e94\u7528\u573a\u666f<\/h2>\n<ol>\n<li><strong>\u672c\u5730\u5f00\u53d1\u4e0e\u6a21\u578b\u8bc4\u4f30<\/strong><br \/>\n\u7814\u7a76\u4eba\u5458\u548c\u5f00\u53d1\u8005\u53ef\u4ee5\u5728\u672c\u5730\u73af\u5883\u4e2d\u5feb\u901f\u90e8\u7f72\u548c\u5207\u6362\u4e0d\u540c\u7684\u5927\u8bed\u8a00\u6a21\u578b\uff0c\u7528\u4e8e\u7b97\u6cd5\u9a8c\u8bc1\u3001\u529f\u80fd\u6d4b\u8bd5\u548c\u6027\u80fd\u8bc4\u4f30\uff0c\u800c\u65e0\u9700\u7f16\u5199\u590d\u6742\u7684\u670d\u52a1\u5668\u90e8\u7f72\u4ee3\u7801\u3002<\/li>\n<li><strong>\u81ea\u52a8\u5316\u90e8\u7f72\u811a\u672c<\/strong><br \/>\n\u5229\u7528\u5176\u547d\u4ee4\u884c\u6a21\u5f0f\uff0c\u53ef\u4ee5\u5c06 vllm-cli \u96c6\u6210\u5230 CI\/CD \u6d41\u7a0b\u6216\u81ea\u52a8\u5316\u8fd0\u7ef4\u811a\u672c\u4e2d\u3002\u4f8b\u5982\uff0c\u5f53\u65b0\u6a21\u578b\u8bad\u7ec3\u5b8c\u6210\u540e\uff0c\u53ef\u4ee5\u81ea\u52a8\u89e6\u53d1\u4e00\u4e2a\u811a\u672c\u6765\u90e8\u7f72\u8be5\u6a21\u578b\u5e76\u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5\u3002<\/li>\n<li><strong>\u6559\u5b66\u4e0e\u6f14\u793a<\/strong><br \/>\n\u5728\u6559\u5b66\u6216\u4ea7\u54c1\u6f14\u793a\u573a\u666f\u4e2d\uff0c\u53ef\u4ee5\u5229\u7528\u4ea4\u4e92\u5f0f\u754c\u9762\u8f7b\u677e\u5730\u542f\u52a8\u4e00\u4e2a\u5927\u8bed\u8a00\u6a21\u578b\u670d\u52a1\uff0c\u5411\u4ed6\u4eba\u76f4\u89c2\u5730\u5c55\u793a\u6a21\u578b\u7684\u6548\u679c\uff0c\u65e0\u9700\u5173\u5fc3\u5e95\u5c42\u590d\u6742\u7684\u914d\u7f6e\u7ec6\u8282\u3002<\/li>\n<li><strong>\u8f7b\u91cf\u7ea7\u5e94\u7528\u540e\u7aef<\/strong><br \/>\n\u5bf9\u4e8e\u4e00\u4e9b\u5185\u90e8\u5de5\u5177\u6216\u8f7b\u91cf\u7ea7\u7684\u5e94\u7528\u7a0b\u5e8f\uff0c\u53ef\u4ee5\u4f7f\u7528 vllm-cli \u5feb\u901f\u642d\u5efa\u4e00\u4e2a\u7a33\u5b9a\u7684\u5927\u8bed\u8a00\u6a21\u578b\u63a8\u7406\u540e\u7aef\uff0c\u6ee1\u8db3\u5c0f\u89c4\u6a21\u7684\u8c03\u7528\u9700\u6c42\u3002<\/li>\n<\/ol>\n<h2>QA<\/h2>\n<ol>\n<li><strong>vllm-cli \u652f\u6301\u54ea\u4e9b\u7c7b\u578b\u7684\u786c\u4ef6\uff1f<\/strong><br \/>\n\u76ee\u524d\uff0cvllm-cli \u4e3b\u8981\u652f\u6301\u642d\u8f7d CUDA \u7684 NVIDIA GPU\u3002\u5bf9 AMD GPU \u7684\u652f\u6301\u4ecd\u5728\u5f00\u53d1\u8def\u7ebf\u56fe\u4e0a\u3002<\/li>\n<li><strong>\u5982\u679c\u4e00\u4e2a\u6a21\u578b\u52a0\u8f7d\u5931\u8d25\u4e86\uff0c\u6211\u5e94\u8be5\u600e\u4e48\u529e\uff1f<\/strong><br \/>\n\u9996\u5148\uff0c\u4f7f\u7528\u5de5\u5177\u63d0\u4f9b\u7684\u65e5\u5fd7\u67e5\u770b\u529f\u80fd\u68c0\u67e5\u8be6\u7ec6\u7684\u9519\u8bef\u4fe1\u606f\uff0c\u8fd9\u901a\u5e38\u4f1a\u6307\u660e\u95ee\u9898\u6240\u5728\u3002\u5176\u6b21\uff0c\u786e\u8ba4\u4f60\u7684 GPU \u578b\u53f7\u548c vLLM \u7248\u672c\u662f\u5426\u4e0e\u8be5\u6a21\u578b\u517c\u5bb9\u3002\u6700\u540e\uff0c\u53ef\u4ee5\u67e5\u9605 vLLM \u7684\u5b98\u65b9\u6587\u6863\uff0c\u4e86\u89e3\u8be5\u6a21\u578b\u662f\u5426\u9700\u8981\u7279\u6b8a\u7684\u542f\u52a8\u53c2\u6570\uff0c\u4f8b\u5982\u7279\u5b9a\u7684\u91cf\u5316\u65b9\u6cd5\u6216\u4fe1\u4efb\u8fdc\u7a0b\u4ee3\u7801\u3002<\/li>\n<li><strong>\u8fd9\u4e2a\u5de5\u5177\u662f\u5982\u4f55\u53d1\u73b0\u6211\u672c\u5730\u7684 HuggingFace \u6a21\u578b\u7684\uff1f<\/strong><br \/>\nvllm-cli \u5185\u90e8\u96c6\u6210\u4e86\u4e00\u4e2a\u540d\u4e3a\u00a0<code>hf-model-tool<\/code>\u00a0\u7684\u8f85\u52a9\u5de5\u5177\u3002\u5b83\u4f1a\u81ea\u52a8\u626b\u63cf HuggingFace \u7684\u9ed8\u8ba4\u7f13\u5b58\u76ee\u5f55\u4ee5\u53ca\u7528\u6237\u624b\u52a8\u914d\u7f6e\u7684\u5176\u4ed6\u6a21\u578b\u76ee\u5f55\uff0c\u6765\u53d1\u73b0\u5e76\u7ba1\u7406\u672c\u5730\u5b58\u50a8\u7684\u6240\u6709\u6a21\u578b\u6587\u4ef6\u3002<\/li>\n<li><strong>\u6211\u53ef\u4ee5\u5728\u6ca1\u6709 GPU \u7684\u60c5\u51b5\u4e0b\u4f7f\u7528\u5b83\u5417\uff1f<\/strong><br \/>\n\u4e0d\u53ef\u4ee5\u3002vllm-cli \u4f9d\u8d56\u4e8e vLLM \u5f15\u64ce\uff0c\u800c vLLM \u672c\u8eab\u7684\u8bbe\u8ba1\u662f\u7528\u4e8e\u5728 GPU \u4e0a\u9ad8\u6548\u8fd0\u884c\u5927\u8bed\u8a00\u6a21\u578b\u7684\uff0c\u56e0\u6b64\u5fc5\u987b\u8981\u6709\u652f\u6301 CUDA \u7684 NVIDIA GPU \u786c\u4ef6\u3002<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>vllm-cli \u662f\u4e00\u4e2a\u4e3a vLLM \u8bbe\u8ba1\u7684\u547d\u4ee4\u884c\u754c\u9762\u5de5\u5177\uff0c\u5b83\u8ba9\u90e8\u7f72\u548c\u7ba1\u7406\u5927\u8bed\u8a00\u6a21\u578b\u53d8\u5f97\u66f4\u52a0\u7b80\u5355\u3002\u8fd9\u4e2a\u5de5\u5177\u540c\u65f6\u63d0\u4f9b\u4e86\u4ea4\u4e92\u5f0f\u83dc\u5355\u754c\u9762\u548c\u4f20\u7edf\u7684\u547d\u4ee4\u884c\u6a21\u5f0f\u3002\u7528\u6237\u53ef\u4ee5\u901a\u8fc7\u5b83\u7ba1\u7406\u672c\u5730\u548c\u8fdc\u7a0b\u6a21\u578b\u3001\u4f7f\u7528\u9884\u8bbe\u6216\u81ea\u5b9a\u4e49\u7684\u914d\u7f6e\u65b9\u6848\u3001\u5e76\u5b9e\u65f6\u76d1\u63a7\u6a21\u578b\u670d\u52a1\u5668\u7684\u8fd0\u884c\u72b6&#8230;<\/p>\n","protected":false},"author":1,"featured_media":62400,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[425,20,407,497],"tags":[],"class_list":["post-41476","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-professional","category-tool","category-pc-client","category-deployment-model"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/41476","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/comments?post=41476"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/41476\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media\/62400"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media?parent=41476"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/categories?post=41476"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/tags?post=41476"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}