{"id":46472,"date":"2025-07-30T03:16:16","date_gmt":"2025-07-29T19:16:16","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/46472.html"},"modified":"2025-07-30T03:16:16","modified_gmt":"2025-07-29T19:16:16","slug":"%e7%ba%a2%e5%b8%bdai%e6%8e%a8%e7%90%86%e6%9c%8d%e5%8a%a1%e5%99%a8%e4%b8%89%e5%a4%a7%e7%89%b9%e7%82%b9","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/46472.html","title":{"rendered":"\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u4e09\u5927\u7279\u70b9"},"content":{"rendered":"<p>\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u80fd&#xff08;Gen AI&#xff09;\u7684\u8fc5\u731b\u53d1\u5c55&#xff0c;\u5bf9\u5927\u578b\u8bed\u8a00\u6a21\u578b&#xff08;LLM&#xff09;\u7684\u90e8\u7f72\u63d0\u51fa\u4e86\u66f4\u9ad8\u7684\u6027\u80fd\u3001\u7075\u6d3b\u6027\u548c\u6548\u7387\u8981\u6c42\u3002<\/p>\n<p>\u65e0\u8bba\u90e8\u7f72\u5728\u4f55\u79cd\u73af\u5883\u4e2d&#xff0c;\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u90fd\u4e3a\u7528\u6237\u63d0\u4f9b\u7ecf\u8fc7\u5f3a\u5316\u5e76\u83b7\u5f97\u5b98\u65b9\u652f\u6301\u7684vLLM\u53d1\u884c\u7248&#xff0c;\u914d\u5957\u667a\u80fdLLM\u538b\u7f29\u5de5\u5177&#xff0c;\u4ee5\u53ca\u5728Hugging Face\u5e73\u53f0\u4e0a\u4f18\u5316\u7684\u6a21\u578b\u4ed3\u5e93\u3002\u7ed3\u5408\u7ea2\u5e3d\u7684\u4f01\u4e1a\u7ea7\u6280\u672f\u652f\u6301\u4e0e\u7075\u6d3b\u7684\u7b2c\u4e09\u65b9\u652f\u6301\u653f\u7b56&#xff0c;\u4e3a\u4f01\u4e1a\u90e8\u7f72\u751f\u6210\u5f0fAI\u5e94\u7528\u63d0\u4f9b\u5f3a\u6709\u529b\u7684\u652f\u6491\u3002<\/p>\n<\/p>\n<p>\u501f\u52a9vLLM\u5185\u6838\u4e0e\u5148\u8fdb\u5e76\u884c\u6280\u672f&#xff0c;\u52a0\u901fAI\u63a8\u7406\u6027\u80fd<\/p>\n<\/p>\n<p>\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u7684\u6838\u5fc3\u662fvLLM\u63a8\u7406\u5f15\u64ce\u3002vLLM\u56e0\u5176\u9ad8\u541e\u5410\u91cf\u4e0e\u5185\u5b58\u6548\u7387\u4f18\u5f02\u800c\u5e7f\u53d7\u8ba4\u53ef&#xff0c;\u6838\u5fc3\u6280\u672f\u5305\u62ec\u6e90\u81ea\u52a0\u5dde\u5927\u5b66\u4f2f\u514b\u5229\u5206\u6821\u7684PagedAttention&#xff08;\u7528\u4e8e\u4f18\u5316GPU\u5185\u5b58\u7ba1\u7406&#xff09;\u4ee5\u53ca\u6301\u7eed\u6279\u5904\u7406&#xff08;Continuous Batching&#xff09;&#xff0c;\u901a\u5e38\u80fd\u5e26\u6765\u6570\u500d\u4e8e\u4f20\u7edf\u63a8\u7406\u65b9\u6cd5\u7684\u6027\u80fd\u63d0\u5347\u3002\u8be5\u670d\u52a1\u5668\u8fd8\u901a\u5e38\u63d0\u4f9b\u4e00\u4e2a\u517c\u5bb9OpenAI\u7684API\u7aef\u70b9&#xff0c;\u4fbf\u4e8e\u5feb\u901f\u96c6\u6210\u3002<\/p>\n<\/p>\n<p>\u4e3a\u5e94\u5bf9\u5f53\u524d\u4f53\u91cf\u5e9e\u5927\u3001\u7ed3\u6784\u590d\u6742\u7684\u751f\u6210\u5f0fAI\u6a21\u578b&#xff0c;vLLM\u878d\u5408\u4e86\u591a\u79cd\u5148\u8fdb\u7684\u63a8\u7406\u4f18\u5316\u6280\u672f&#xff0c;\u5305\u62ec&#xff1a;<\/p>\n<\/p>\n<ul>\n<li>\n<p>\u5f20\u91cf\u5e76\u884c&#xff08;Tensor Parallelism&#xff0c;TP&#xff09;&#xff1a;\u5c06\u5355\u4e2a\u6a21\u578b\u5c42\u62c6\u5206\u5e76\u5e76\u884c\u5206\u5e03\u5230\u540c\u4e00\u8282\u70b9\u5185\u7684\u591a\u4e2aGPU\u4e0a\u6267\u884c&#xff0c;\u4ece\u800c\u964d\u4f4e\u5ef6\u8fdf\u5e76\u63d0\u5347\u8ba1\u7b97\u541e\u5410\u80fd\u529b\u3002<\/p>\n<\/li>\n<li>\n<p>\u6d41\u6c34\u7ebf\u5e76\u884c&#xff08;Pipeline Parallelism&#xff0c;PP&#xff09;&#xff1a;\u5c06\u6a21\u578b\u7684\u4e0d\u540c\u5c42\u5212\u5206\u4e3a\u82e5\u5e72\u9636\u6bb5&#xff0c;\u5206\u522b\u90e8\u7f72\u5728\u4e0d\u540cGPU\u6216\u8282\u70b9\u4e0a&#xff0c;\u9002\u7528\u4e8e\u5355\u4e00\u591aGPU\u8282\u70b9\u4e5f\u65e0\u6cd5\u5bb9\u7eb3\u7684\u8d85\u5927\u6a21\u578b\u3002<\/p>\n<\/li>\n<li>\n<p>\u4e13\u5bb6\u5e76\u884c&#xff08;Expert Parallelism&#xff0c;EP&#xff09;&#xff1a;\u9488\u5bf9\u6df7\u5408\u4e13\u5bb6&#xff08;Mixture of Experts&#xff0c;MoE&#xff09;\u6a21\u578b\u8fdb\u884c\u4e13\u95e8\u4f18\u5316&#xff0c;\u80fd\u591f\u9ad8\u6548\u5904\u7406\u5176\u72ec\u7279\u7684\u8def\u7531\u903b\u8f91\u548c\u8ba1\u7b97\u8d44\u6e90\u5206\u914d\u3002<\/p>\n<\/li>\n<li>\n<p>\u6570\u636e\u5e76\u884c&#xff08;Data Parallelism&#xff0c;DP&#xff09;&#xff1a;\u652f\u6301\u5c06\u4e0d\u540c\u7684\u63a8\u7406\u8bf7\u6c42\u5206\u53d1\u81f3\u591a\u4e2avLLM\u5b9e\u4f8b\u3002\u5728\u8fdb\u5165MoE\u5c42\u65f6&#xff0c;\u5404\u6570\u636e\u5e76\u884c\u5f15\u64ce\u534f\u540c\u5de5\u4f5c&#xff0c;\u5c06\u4e13\u5bb6\u6a21\u5757\u5728\u6240\u6709\u6570\u636e\u5e76\u884c\u4e0e\u5f20\u91cf\u5e76\u884c\u7684\u5de5\u4f5c\u5668\u4e4b\u95f4\u8fdb\u884c\u5207\u5206\u3002\u6b64\u673a\u5236\u7279\u522b\u9002\u7528\u4e8e\u5982DeepSeek V3\u6216Qwen3\u8fd9\u7c7bKV\u6ce8\u610f\u529b\u5934\u8f83\u5c11\u7684\u6a21\u578b&#xff0c;\u53ef\u907f\u514d\u5f20\u91cf\u5e76\u884c\u9020\u6210\u7684KV\u7f13\u5b58\u5197\u4f59&#xff0c;\u63d0\u5347\u8d44\u6e90\u5229\u7528\u7387\u4e0e\u6269\u5c55\u80fd\u529b\u3002<\/p>\n<\/li>\n<li>\n<p>\u91cf\u5316&#xff08;Quantization&#xff09;&#xff1a;AI\u63a8\u7406\u670d\u52a1\u5668\u5185\u7f6e\u7684LLM Compressor\u63d0\u4f9b\u7edf\u4e00\u7684\u6a21\u578b\u538b\u7f29\u5e93&#xff0c;\u652f\u6301\u6743\u91cd&#043;\u6fc0\u6d3b\u91cf\u5316\u6216\u4ec5\u6743\u91cd\u91cf\u5316&#xff0c;\u4ece\u800c\u52a0\u901fvLLM\u63a8\u7406\u6d41\u7a0b\u3002vLLM\u540c\u65f6\u63d0\u4f9b\u81ea\u5b9a\u4e49\u5185\u6838&#xff08;\u5982Marlin\u548cMachete&#xff09;\u4ee5\u8fdb\u4e00\u6b65\u63d0\u5347\u91cf\u5316\u6a21\u578b\u7684\u8fd0\u884c\u6548\u7387\u3002<\/p>\n<\/li>\n<li>\n<p>\u63a8\u6d4b\u89e3\u7801&#xff08;Speculative Decoding&#xff09;&#xff1a;\u901a\u8fc7\u5f15\u5165\u4e00\u4e2a\u5c0f\u578b\u8349\u7a3f\u6a21\u578b\u9884\u6d4b\u591a\u4e2a\u672a\u6765token&#xff0c;\u4e3b\u6a21\u578b\u4ec5\u5bf9\u5176\u8fdb\u884c\u9a8c\u8bc1\u6216\u4fee\u6b63&#xff0c;\u4ece\u800c\u663e\u8457\u964d\u4f4e\u6574\u4f53\u89e3\u7801\u5ef6\u8fdf&#xff0c;\u63d0\u9ad8\u63a8\u7406\u541e\u5410\u91cf&#xff0c;\u540c\u65f6\u4fdd\u6301\u751f\u6210\u8d28\u91cf\u4e0d\u53d7\u5f71\u54cd\u3002<\/p>\n<\/li>\n<\/ul>\n<p>\u503c\u5f97\u4e00\u63d0\u7684\u662f&#xff0c;\u4e0a\u8ff0\u4f18\u5316\u6280\u672f\u901a\u5e38\u53ef\u7075\u6d3b\u7ec4\u5408\u4f7f\u7528&#xff0c;\u4f8b\u5982\u8282\u70b9\u95f4\u5e94\u7528\u6d41\u6c34\u7ebf\u5e76\u884c\u3001\u8282\u70b9\u5185\u5e94\u7528\u5f20\u91cf\u5e76\u884c&#xff0c;\u4ee5\u9002\u5e94\u590d\u6742\u7684\u786c\u4ef6\u62d3\u6251\u7ed3\u6784&#xff0c;\u5728\u5927\u89c4\u6a21\u63a8\u7406\u573a\u666f\u4e2d\u9ad8\u6548\u6269\u5c55LLM\u7684\u8ba1\u7b97\u80fd\u529b\u3002<\/p>\n<\/p>\n<p>\u901a\u8fc7\u5bb9\u5668\u5316\u5b9e\u73b0\u90e8\u7f72\u53ef\u79fb\u690d\u6027<\/p>\n<\/p>\n<p>\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u4ee5\u6807\u51c6\u5bb9\u5668\u955c\u50cf\u5f62\u5f0f\u4ea4\u4ed8&#xff0c;\u5177\u5907\u51fa\u8272\u7684\u90e8\u7f72\u7075\u6d3b\u6027\u3002\u8fd9\u79cd\u5bb9\u5668\u5316\u4ea4\u4ed8\u65b9\u5f0f\u662f\u5b9e\u73b0\u6df7\u5408\u4e91\u73af\u5883\u4e0b\u53ef\u79fb\u690d\u6027\u7684\u6838\u5fc3&#xff0c;\u786e\u4fdd\u65e0\u8bba\u90e8\u7f72\u5728\u7ea2\u5e3dOpenShift\u3001\u7ea2\u5e3d\u4f01\u4e1aLinux&#xff08;RHEL&#xff09;\u3001\u975e\u7ea2\u5e3dKubernetes\u5e73\u53f0&#xff0c;\u8fd8\u662f\u5176\u4ed6\u6807\u51c6Linux\u7cfb\u7edf\u4e0a&#xff0c;\u5747\u53ef\u63d0\u4f9b\u4e00\u81f4\u7684\u63a8\u7406\u8fd0\u884c\u73af\u5883\u3002\u5b83\u4e3a\u5728\u4efb\u610f\u4e1a\u52a1\u573a\u666f\u4e2d\u90e8\u7f72\u5927\u578b\u8bed\u8a00\u6a21\u578b&#xff08;LLM&#xff09;\u5960\u5b9a\u4e86\u6807\u51c6\u5316\u3001\u53ef\u9884\u6d4b\u7684\u57fa\u7840&#xff0c;\u6709\u6548\u7b80\u5316\u4e86\u8de8\u5f02\u6784\u57fa\u7840\u8bbe\u65bd\u7684\u8fd0\u7ef4\u5de5\u4f5c\u3002<\/p>\n<\/p>\n<p>\u591a\u52a0\u901f\u5668\u652f\u6301<\/p>\n<\/p>\n<p>\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u81ea\u8bbe\u8ba1\u4e4b\u521d\u4fbf\u5c06\u591a\u52a0\u901f\u5668\u652f\u6301\u4f5c\u4e3a\u6838\u5fc3\u80fd\u529b&#xff0c;\u80fd\u591f\u65e0\u7f1d\u517c\u5bb9\u591a\u79cd\u786c\u4ef6\u52a0\u901f\u5668&#xff0c;\u5305\u62ecNVIDIA GPU\u3001AMD GPU\u548cGoogle TPU\u3002\u901a\u8fc7\u6784\u5efa\u7edf\u4e00\u7684\u63a8\u7406\u670d\u52a1\u5c42&#xff0c;\u5e73\u53f0\u6709\u6548\u5c4f\u853d\u5e95\u5c42\u786c\u4ef6\u5dee\u5f02&#xff0c;\u5e26\u6765\u6781\u5927\u7684\u7075\u6d3b\u6027\u548c\u4f18\u5316\u7a7a\u95f4\u3002<\/p>\n<\/p>\n<p>\u8fd9\u4e00\u80fd\u529b\u8ba9\u7528\u6237\u80fd\u591f&#xff1a;<\/p>\n<\/p>\n<ul>\n<li>\n<p>\u4f18\u5316\u6027\u80fd\u4e0e\u6210\u672c&#xff1a;\u6839\u636e\u6a21\u578b\u7279\u6027\u3001\u5ef6\u8fdf\u8981\u6c42\u548c\u6210\u672c\u9884\u7b97&#xff0c;\u5728\u6700\u9002\u5408\u7684\u52a0\u901f\u5668\u4e0a\u8fd0\u884c\u63a8\u7406\u4efb\u52a1&#xff0c;\u5b9e\u73b0\u66f4\u9ad8\u6027\u80fd\u548c\u8d44\u6e90\u5229\u7528\u6548\u7387\u3002<\/p>\n<\/li>\n<li>\n<p>\u4fdd\u969c\u672a\u6765\u9002\u5e94\u6027&#xff1a;\u652f\u6301\u65b0\u4e00\u4ee3\u52a0\u901f\u5668\u7684\u65e0\u7f1d\u96c6\u6210&#xff0c;\u65e0\u9700\u4fee\u6539\u57fa\u7840\u67b6\u6784\u6216\u5e94\u7528\u4ee3\u7801&#xff0c;\u786e\u4fdd\u5e73\u53f0\u5177\u5907\u6301\u7eed\u6f14\u8fdb\u80fd\u529b\u3002<\/p>\n<\/li>\n<li>\n<p>\u7075\u6d3b\u6269\u5c55\u63a8\u7406\u80fd\u529b&#xff1a;\u53ef\u6309\u9700\u6dfb\u52a0\u540c\u7c7b\u6216\u5f02\u6784\u52a0\u901f\u5668&#xff0c;\u8f7b\u677e\u5e94\u5bf9\u4e1a\u52a1\u589e\u957f\u548c\u6a21\u578b\u590d\u6742\u5ea6\u63d0\u5347\u3002<\/p>\n<\/li>\n<li>\n<p>\u964d\u4f4e\u5382\u5546\u4f9d\u8d56&#xff1a;\u517c\u5bb9\u591a\u5bb6\u52a0\u901f\u5668\u4f9b\u5e94\u5546&#xff0c;\u907f\u514d\u5bf9\u5355\u4e00\u786c\u4ef6\u5e73\u53f0\u7684\u7ed1\u5b9a&#xff0c;\u589e\u5f3a\u91c7\u8d2d\u7075\u6d3b\u6027\u4e0e\u6210\u672c\u63a7\u5236\u80fd\u529b\u3002<\/p>\n<\/li>\n<li>\n<p>\u7b80\u5316\u8fd0\u7ef4\u7ba1\u7406&#xff1a;\u5728\u4e0d\u540c\u786c\u4ef6\u4e0a\u63d0\u4f9b\u4e00\u81f4\u7684\u7ba1\u7406\u63a5\u53e3&#xff0c;\u663e\u8457\u964d\u4f4e\u63a8\u7406\u670d\u52a1\u5728\u5f02\u6784\u73af\u5883\u4e2d\u7684\u8fd0\u7ef4\u8d1f\u62c5\u3002<\/p>\n<\/li>\n<\/ul>\n<p>\u51ed\u501f\u8fd9\u4e00\u9762\u5411\u672a\u6765\u7684\u67b6\u6784\u8bbe\u8ba1&#xff0c;\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u4e0d\u4ec5\u6ee1\u8db3\u5f53\u524d\u751f\u6210\u5f0fAI\u7684\u9ad8\u6027\u80fd\u63a8\u7406\u9700\u6c42&#xff0c;\u4e5f\u4e3a\u4f01\u4e1a\u6784\u5efa\u53ef\u6301\u7eed\u3001\u53ef\u62d3\u5c55\u7684AI\u57fa\u7840\u8bbe\u65bd\u5960\u5b9a\u575a\u5b9e\u57fa\u7840\u3002<\/p>\n<\/p>\n<p>\u7531\u7ea2\u5e3d\u5185\u90e8Neural Magic\u4e13\u4e1a\u6280\u672f\u9a71\u52a8\u7684\u6a21\u578b\u4f18\u5316<\/p>\n<\/p>\n<p>\u9ad8\u6548\u90e8\u7f72\u5927\u578b\u8bed\u8a00\u6a21\u578b&#xff08;LLM&#xff09;\u901a\u5e38\u9700\u8981\u6a21\u578b\u4f18\u5316\u3002AI\u63a8\u7406\u670d\u52a1\u5668\u96c6\u6210\u4e86\u5f3a\u5927\u7684LLM\u538b\u7f29\u80fd\u529b&#xff0c;\u5229\u7528\u5df2\u52a0\u5165\u7ea2\u5e3d\u7684Neural Magic\u7684\u524d\u6cbf\u4f18\u5316\u6280\u672f\u3002\u901a\u8fc7SparseGPT\u7b49\u4e1a\u754c\u9886\u5148\u7684\u91cf\u5316\u4e0e\u7a00\u758f\u5316\u65b9\u6cd5&#xff0c;\u538b\u7f29\u5668\u53ef\u5728\u51c6\u786e\u7387\u51e0\u4e4e\u65e0\u635f\u7684\u524d\u63d0\u4e0b\u5927\u5e45\u51cf\u5c0f\u6a21\u578b\u4f53\u79ef\u548c\u8ba1\u7b97\u8d1f\u62c5\u3002\u8fd9\u63d0\u5347\u4e86\u63a8\u7406\u901f\u5ea6\u4e0e\u8d44\u6e90\u5229\u7528\u6548\u7387&#xff0c;\u663e\u8457\u964d\u4f4e\u5185\u5b58\u5360\u7528&#xff0c;\u4f7f\u6a21\u578b\u5373\u4f7f\u5728GPU\u8d44\u6e90\u53d7\u9650\u7684\u7cfb\u7edf\u4e2d\u4e5f\u80fd\u987a\u7545\u8fd0\u884c\u3002<\/p>\n<\/p>\n<p>\u901a\u8fc7\u4f18\u5316\u7684\u6a21\u578b\u4ed3\u5e93\u5b9e\u73b0\u7b80\u5316\u8bbf\u95ee<\/p>\n<\/p>\n<p>\u4e3a\u8fdb\u4e00\u6b65\u7b80\u5316\u90e8\u7f72&#xff0c;AI\u63a8\u7406\u670d\u52a1\u5668\u63d0\u4f9b\u5bf9\u4e00\u7cfb\u5217\u4e3b\u6d41LLM&#xff08;\u5982Llama\u3001Mistral\u548cGranite\u7cfb\u5217&#xff09;\u7684\u4f18\u5316\u6a21\u578b\u4ed3\u5e93\u8bbf\u95ee&#xff0c;\u6258\u7ba1\u4e8eHugging Face\u7684\u7ea2\u5e3dAI\u9875\u9762\u3002<\/p>\n<\/p>\n<p>\u8fd9\u4e9b\u6a21\u578b\u5e76\u975e\u666e\u901a\u7248\u672c&#xff0c;\u800c\u662f\u7ecf\u8fc7\u96c6\u6210\u538b\u7f29\u6280\u672f\u4e13\u95e8\u4f18\u5316&#xff0c;\u9002\u914dvLLM\u5f15\u64ce\u7684\u9ad8\u6027\u80fd\u63a8\u7406\u3002\u7528\u6237\u53ef\u76f4\u63a5\u4f7f\u7528\u8fd9\u4e9b\u9ad8\u6548\u6a21\u578b&#xff0c;\u5927\u5e45\u7f29\u77ed\u90e8\u7f72\u65f6\u95f4&#xff0c;\u52a0\u5febAI\u5e94\u7528\u843d\u5730\u3002<\/p>\n<\/p>\n<p>\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u6280\u672f\u6982\u89c8<\/p>\n<\/p>\n<p class=\"img-center\"><img loading=\"lazy\" decoding=\"async\" alt=\"\u56fe\u7247\" height=\"742\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729191614-68891dfeefc4b.png\" width=\"1080\" \/><\/p>\n<\/p>\n<p>vLLM\u67b6\u6784\u65e8\u5728\u6700\u5927\u5316LLM\u63a8\u7406\u7684\u541e\u5410\u91cf\u5e76\u6700\u5c0f\u5316\u5ef6\u8fdf&#xff0c;\u5c24\u5176\u9002\u7528\u4e8e\u9ad8\u5e76\u53d1\u3001\u8bf7\u6c42\u957f\u5ea6\u591a\u6837\u7684\u573a\u666f\u3002\u6838\u5fc3\u7ec4\u4ef6EngineCore\u662f\u4e13\u7528\u63a8\u7406\u5f15\u64ce&#xff0c;\u8d1f\u8d23\u524d\u5411\u8ba1\u7b97\u8c03\u5ea6\u3001\u952e\u503c&#xff08;KV&#xff09;\u7f13\u5b58\u7ba1\u7406\u4ee5\u53ca\u591a\u8bf7\u6c42\u4ee4\u724c\u7684\u52a8\u6001\u6279\u5904\u7406\u3002<\/p>\n<\/p>\n<p>EngineCore\u4e0d\u4ec5\u964d\u4f4e\u4e86\u957f\u4e0a\u4e0b\u6587\u7a97\u53e3\u7ba1\u7406\u7684\u5f00\u9500&#xff0c;\u8fd8\u80fd\u667a\u80fd\u9884\u5904\u7406\u6216\u4ea4\u9519\u5904\u7406\u77ed\u65f6\u5ef6\u8bf7\u6c42\u4e0e\u957f\u4efb\u52a1\u3002\u8fd9\u4f9d\u8d56\u4e8e\u961f\u5217\u8c03\u5ea6\u673a\u5236\u4e0ePagedAttention\u2014\u2014\u4e00\u79cd\u4e3a\u6bcf\u4e2a\u8bf7\u6c42\u865a\u62df\u5316KV\u7f13\u5b58\u7684\u65b0\u65b9\u6cd5\u3002\u5176\u7ed3\u679c\u662f\u66f4\u9ad8\u7684GPU\u5185\u5b58\u5229\u7528\u7387\u4e0e\u66f4\u5c11\u7684\u8ba1\u7b97\u7a7a\u95f2\u65f6\u95f4\u3002<\/p>\n<\/p>\n<p>\u4f5c\u4e3a\u63a5\u53e3\u9002\u914d\u5668&#xff0c;EngineCoreClient\u8d1f\u8d23\u8fde\u63a5API&#xff08;\u5982HTTP\u3001gRPC\u7b49&#xff09;\u5e76\u5c06\u8bf7\u6c42\u8f6c\u53d1\u81f3EngineCore\u3002\u591a\u4e2aEngineCoreClient\u53ef\u4e0e\u4e00\u4e2a\u6216\u591a\u4e2aEngineCore\u901a\u4fe1&#xff0c;\u652f\u6301\u5206\u5e03\u5f0f\u6216\u591a\u8282\u70b9\u90e8\u7f72\u3002vLLM\u5c06\u8bf7\u6c42\u5904\u7406\u4e0e\u5e95\u5c42\u63a8\u7406\u89e3\u8026&#xff0c;\u4fbf\u4e8e\u5b9e\u65bd\u5982\u591aEngineCore\u8d1f\u8f7d\u5747\u8861\u6216\u6839\u636e\u9700\u6c42\u6269\u5c55\u5ba2\u6237\u7aef\u7b49\u7b56\u7565\u3002<\/p>\n<\/p>\n<p>\u8be5\u67b6\u6784\u4e0d\u4ec5\u4fbf\u4e8e\u96c6\u6210\u591a\u79cd\u670d\u52a1\u63a5\u53e3&#xff0c;\u8fd8\u652f\u6301\u53ef\u6269\u5c55\u7684\u5206\u5e03\u5f0f\u90e8\u7f72\u3002EngineCoreClient\u53ef\u5728\u72ec\u7acb\u8fdb\u7a0b\u4e2d\u8fd0\u884c&#xff0c;\u901a\u8fc7\u7f51\u7edc\u8fde\u63a5EngineCore&#xff0c;\u4ece\u800c\u5b9e\u73b0\u8d1f\u8f7d\u5747\u8861\u5e76\u964d\u4f4eCPU\u8d1f\u8f7d\u3002<\/p>\n<p>\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u7ed3\u5408\u9886\u5148\u6027\u80fd\u4e0e\u7075\u6d3b\u90e8\u7f72\u80fd\u529b\u3002\u5176\u5bb9\u5668\u5316\u7279\u6027\u8d4b\u4e88\u771f\u6b63\u7684\u6df7\u5408\u4e91\u7075\u6d3b\u6027&#xff0c;\u652f\u6301\u5728\u4efb\u610f\u6570\u636e\u4e0e\u5e94\u7528\u6240\u5728\u73af\u5883\u4e2d\u4e00\u81f4\u90e8\u7f72\u5148\u8fdbAI\u63a8\u7406&#xff0c;\u6253\u9020\u4f01\u4e1aAI\u8d1f\u8f7d\u7684\u5f3a\u5927\u57fa\u7840\u3002<\/p><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb1.1k\u6b21\uff0c\u70b9\u8d5e11\u6b21\uff0c\u6536\u85cf17\u6b21\u3002\u65e0\u8bba\u90e8\u7f72\u5728\u4f55\u79cd\u73af\u5883\u4e2d\uff0c\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u90fd\u4e3a\u7528\u6237\u63d0\u4f9b\u7ecf\u8fc7\u5f3a\u5316\u5e76\u83b7\u5f97\u5b98\u65b9\u652f\u6301\u7684vLLM\u53d1\u884c\u7248\uff0c\u914d\u5957\u667a\u80fdLLM\u538b\u7f29\u5de5\u5177\uff0c\u4ee5\u53ca\u5728Hugging Face\u5e73\u53f0\u4e0a\u4f18\u5316\u7684\u6a21\u578b\u4ed3\u5e93\u3002\u6d41\u6c34\u7ebf\u5e76\u884c\uff08Pipeline Parallelism\uff0cPP\uff09\uff1a\u5c06\u6a21\u578b\u7684\u4e0d\u540c\u5c42\u5212\u5206\u4e3a\u82e5\u5e72\u9636\u6bb5\uff0c\u5206\u522b\u90e8\u7f72\u5728\u4e0d\u540cGPU\u6216\u8282\u70b9\u4e0a\uff0c\u9002\u7528\u4e8e\u5355\u4e00\u591aGPU\u8282\u70b9\u4e5f\u65e0\u6cd5\u5bb9\u7eb3\u7684\u8d85\u5927\u6a21\u578b\u3002\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0c\u4e0a\u8ff0\u4f18\u5316\u6280\u672f\u901a\u5e38\u53ef\u7075\u6d3b\u7ec4\u5408\u4f7f\u7528\uff0c\u4f8b\u5982\u8282\u70b9\u95f4\u5e94\u7528\u6d41\u6c34\u7ebf\u5e76\u884c\u3001\u8282\u70b9\u5185\u5e94\u7528\u5f20\u91cf\u5e76\u884c\uff0c\u4ee5\u9002\u5e94\u590d\u6742\u7684\u786c\u4ef6\u62d3\u6251\u7ed3\u6784\uff0c\u5728\u5927\u89c4\u6a21\u63a8\u7406\u573a\u666f\u4e2d\u9ad8\u6548\u6269\u5c55LLM\u7684\u8ba1\u7b97\u80fd\u529b\u3002<\/p>\n","protected":false},"author":2,"featured_media":46471,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[2681,50],"topic":[],"class_list":["post-46472","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-vllm","tag-50"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u4e09\u5927\u7279\u70b9 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/46472.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u4e09\u5927\u7279\u70b9 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb1.1k\u6b21\uff0c\u70b9\u8d5e11\u6b21\uff0c\u6536\u85cf17\u6b21\u3002\u65e0\u8bba\u90e8\u7f72\u5728\u4f55\u79cd\u73af\u5883\u4e2d\uff0c\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u90fd\u4e3a\u7528\u6237\u63d0\u4f9b\u7ecf\u8fc7\u5f3a\u5316\u5e76\u83b7\u5f97\u5b98\u65b9\u652f\u6301\u7684vLLM\u53d1\u884c\u7248\uff0c\u914d\u5957\u667a\u80fdLLM\u538b\u7f29\u5de5\u5177\uff0c\u4ee5\u53ca\u5728Hugging Face\u5e73\u53f0\u4e0a\u4f18\u5316\u7684\u6a21\u578b\u4ed3\u5e93\u3002\u6d41\u6c34\u7ebf\u5e76\u884c\uff08Pipeline Parallelism\uff0cPP\uff09\uff1a\u5c06\u6a21\u578b\u7684\u4e0d\u540c\u5c42\u5212\u5206\u4e3a\u82e5\u5e72\u9636\u6bb5\uff0c\u5206\u522b\u90e8\u7f72\u5728\u4e0d\u540cGPU\u6216\u8282\u70b9\u4e0a\uff0c\u9002\u7528\u4e8e\u5355\u4e00\u591aGPU\u8282\u70b9\u4e5f\u65e0\u6cd5\u5bb9\u7eb3\u7684\u8d85\u5927\u6a21\u578b\u3002\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0c\u4e0a\u8ff0\u4f18\u5316\u6280\u672f\u901a\u5e38\u53ef\u7075\u6d3b\u7ec4\u5408\u4f7f\u7528\uff0c\u4f8b\u5982\u8282\u70b9\u95f4\u5e94\u7528\u6d41\u6c34\u7ebf\u5e76\u884c\u3001\u8282\u70b9\u5185\u5e94\u7528\u5f20\u91cf\u5e76\u884c\uff0c\u4ee5\u9002\u5e94\u590d\u6742\u7684\u786c\u4ef6\u62d3\u6251\u7ed3\u6784\uff0c\u5728\u5927\u89c4\u6a21\u63a8\u7406\u573a\u666f\u4e2d\u9ad8\u6548\u6269\u5c55LLM\u7684\u8ba1\u7b97\u80fd\u529b\u3002\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/46472.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2025-07-29T19:16:16+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729191614-68891dfeefc4b.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"2 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/46472.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/46472.html\",\"name\":\"\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u4e09\u5927\u7279\u70b9 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2025-07-29T19:16:16+00:00\",\"dateModified\":\"2025-07-29T19:16:16+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/46472.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/46472.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/46472.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u4e09\u5927\u7279\u70b9\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u4e09\u5927\u7279\u70b9 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/46472.html","og_locale":"zh_CN","og_type":"article","og_title":"\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u4e09\u5927\u7279\u70b9 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb1.1k\u6b21\uff0c\u70b9\u8d5e11\u6b21\uff0c\u6536\u85cf17\u6b21\u3002\u65e0\u8bba\u90e8\u7f72\u5728\u4f55\u79cd\u73af\u5883\u4e2d\uff0c\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u90fd\u4e3a\u7528\u6237\u63d0\u4f9b\u7ecf\u8fc7\u5f3a\u5316\u5e76\u83b7\u5f97\u5b98\u65b9\u652f\u6301\u7684vLLM\u53d1\u884c\u7248\uff0c\u914d\u5957\u667a\u80fdLLM\u538b\u7f29\u5de5\u5177\uff0c\u4ee5\u53ca\u5728Hugging Face\u5e73\u53f0\u4e0a\u4f18\u5316\u7684\u6a21\u578b\u4ed3\u5e93\u3002\u6d41\u6c34\u7ebf\u5e76\u884c\uff08Pipeline Parallelism\uff0cPP\uff09\uff1a\u5c06\u6a21\u578b\u7684\u4e0d\u540c\u5c42\u5212\u5206\u4e3a\u82e5\u5e72\u9636\u6bb5\uff0c\u5206\u522b\u90e8\u7f72\u5728\u4e0d\u540cGPU\u6216\u8282\u70b9\u4e0a\uff0c\u9002\u7528\u4e8e\u5355\u4e00\u591aGPU\u8282\u70b9\u4e5f\u65e0\u6cd5\u5bb9\u7eb3\u7684\u8d85\u5927\u6a21\u578b\u3002\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0c\u4e0a\u8ff0\u4f18\u5316\u6280\u672f\u901a\u5e38\u53ef\u7075\u6d3b\u7ec4\u5408\u4f7f\u7528\uff0c\u4f8b\u5982\u8282\u70b9\u95f4\u5e94\u7528\u6d41\u6c34\u7ebf\u5e76\u884c\u3001\u8282\u70b9\u5185\u5e94\u7528\u5f20\u91cf\u5e76\u884c\uff0c\u4ee5\u9002\u5e94\u590d\u6742\u7684\u786c\u4ef6\u62d3\u6251\u7ed3\u6784\uff0c\u5728\u5927\u89c4\u6a21\u63a8\u7406\u573a\u666f\u4e2d\u9ad8\u6548\u6269\u5c55LLM\u7684\u8ba1\u7b97\u80fd\u529b\u3002","og_url":"https:\/\/www.wsisp.com\/helps\/46472.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2025-07-29T19:16:16+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729191614-68891dfeefc4b.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"2 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/46472.html","url":"https:\/\/www.wsisp.com\/helps\/46472.html","name":"\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u4e09\u5927\u7279\u70b9 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2025-07-29T19:16:16+00:00","dateModified":"2025-07-29T19:16:16+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/46472.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/46472.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/46472.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u7ea2\u5e3dAI\u63a8\u7406\u670d\u52a1\u5668\u4e09\u5927\u7279\u70b9"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/46472","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=46472"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/46472\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/46471"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=46472"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=46472"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=46472"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=46472"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}