{"id":14734,"date":"2024-12-03T14:45:03","date_gmt":"2024-12-03T06:45:03","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=14734"},"modified":"2024-12-03T14:45:03","modified_gmt":"2024-12-03T06:45:03","slug":"shiyonggpustackweidif","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/ja\/shiyonggpustackweidif\/","title":{"rendered":"\u4f7f\u7528GPUStack\u4e3aDify\u5feb\u901f\u90e8\u7f72RAG\u4e09\u4ef6\u5957"},"content":{"rendered":"<p><a href=\"https:\/\/www.kdjingpai.com\/gpustack\/\">GPUStack<\/a> \u662f\u4e00\u4e2a\u5f00\u6e90\u7684\u5927\u6a21\u578b\u5373\u670d\u52a1\u5e73\u53f0\uff0c\u53ef\u4ee5\u9ad8\u6548\u6574\u5408\u5e76\u5229\u7528 Nvidia\u3001Apple Metal\u3001\u534e\u4e3a\u6607\u817e\u548c\u6469\u5c14\u7ebf\u7a0b\u7b49\u5404\u79cd\u5f02\u6784\u7684 GPU\/NPU \u8d44\u6e90\uff0c\u63d0\u4f9b\u672c\u5730\u79c1\u6709\u90e8\u7f72\u5927\u6a21\u578b\u89e3\u51b3\u65b9\u6848\u3002<\/p>\n<p>GPUStack \u53ef\u4ee5\u652f\u6301 <a href=\"https:\/\/www.kdjingpai.com\/rag\/\">RAG<\/a> \u7cfb\u7edf\u4e2d\u6240\u9700\u8981\u7684\u4e09\u79cd\u5173\u952e\u6a21\u578b\uff1aChat \u00a0\u5bf9\u8bdd\u6a21\u578b\uff08\u5927\u8bed\u8a00\u6a21\u578b\uff09\u3001Embedding \u6587\u672c\u5d4c\u5165\u6a21\u578b\u548c Rerank \u91cd\u6392\u5e8f\u6a21\u578b\u4e09\u4ef6\u5957\uff0c\u53ea\u9700\u8981\u975e\u5e38\u7b80\u5355\u7684\u50bb\u74dc\u5316\u64cd\u4f5c\u5c31\u80fd\u90e8\u7f72 RAG \u7cfb\u7edf\u6240\u9700\u8981\u7684\u672c\u5730\u79c1\u6709\u6a21\u578b\u3002<\/p>\n<p>\u4e0b\u9762\u4ecb\u7ecd\u5982\u4f55\u5b89\u88c5 GPUStack \u548c Dify\uff0c\u5e76\u4f7f\u7528 <a href=\"https:\/\/www.kdjingpai.com\/dify\/\">Dify<\/a> \u6765\u5bf9\u63a5 GPUStack \u90e8\u7f72\u7684\u5bf9\u8bdd\u6a21\u578b\u3001Embedding \u6a21\u578b\u548c Reranker \u6a21\u578b\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>\u5b89\u88c5 GPUStack<\/h2>\n<p>\u5728 Linux \u6216 macOS \u4e0a\u901a\u8fc7\u4ee5\u4e0b\u547d\u4ee4\u5728\u7ebf\u5b89\u88c5\uff0c\u5728\u5b89\u88c5\u8fc7\u7a0b\u4e2d\u9700\u8981\u8f93\u5165 sudo \u5bc6\u7801\uff1a<br \/>\n<code>\u00a0\u00a0curl\u00a0-sfL\u00a0https:\/\/get.gpustack.ai\u00a0|\u00a0sh\u00a0-<br \/>\n<\/code><\/p>\n<p>\u5982\u679c\u73af\u5883\u8fde\u63a5\u4e0d\u4e86 GitHub\uff0c\u65e0\u6cd5\u4e0b\u8f7d\u4e00\u4e9b\u4e8c\u8fdb\u5236\u6587\u4ef6\uff0c\u4f7f\u7528\u4ee5\u4e0b\u547d\u4ee4\u5b89\u88c5\uff0c\u7528 <code>--tools-download-base-url<\/code> \u53c2\u6570\u6307\u5b9a\u4ece\u817e\u8baf\u4e91\u5bf9\u8c61\u5b58\u50a8\u4e0b\u8f7d\uff1a<br \/>\n<code>curl\u00a0-sfL\u00a0https:\/\/get.gpustack.ai\u00a0|\u00a0sh\u00a0-\u00a0--tools-download-base-url\u00a0\"https:\/\/gpustack-1303613262.cos.ap-guangzhou.myqcloud.com\"<br \/>\n<\/code><\/p>\n<p>\u5728 Windows \u4e0a\u4ee5\u7ba1\u7406\u5458\u8eab\u4efd\u8fd0\u884c Powershell\uff0c\u901a\u8fc7\u4ee5\u4e0b\u547d\u4ee4\u5728\u7ebf\u5b89\u88c5\uff1a<br \/>\n<code>Invoke-Expression (Invoke-WebRequest -Uri \"https:\/\/get.gpustack.ai\" -UseBasicParsing).Content<br \/>\n<\/code><\/p>\n<p>\u5982\u679c\u73af\u5883\u8fde\u63a5\u4e0d\u4e86 GitHub\uff0c\u65e0\u6cd5\u4e0b\u8f7d\u4e00\u4e9b\u4e8c\u8fdb\u5236\u6587\u4ef6\uff0c\u4f7f\u7528\u4ee5\u4e0b\u547d\u4ee4\u5b89\u88c5\uff0c\u7528 <code>--tools-download-base-url<\/code> \u53c2\u6570\u6307\u5b9a\u4ece\u817e\u8baf\u4e91\u5bf9\u8c61\u5b58\u50a8\u4e0b\u8f7d\uff1a<br \/>\n<code>Invoke-Expression\u00a0\"&amp;\u00a0{\u00a0$((Invoke-WebRequest\u00a0-Uri\u00a0'https:\/\/get.gpustack.ai'\u00a0-UseBasicParsing).Content)\u00a0}\u00a0--tools-download-base-url\u00a0'https:\/\/gpustack-1303613262.cos.ap-guangzhou.myqcloud.com'\"<br \/>\n<\/code><\/p>\n<p>\u5f53\u770b\u5230\u4ee5\u4e0b\u8f93\u51fa\u65f6\uff0c\u8bf4\u660e\u5df2\u7ecf\u6210\u529f\u90e8\u7f72\u5e76\u542f\u52a8\u4e86 GPUStack\uff1a<\/p>\n<pre>[INFO]\u00a0\u00a0Install\u00a0complete.\r\n\r\nGPUStack\u00a0UI\u00a0is\u00a0available\u00a0at\u00a0http:\/\/localhost.\r\nDefault\u00a0username\u00a0is\u00a0'admin'.\r\nTo\u00a0get\u00a0the\u00a0default\u00a0password,\u00a0run\u00a0'cat\u00a0\/var\/lib\/gpustack\/initial_admin_password'.\r\n\r\nCLI\u00a0\"gpustack\"\u00a0is\u00a0available\u00a0from\u00a0the\u00a0<a href=\"https:\/\/www.kdjingpai.com\/cohere\/\">command<\/a>\u00a0line.\u00a0(You\u00a0may\u00a0need\u00a0to\u00a0open\u00a0a\u00a0new\u00a0terminal\u00a0or\u00a0re-login\u00a0for\u00a0the\u00a0PATH\u00a0changes\u00a0to\u00a0take\u00a0effect.)<\/pre>\n<p>\u63a5\u4e0b\u6765\u6309\u7167\u811a\u672c\u8f93\u51fa\u7684\u6307\u5f15\uff0c\u62ff\u5230\u767b\u5f55 GPUStack \u7684\u521d\u59cb\u5bc6\u7801\uff0c\u6267\u884c\u4ee5\u4e0b\u547d\u4ee4\uff1a<br \/>\n\u5728 Linux \u6216 macOS \u4e0a\uff1a<br \/>\n<code>cat\u00a0\/var\/lib\/gpustack\/initial_admin_password<br \/>\n<\/code><\/p>\n<p>\u5728 Windows \u4e0a\uff1a<br \/>\n<code>Get-Content -Path (Join-Path -Path $env:APPDATA -ChildPath \"gpustackinitial_admin_password\") -Raw<br \/>\n<\/code><\/p>\n<p>\u5728\u6d4f\u89c8\u5668\u8bbf\u95ee GPUStack UI\uff0c\u7528\u6237\u540d admin\uff0c\u5bc6\u7801\u4e3a\u4e0a\u9762\u83b7\u5f97\u7684\u521d\u59cb\u5bc6\u7801\u3002<\/p>\n<p>\u91cd\u65b0\u8bbe\u7f6e\u5bc6\u7801\u540e\uff0c\u8fdb\u5165 GPUStack\uff1a<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/a15222d031a01c4.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u7eb3\u7ba1 GPU \u8d44\u6e90<\/h2>\n<p>GPUStack \u652f\u6301 Linux\u3001Windows \u548c macOS \u8bbe\u5907\u7684 GPU \u8d44\u6e90\uff0c\u901a\u8fc7\u4ee5\u4e0b\u6b65\u9aa4\u6765\u7eb3\u7ba1\u8fd9\u4e9b GPU \u8d44\u6e90\u3002<\/p>\n<p>\u5176\u4ed6\u8282\u70b9\u9700\u8981\u901a\u8fc7\u8ba4\u8bc1 <a href=\"https:\/\/www.kdjingpai.com\/tokenization\/\">Token<\/a> \u52a0\u5165 GPUStack \u96c6\u7fa4\uff0c\u5728 GPUStack Server \u8282\u70b9\u6267\u884c\u4ee5\u4e0b\u547d\u4ee4\u83b7\u53d6 Token\uff1a<\/p>\n<p>\u5728 Linux \u6216 macOS \u4e0a\uff1a<br \/>\n<code>cat\u00a0\/var\/lib\/gpustack\/token<br \/>\n<\/code><\/p>\n<p>\u5728 Windows \u4e0a\uff1a<br \/>\n<code>Get-Content -Path (Join-Path -Path $env:APPDATA -ChildPath \"gpustacktoken\") -Raw<br \/>\n<\/code><\/p>\n<p>\u62ff\u5230 Token \u540e\uff0c\u5728\u5176\u4ed6\u8282\u70b9\u4e0a\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u6dfb\u52a0 Worker \u5230 GPUStack\uff0c\u7eb3\u7ba1\u8fd9\u4e9b\u8282\u70b9\u7684 GPU\uff08\u5c06\u5176\u4e2d\u7684 http:\/\/YOUR_IP_ADDRESS \u66ff\u6362\u4e3a\u4f60\u7684 GPUStack \u8bbf\u95ee\u5730\u5740\uff0c\u5c06 YOUR_TOKEN \u66ff\u6362\u4e3a\u7528\u4e8e\u6dfb\u52a0 Worker \u7684\u8ba4\u8bc1 Token\uff09\uff1a<\/p>\n<p>\u5728 Linux \u6216 macOS \u4e0a\uff1a<br \/>\n<code>curl\u00a0-sfL\u00a0https:\/\/get.gpustack.ai\u00a0|\u00a0sh\u00a0-\u00a0--server-url\u00a0http:\/\/YOUR_IP_ADDRESS\u00a0--token\u00a0YOUR_TOKEN\u00a0--tools-download-base-url\u00a0\"https:\/\/gpustack-1303613262.cos.ap-guangzhou.myqcloud.com\"<br \/>\n<\/code><\/p>\n<p>\u5728 Windows \u4e0a\uff1a<br \/>\n<code>Invoke-Expression \"&amp; { $((Invoke-WebRequest -Uri \"https:\/\/get.gpustack.ai\" -UseBasicParsing).Content) } --server-url http:\/\/YOUR_IP_ADDRESS --token YOUR_TOKEN --tools-download-base-url 'https:\/\/gpustack-1303613262.cos.ap-guangzhou.myqcloud.com'\"<br \/>\n<\/code><\/p>\n<p>\u901a\u8fc7\u4ee5\u4e0a\u6b65\u9aa4\uff0c\u6211\u4eec\u5df2\u7ecf\u521b\u5efa\u4e86\u4e00\u4e2a GPUStack \u73af\u5883\u5e76\u7eb3\u7ba1\u4e86\u591a\u4e2a GPU \u8282\u70b9\uff0c\u63a5\u4e0b\u6765\u53ef\u4ee5\u4f7f\u7528\u8fd9\u4e9b GPU \u8d44\u6e90\u6765\u90e8\u7f72\u79c1\u6709\u5927\u6a21\u578b\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>\u90e8\u7f72\u79c1\u6709\u5927\u6a21\u578b<\/h2>\n<p>\u8bbf\u95ee GPUStack\uff0c\u5728 Models \u83dc\u5355\u4e2d\u90e8\u7f72\u6a21\u578b\u3002GPUStack \u652f\u6301\u4ece HuggingFace\u3001Ollama Library\u3001ModelScope \u548c\u79c1\u6709\u6a21\u578b\u4ed3\u5e93\u90e8\u7f72\u6a21\u578b\uff0c\u56fd\u5185\u7f51\u7edc\u5efa\u8bae\u4ece ModelScope \u90e8\u7f72\u3002<\/p>\n<p>GPUStack \u652f\u6301 <a href=\"https:\/\/www.kdjingpai.com\/vllm\/\">vLLM<\/a> \u548c llama-box \u63a8\u7406\u540e\u7aef\uff0cvLLM \u4e13\u95e8\u9488\u5bf9\u751f\u4ea7\u63a8\u7406\u8fdb\u884c\u4e86\u4f18\u5316\uff0c\u5728\u5e76\u53d1\u548c\u6027\u80fd\u65b9\u9762\u66f4\u80fd\u6ee1\u8db3\u751f\u4ea7\u9700\u6c42\uff0c\u4f46 vLLM \u53ea\u652f\u6301 Linux \u7cfb\u7edf\u3002llama-box \u5219\u662f\u4e00\u4e2a\u7075\u6d3b\u3001\u517c\u5bb9\u591a\u5e73\u53f0\u7684\u63a8\u7406\u5f15\u64ce\uff0c\u662f <a href=\"https:\/\/www.kdjingpai.com\/llamacpp\/\">llama.cpp<\/a> \u7684\u4f18\u5316\u7248\u672c\uff0c\u5bf9\u6027\u80fd\u548c\u7a33\u5b9a\u6027\u8fdb\u884c\u4e86\u9488\u5bf9\u6027\u7684\u4f18\u5316\uff0c\u652f\u6301 Linux\u3001Windows \u548c macOS \u7cfb\u7edf\uff0c\u4e0d\u6b62\u652f\u6301\u5404\u79cd GPU \u73af\u5883\uff0c\u4e5f\u652f\u6301\u5728 CPU \u73af\u5883\u8fd0\u884c\u5927\u6a21\u578b\uff0c\u66f4\u9002\u5408\u9700\u8981\u591a\u5e73\u53f0\u517c\u5bb9\u6027\u7684\u573a\u666f\u3002<\/p>\n<p>GPUStack \u4f1a\u5728\u90e8\u7f72\u6a21\u578b\u65f6\u81ea\u52a8\u6839\u636e\u6a21\u578b\u6587\u4ef6\u7684\u7c7b\u578b\u9009\u62e9\u9002\u5f53\u7684\u63a8\u7406\u540e\u7aef\uff0c\u5982\u679c\u6a21\u578b\u4e3a GGUF \u683c\u5f0f\uff0cGPUStack \u4f1a\u4f7f\u7528 llama-box \u4f5c\u4e3a\u540e\u7aef\u8fd0\u884c\u6a21\u578b\u670d\u52a1\uff0c\u5982\u679c\u4e3a\u975e GGUF \u683c\u5f0f\uff0c GPUStack \u4f1a\u4f7f\u7528 vLLM \u4f5c\u4e3a\u540e\u7aef\u8fd0\u884c\u6a21\u578b\u670d\u52a1\u3002<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/b89c698e130306a.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u90e8\u7f72 Dify \u5bf9\u63a5\u6240\u9700\u8981\u7684\u6587\u672c\u5bf9\u8bdd\u6a21\u578b\u3001Embedding \u6587\u672c\u5d4c\u5165\u6a21\u578b\u3001Reranker \u6a21\u578b\uff0c\u8bb0\u5f97\u90e8\u7f72\u65f6\u52fe\u9009 GGUF \u683c\u5f0f\uff1a<\/p>\n<ul>\n<li>Qwen\/Qwen2.5-7B-Instruct-GGUF<\/li>\n<li>gpustack\/bge-m3-GGUF<\/li>\n<li>gpustack\/bge-reranker-v2-m3-GGUF<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/0b90c5fcb22ffb6.png\" \/><\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/0c45cdd6524fbea.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>GPUStack \u8fd8\u652f\u6301 VLM \u591a\u6a21\u6001\u6a21\u578b\uff0c\u90e8\u7f72 VLM \u591a\u6a21\u6001\u6a21\u578b\u9700\u8981\u4f7f\u7528 vLLM \u63a8\u7406\u540e\u7aef\uff1a<br \/>\nQwen2-VL-2B-Instruct <img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/ec2bb36d70a398a.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/b9a8a4c3c641c54.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u6a21\u578b\u90e8\u7f72\u597d\u540e\uff0cRAG \u7cfb\u7edf\u6216\u5176\u4ed6\u751f\u6210\u5f0f AI \u5e94\u7528\u53ef\u4ee5\u901a\u8fc7 GPUStack \u63d0\u4f9b\u7684 OpenAI \/ Jina \u517c\u5bb9 API \u5bf9\u63a5 GPUStack \u90e8\u7f72\u7684\u6a21\u578b\uff0c\u63a5\u4e0b\u6765\u4f7f\u7528 Dify \u6765\u5bf9\u63a5 GPUStack \u90e8\u7f72\u7684\u6a21\u578b\u3002<\/p>\n<h2>\nDify \u96c6\u6210 GPUStack \u6a21\u578b<\/h2>\n<h3>\u5b89\u88c5 Dify<\/h3>\n<p>\u91c7\u7528 Docker \u65b9\u5f0f\u8fd0\u884c Dify\uff0c\u9700\u8981\u51c6\u5907\u597d Docker \u73af\u5883\uff0c\u6ce8\u610f\u907f\u514d Dify \u548c GPUStack \u7684 80 \u7aef\u53e3\u51b2\u7a81\uff0c\u4f7f\u7528\u5176\u4ed6\u4e3b\u673a\u6216\u4fee\u6539\u7aef\u53e3\u3002\u6267\u884c\u4ee5\u4e0b\u547d\u4ee4\u5b89\u88c5 Dify\uff1a<br \/>\n<code>git\u00a0clone\u00a0-b\u00a00.10.1\u00a0https:\/\/github.com\/langgenius\/dify.git<br \/>\ncd\u00a0dify\/docker\/<br \/>\ncp\u00a0.env.example\u00a0.env<br \/>\ndocker\u00a0compose\u00a0up\u00a0-d<br \/>\n<\/code>\u8bbf\u95ee Dify \u7684 UI \u754c\u9762 http:\/\/localhost\uff0c\u521d\u59cb\u5316\u7ba1\u7406\u5458\u8d26\u6237\u5e76\u767b\u5f55\u3002<br \/>\n\u96c6\u6210 GPUStack \u6a21\u578b\u9996\u5148\u6dfb\u52a0 Chat \u5bf9\u8bdd\u6a21\u578b\uff0c\u5728 Dify \u53f3\u4e0a\u89d2\u9009\u62e9\u201c\u8bbe\u7f6e-\u6a21\u578b\u4f9b\u5e94\u5546\u201d\uff0c\u5728\u5217\u8868\u4e2d\u627e\u5230 GPUStack \u7c7b\u578b\uff0c\u9009\u62e9\u6dfb\u52a0\u6a21\u578b\uff1a<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/547eede3af72616.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u586b\u5199 GPUStack \u4e0a\u90e8\u7f72\u7684 LLM \u6a21\u578b\u540d\u79f0\uff08\u4f8b\u5982 qwen2.5-7b-instruct\uff09\u3001GPUStack \u7684\u8bbf\u95ee\u5730\u5740\uff08\u4f8b\u5982 http:\/\/192.168.0.111\uff09\u548c\u751f\u6210\u7684 API Key\uff0c\u8fd8\u6709\u6a21\u578b\u8bbe\u7f6e\u7684\u4e0a\u4e0b\u6587\u957f\u5ea6 8192 \u548c max <a href=\"https:\/\/www.kdjingpai.com\/tokenization\/\">tokens<\/a> 2048\uff1a<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/44ac0f469ec370b.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u63a5\u4e0b\u6765\u6dfb\u52a0 Embedding \u6a21\u578b\uff0c\u5728\u6a21\u578b\u4f9b\u5e94\u5546\u7684\u6700\u4e0a\u65b9\u7ee7\u7eed\u9009\u62e9 GPUStack \u7c7b\u578b\uff0c\u9009\u62e9\u6dfb\u52a0\u6a21\u578b\uff1a<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/53070f04383b25a.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u6dfb\u52a0 Text Embedding \u7c7b\u578b\u7684\u6a21\u578b\uff0c\u586b\u5199 GPUStack \u4e0a\u90e8\u7f72\u7684 Embedding \u6a21\u578b\u540d\u79f0\uff08\u4f8b\u5982 bge-m3\uff09\u3001GPUStack \u7684\u8bbf\u95ee\u5730\u5740\uff08\u4f8b\u5982 http:\/\/192.168.0.111\uff09\u548c\u751f\u6210\u7684 API Key\uff0c\u8fd8\u6709\u6a21\u578b\u8bbe\u7f6e\u7684\u4e0a\u4e0b\u6587\u957f\u5ea6 8192\uff1a<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/531712164fb5c94.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u63a5\u4e0b\u6765\u6dfb\u52a0 Rerank \u6a21\u578b\uff0c\u7ee7\u7eed\u9009\u62e9 GPUStack \u7c7b\u578b\uff0c\u9009\u62e9\u6dfb\u52a0\u6a21\u578b\uff0c\u6dfb\u52a0 Rerank \u7c7b\u578b\u7684\u6a21\u578b\uff0c\u586b\u5199 GPUStack \u4e0a\u90e8\u7f72\u7684 Rerank \u6a21\u578b\u540d\u79f0\uff08\u4f8b\u5982 bge-reranker-v2-m3\uff09\u3001GPUStack \u7684\u8bbf\u95ee\u5730\u5740\uff08\u4f8b\u5982 http:\/\/192.168.0.111\uff09\u548c\u751f\u6210\u7684 API Key\uff0c\u8fd8\u6709\u6a21\u578b\u8bbe\u7f6e\u7684\u4e0a\u4e0b\u6587\u957f\u5ea6 8192\uff1a<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/a3594a7313a61dd.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u6dfb\u52a0\u540e\u91cd\u65b0\u5237\u65b0\uff0c\u7136\u540e\u5728\u6a21\u578b\u4f9b\u5e94\u5546\u786e\u8ba4\u7cfb\u7edf\u6a21\u578b\u914d\u7f6e\u4e3a\u4e0a\u9762\u6dfb\u52a0\u7684\u4e09\u4e2a\u6a21\u578b\uff1a<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/b03224f8e2cee97.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u5728 RAG \u7cfb\u7edf\u4f7f\u7528\u6a21\u578b\u9009\u62e9 Dfiy \u7684\u77e5\u8bc6\u5e93\uff0c\u9009\u62e9\u521b\u5efa\u77e5\u8bc6\u5e93\uff0c\u5bfc\u5165\u4e00\u4e2a\u6587\u672c\u6587\u4ef6\uff0c\u786e\u8ba4 Embedding \u6a21\u578b\u9009\u9879\uff0c\u68c0\u7d22\u8bbe\u7f6e\u4f7f\u7528\u63a8\u8350\u7684\u6df7\u5408\u68c0\u7d22\uff0c\u5e76\u5f00\u542f Rerank \u6a21\u578b\uff1a<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/ff80e23a6684e4a.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u4fdd\u5b58\uff0c\u5f00\u59cb\u5c06\u6587\u6863\u8fdb\u884c\u5411\u91cf\u5316\u8fc7\u7a0b\uff0c\u5411\u91cf\u5316\u5b8c\u6210\u540e\uff0c\u77e5\u8bc6\u5e93\u5373\u53ef\u4ee5\u4f7f\u7528\u4e86\u3002<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/c024059e3738ebd.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u53ef\u4ee5\u901a\u8fc7\u53ec\u56de\u6d4b\u8bd5\u786e\u8ba4\u77e5\u8bc6\u5e93\u7684\u53ec\u56de\u6548\u679c\uff0cRerank \u6a21\u578b\u5c06\u8fdb\u884c\u7cbe\u6392\u4ee5\u53ec\u56de\u66f4\u6709\u76f8\u5173\u6027\u7684\u6587\u6863\uff0c\u4ee5\u8fbe\u5230\u66f4\u597d\u7684\u53ec\u56de\u6548\u679c\uff1a<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/728e7a421e791c2.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u63a5\u4e0b\u6765\u5728\u804a\u5929\u5ba4\u521b\u5efa\u4e00\u4e2a\u804a\u5929\u52a9\u624b\u5e94\u7528\uff1a<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/6d97fbc067f9e38.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u5728\u4e0a\u4e0b\u6587\u4e2d\u6dfb\u52a0\u76f8\u5173\u77e5\u8bc6\u5e93\u5373\u53ef\u4f7f\u7528\uff0c\u6b64\u65f6 Chat \u6a21\u578b\u3001Embedding \u6a21\u578b\u548c Reranker \u6a21\u578b\u5c06\u5171\u540c\u4e3a RAG \u5e94\u7528\u63d0\u4f9b\u652f\u6491\uff0cEmbedding \u6a21\u578b\u8d1f\u8d23\u5411\u91cf\u5316\u3001 Reranker \u6a21\u578b\u8d1f\u8d23\u5bf9\u53ec\u56de\u7684\u5185\u5bb9\u8fdb\u884c\u7cbe\u6392\uff0cChat \u6a21\u578b\u8d1f\u8d23\u6839\u636e\u95ee\u9898\u548c\u53ec\u56de\u7684\u4e0a\u4e0b\u6587\u5185\u5bb9\u8fdb\u884c\u56de\u7b54\uff1a<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/7911d06dfba3bf9.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u4ee5\u4e0a\u4e3a\u4f7f\u7528 Dify \u5bf9\u63a5 GPUStack \u6a21\u578b\u7684\u793a\u4f8b\uff0c\u5176\u4ed6 RAG \u7cfb\u7edf\u4e5f\u53ef\u4ee5\u901a\u8fc7 OpenAI \/ Jina \u517c\u5bb9 API \u5bf9\u63a5 GPUStack\uff0c\u5373\u53ef\u5229\u7528 GPUStack \u5e73\u53f0\u90e8\u7f72\u7684\u5404\u79cd Chat \u6a21\u578b\u3001Embedding \u6a21\u578b\u548c Reranker \u6a21\u578b\u6765\u652f\u6491 RAG \u7cfb\u7edf\u3002<\/p>\n<p>\u4ee5\u4e0b\u4e3a GPUStack \u529f\u80fd\u7684\u7b80\u5355\u4ecb\u7ecd\u3002<\/p>\n<h2>\nGPUStack \u529f\u80fd\u4ecb\u7ecd<\/h2>\n<ul>\n<li>\u5f02\u6784 GPU \u652f\u6301\uff1a\u652f\u6301\u5f02\u6784 GPU \u8d44\u6e90\uff0c\u5f53\u524d\u652f\u6301 Nvidia\u3001Apple Metal\u3001\u534e\u4e3a\u6607\u817e\u548c\u6469\u5c14\u7ebf\u7a0b\u7b49\u5404\u79cd\u7c7b\u578b\u7684 GPU\/NPU<\/li>\n<li>\u591a\u63a8\u7406\u540e\u7aef\u652f\u6301\uff1a\u652f\u6301 vLLM \u548c llama-box (llama.cpp) \u63a8\u7406\u540e\u7aef\uff0c\u517c\u987e\u751f\u4ea7\u6027\u80fd\u9700\u6c42\u4e0e\u591a\u5e73\u53f0\u517c\u5bb9\u6027\u9700\u6c42<\/li>\n<li>\u591a\u5e73\u53f0\u652f\u6301\uff1a\u652f\u6301 Linux\u3001Windows \u548c macOS \u5e73\u53f0\uff0c\u8986\u76d6 amd64 \u548c arm64 \u67b6\u6784<\/li>\n<li>\u591a\u6a21\u578b\u7c7b\u578b\u652f\u6301\uff1a\u652f\u6301 LLM \u6587\u672c\u6a21\u578b\u3001VLM \u591a\u6a21\u6001\u6a21\u578b\u3001Embedding \u6587\u672c\u5d4c\u5165\u6a21\u578b \u548c Reranker \u91cd\u6392\u5e8f\u6a21\u578b\u7b49\u5404\u79cd\u7c7b\u578b\u7684\u6a21\u578b<\/li>\n<li>\u591a\u6a21\u578b\u4ed3\u5e93\u652f\u6301\uff1a\u652f\u6301\u4ece HuggingFace\u3001Ollama Library\u3001ModelScope \u548c\u79c1\u6709\u6a21\u578b\u4ed3\u5e93\u90e8\u7f72\u6a21\u578b<\/li>\n<li>\u4e30\u5bcc\u7684\u81ea\u52a8\/\u624b\u52a8\u8c03\u5ea6\u7b56\u7565\uff1a\u652f\u6301\u7d27\u51d1\u8c03\u5ea6\u3001\u5206\u6563\u8c03\u5ea6\u3001\u6307\u5b9a Worker \u6807\u7b7e\u8c03\u5ea6\u3001\u6307\u5b9a GPU \u8c03\u5ea6\u7b49\u5404\u79cd\u8c03\u5ea6\u7b56\u7565<\/li>\n<li>\u5206\u5e03\u5f0f\u63a8\u7406\uff1a\u5982\u679c\u5355\u4e2a GPU \u65e0\u6cd5\u8fd0\u884c\u8f83\u5927\u7684\u6a21\u578b\uff0c\u53ef\u4ee5\u901a\u8fc7 GPUStack \u7684\u5206\u5e03\u5f0f\u63a8\u7406\u529f\u80fd\uff0c\u81ea\u52a8\u5c06\u6a21\u578b\u8fd0\u884c\u5728\u8de8\u4e3b\u673a\u7684\u591a\u4e2a GPU \u4e0a<\/li>\n<li>CPU \u63a8\u7406\uff1a\u5982\u679c\u6ca1\u6709 GPU \u6216 GPU \u8d44\u6e90\u4e0d\u8db3\uff0cGPUStack \u53ef\u4ee5\u7528 CPU \u8d44\u6e90\u6765\u8fd0\u884c\u5927\u6a21\u578b\uff0c\u652f\u6301 GPU&amp;CPU \u6df7\u5408\u63a8\u7406\u548c\u7eaf CPU \u63a8\u7406\u4e24\u79cd CPU \u63a8\u7406\u6a21\u5f0f<\/li>\n<li>\u591a\u6a21\u578b\u5bf9\u6bd4\uff1aGPUStack \u5728 <a href=\"https:\/\/www.kdjingpai.com\/playground\/\">Playground<\/a> \u4e2d\u63d0\u4f9b\u4e86\u591a\u6a21\u578b\u5bf9\u6bd4\u89c6\u56fe\uff0c\u53ef\u4ee5\u540c\u65f6\u5bf9\u6bd4\u591a\u4e2a\u6a21\u578b\u7684\u95ee\u7b54\u5185\u5bb9\u548c\u6027\u80fd\u6570\u636e\uff0c\u4ee5\u8bc4\u4f30\u4e0d\u540c\u6a21\u578b\u3001\u4e0d\u540c\u6743\u91cd\u3001\u4e0d\u540c Prompt \u53c2\u6570\u3001\u4e0d\u540c\u91cf\u5316\u3001\u4e0d\u540c GPU\u3001\u4e0d\u540c\u63a8\u7406\u540e\u7aef\u7684\u6a21\u578b Serving \u6548\u679c<\/li>\n<li>GPU \u548c LLM \u89c2\u6d4b\u6307\u6807\uff1a\u63d0\u4f9b\u5168\u9762\u7684\u6027\u80fd\u3001\u5229\u7528\u7387\u3001\u72b6\u6001\u76d1\u63a7\u548c\u4f7f\u7528\u6570\u636e\u6307\u6807\uff0c\u4ee5\u8bc4\u4f30 GPU \u548c LLM \u7684\u5229\u7528\u60c5\u51b5<\/li>\n<\/ul>\n<p>GPUStack \u63d0\u4f9b\u4e86\u5efa\u8bbe\u4e00\u4e2a\u79c1\u6709\u5927\u6a21\u578b\u5373\u670d\u52a1\u5e73\u53f0\u6240\u9700\u8981\u7684\u5404\u9879\u4f01\u4e1a\u7ea7\u529f\u80fd\uff0c\u4f5c\u4e3a\u4e00\u4e2a\u5f00\u6e90\u9879\u76ee\uff0c\u53ea\u9700\u8981\u975e\u5e38\u7b80\u5355\u7684\u5b89\u88c5\u8bbe\u7f6e\uff0c\u5c31\u53ef\u4ee5\u5f00\u7bb1\u5373\u7528\u5730\u6784\u5efa\u4f01\u4e1a\u79c1\u6709\u5927\u6a21\u578b\u5373\u670d\u52a1\u5e73\u53f0\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>\u603b\u7ed3<\/h2>\n<p>\u4ee5\u4e0a\u4e3a\u5b89\u88c5 GPUStack \u548c\u4f7f\u7528 Dify \u96c6\u6210 GPUStack \u6a21\u578b\u7684\u914d\u7f6e\u6559\u7a0b\uff0c\u9879\u76ee\u7684\u5f00\u6e90\u5730\u5740\u4e3a\uff1ahttps:\/\/github.com\/gpustack\/gpustack\u3002<\/p>\n<p>GPUStack \u4f5c\u4e3a\u4e00\u4e2a\u4f4e\u95e8\u69db\u3001\u6613\u4e0a\u624b\u3001\u5f00\u7bb1\u5373\u7528\u7684<span class=\"\">\u5f00\u6e90\u5e73\u53f0<\/span>\uff0c\u53ef\u4ee5\u5e2e\u52a9\u4f01\u4e1a\u5feb\u901f\u6574\u5408\u548c\u5229\u7528\u5f02\u6784 GPU \u8d44\u6e90\uff0c\u5728\u77ed\u65f6\u95f4\u5185\u5feb\u901f\u642d\u5efa\u8d77\u4e00\u4e2a\u4f01\u4e1a\u7ea7\u7684\u79c1\u6709\u5927\u6a21\u578b\u5373\u670d\u52a1\u5e73\u53f0\u3002<\/p>\n<p>&nbsp;<\/p>\n","protected":false},"excerpt":{"rendered":"<p>GPUStack \u662f\u4e00\u4e2a\u5f00\u6e90\u7684\u5927\u6a21\u578b\u5373\u670d\u52a1\u5e73\u53f0\uff0c\u53ef\u4ee5\u9ad8\u6548\u6574\u5408\u5e76\u5229\u7528 Nvidia\u3001Apple Metal\u3001\u534e\u4e3a\u6607\u817e\u548c\u6469\u5c14\u7ebf\u7a0b\u7b49\u5404\u79cd\u5f02\u6784\u7684 GPU\/NPU \u8d44\u6e90\uff0c\u63d0\u4f9b\u672c\u5730\u79c1\u6709\u90e8\u7f72\u5927\u6a21\u578b\u89e3\u51b3\u65b9\u6848\u3002 GPUStack \u53ef\u4ee5\u652f\u6301 RAG \u7cfb\u7edf\u4e2d\u6240\u9700&#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[182],"tags":[],"class_list":["post-14734","post","type-post","status-publish","format-standard","hentry","category-shicao"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts\/14734","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/comments?post=14734"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts\/14734\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/media?parent=14734"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/categories?post=14734"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/tags?post=14734"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}