{"id":13656,"date":"2024-11-20T13:51:13","date_gmt":"2024-11-20T05:51:13","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=13656"},"modified":"2024-11-20T13:51:13","modified_gmt":"2024-11-20T05:51:13","slug":"aphrodite-enginea","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/pt\/aphrodite-enginea\/","title":{"rendered":"Aphrodite Engine\uff1a\u9ad8\u6548\u7684LLM\u63a8\u7406\u5f15\u64ce\uff0c\u652f\u6301\u591a\u79cd\u91cf\u5316\u683c\u5f0f\u548c\u5206\u5e03\u5f0f\u63a8\u7406\u3002"},"content":{"rendered":"<p>Aphrodite Engine\u662fPygmalionAI\u7684\u5b98\u65b9\u540e\u7aef\u5f15\u64ce\uff0c\u65e8\u5728\u4e3aPygmalionAI\u7f51\u7ad9\u63d0\u4f9b\u63a8\u7406\u7aef\u70b9\uff0c\u5e76\u652f\u6301Hugging Face\u517c\u5bb9\u6a21\u578b\u7684\u5feb\u901f\u90e8\u7f72\u3002\u8be5\u5f15\u64ce\u5229\u7528vLLM\u7684Paged Attention\u6280\u672f\uff0c\u5b9e\u73b0\u4e86\u9ad8\u6548\u7684K\/V\u7ba1\u7406\u548c\u8fde\u7eed\u6279\u5904\u7406\uff0c\u663e\u8457\u63d0\u5347\u4e86\u63a8\u7406\u901f\u5ea6\u548c\u5185\u5b58\u5229\u7528\u7387\u3002Aphrodite Engine\u652f\u6301\u591a\u79cd\u91cf\u5316\u683c\u5f0f\u548c\u5206\u5e03\u5f0f\u63a8\u7406\uff0c\u9002\u7528\u4e8e\u5404\u79cd\u73b0\u4ee3GPU\u548cTPU\u8bbe\u5907\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li><strong>\u8fde\u7eed\u6279\u5904\u7406<\/strong>\uff1a\u9ad8\u6548\u5904\u7406\u591a\u4e2a\u8bf7\u6c42\uff0c\u63d0\u5347\u63a8\u7406\u901f\u5ea6\u3002<\/li>\n<li><strong>Paged Attention<\/strong>\uff1a\u4f18\u5316K\/V\u7ba1\u7406\uff0c\u63d0\u5347\u5185\u5b58\u5229\u7528\u7387\u3002<\/li>\n<li><strong>CUDA\u4f18\u5316\u5185\u6838<\/strong>\uff1a\u63d0\u9ad8\u63a8\u7406\u6027\u80fd\u3002<\/li>\n<li><strong>\u91cf\u5316\u652f\u6301<\/strong>\uff1a\u652f\u6301AQLM\u3001AWQ\u3001Bitsandbytes\u7b49\u591a\u79cd\u91cf\u5316\u683c\u5f0f\u3002<\/li>\n<li><strong>\u5206\u5e03\u5f0f\u63a8\u7406<\/strong>\uff1a\u652f\u63018-bit KV\u7f13\u5b58\uff0c\u9002\u7528\u4e8e\u9ad8\u4e0a\u4e0b\u6587\u957f\u5ea6\u548c\u9ad8\u541e\u5410\u91cf\u9700\u6c42\u3002<\/li>\n<li><strong>\u591a\u8bbe\u5907\u652f\u6301<\/strong>\uff1a\u517c\u5bb9NVIDIA\u3001AMD\u3001Intel GPU\u53caGoogle TPU\u3002<\/li>\n<li><strong>Docker\u90e8\u7f72<\/strong>\uff1a\u63d0\u4f9bDocker\u955c\u50cf\uff0c\u7b80\u5316\u90e8\u7f72\u6d41\u7a0b\u3002<\/li>\n<li><strong>API\u517c\u5bb9<\/strong>\uff1a\u652f\u6301OpenAI\u517c\u5bb9API\uff0c\u4fbf\u4e8e\u96c6\u6210\u5230\u73b0\u6709\u7cfb\u7edf\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<h3>\u5b89\u88c5\u6d41\u7a0b<\/h3>\n<ol>\n<li><strong>\u5b89\u88c5\u4f9d\u8d56<\/strong>\uff1a\n<ul>\n<li>\u786e\u4fdd\u7cfb\u7edf\u5b89\u88c5\u4e86Python 3.8\u81f33.12\u7248\u672c\u3002<\/li>\n<li>\u5bf9\u4e8eLinux\u7528\u6237\uff0c\u5efa\u8bae\u4f7f\u7528\u4ee5\u4e0b\u547d\u4ee4\u5b89\u88c5\u4f9d\u8d56\uff1a<\/li>\n<\/ul>\n<pre><code> sudo apt update &amp;&amp; sudo apt install python3 python3-pip git wget curl bzip2 tar\r\n<\/code><\/pre>\n<ul>\n<li>\u5bf9\u4e8eWindows\u7528\u6237\uff0c\u5efa\u8bae\u4f7f\u7528WSL2\u5b89\u88c5\uff1a<\/li>\n<\/ul>\n<pre><code> wsl --install\r\nsudo apt update &amp;&amp; sudo apt install python3 python3-pip git wget curl bzip2 tar\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u5b89\u88c5Aphrodite Engine<\/strong>\uff1a\n<ul>\n<li>\u4f7f\u7528pip\u5b89\u88c5\uff1a<\/li>\n<\/ul>\n<pre><code> pip install -U aphrodite-engine\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u542f\u52a8\u6a21\u578b<\/strong>\uff1a\n<ul>\n<li>\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u542f\u52a8\u6a21\u578b\uff1a <code>bash<br \/>\naphrodite run meta-llama\/Meta-Llama-3.1-8B-Instruct<br \/>\n<\/code><\/li>\n<li>\u8fd9\u5c06\u521b\u5efa\u4e00\u4e2aOpenAI\u517c\u5bb9\u7684API\u670d\u52a1\u5668\uff0c\u9ed8\u8ba4\u7aef\u53e3\u4e3a2242\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n<h3>\u4f7f\u7528Docker\u90e8\u7f72<\/h3>\n<ol>\n<li><strong>\u62c9\u53d6Docker\u955c\u50cf<\/strong>\uff1a<\/li>\n<\/ol>\n<pre><code>   docker pull alpindale\/aphrodite-openai:latest\r\n<\/code><\/pre>\n<ol start=\"2\">\n<li><strong>\u8fd0\u884cDocker\u5bb9\u5668<\/strong>\uff1a<\/li>\n<\/ol>\n<pre><code>   docker run --runtime nvidia --gpus all \\\r\n-v ~\/.cache\/huggingface:\/root\/.cache\/huggingface \\\r\n-p 2242:2242 \\\r\n--ipc=host \\\r\nalpindale\/aphrodite-openai:latest \\\r\n--model NousResearch\/Meta-Llama-3.1-8B-Instruct \\\r\n--tensor-parallel-size 8 \\\r\n--api-keys \"sk-empty\"\r\n<\/code><\/pre>\n<h3>\u4e3b\u8981\u529f\u80fd\u64cd\u4f5c\u6d41\u7a0b<\/h3>\n<ol>\n<li><strong>\u8fde\u7eed\u6279\u5904\u7406<\/strong>\uff1a\n<ul>\n<li>Aphrodite Engine\u901a\u8fc7\u8fde\u7eed\u6279\u5904\u7406\u6280\u672f\uff0c\u53ef\u4ee5\u540c\u65f6\u5904\u7406\u591a\u4e2a\u8bf7\u6c42\uff0c\u663e\u8457\u63d0\u5347\u63a8\u7406\u901f\u5ea6\u3002\u7528\u6237\u53ea\u9700\u5728\u542f\u52a8\u65f6\u6307\u5b9a\u6279\u5904\u7406\u53c2\u6570\u5373\u53ef\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>Paged Attention<\/strong>\uff1a\n<ul>\n<li>\u8be5\u6280\u672f\u4f18\u5316\u4e86K\/V\u7ba1\u7406\uff0c\u63d0\u5347\u4e86\u5185\u5b58\u5229\u7528\u7387\u3002\u7528\u6237\u65e0\u9700\u989d\u5916\u914d\u7f6e\uff0c\u7cfb\u7edf\u4f1a\u81ea\u52a8\u5e94\u7528\u8be5\u4f18\u5316\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u91cf\u5316\u652f\u6301<\/strong>\uff1a\n<ul>\n<li>\u652f\u6301\u591a\u79cd\u91cf\u5316\u683c\u5f0f\uff0c\u5982AQLM\u3001AWQ\u3001Bitsandbytes\u7b49\u3002\u7528\u6237\u53ef\u4ee5\u5728\u542f\u52a8\u6a21\u578b\u65f6\u6307\u5b9a\u6240\u9700\u7684\u91cf\u5316\u683c\u5f0f\uff1a<\/li>\n<\/ul>\n<pre><code> aphrodite run --quant-format AQLM meta-llama\/Meta-Llama-3.1-8B-Instruct\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u5206\u5e03\u5f0f\u63a8\u7406<\/strong>\uff1a\n<ul>\n<li>\u652f\u63018-bit KV\u7f13\u5b58\uff0c\u9002\u7528\u4e8e\u9ad8\u4e0a\u4e0b\u6587\u957f\u5ea6\u548c\u9ad8\u541e\u5410\u91cf\u9700\u6c42\u3002\u7528\u6237\u53ef\u4ee5\u901a\u8fc7\u4ee5\u4e0b\u547d\u4ee4\u542f\u52a8\u5206\u5e03\u5f0f\u63a8\u7406\uff1a<\/li>\n<\/ul>\n<pre><code> aphrodite run --tensor-parallel-size 8 meta-llama\/Meta-Llama-3.1-8B-Instruct\r\n<\/code><\/pre>\n<\/li>\n<li><strong>API\u96c6\u6210<\/strong>\uff1a\n<ul>\n<li>Aphrodite Engine\u63d0\u4f9bOpenAI\u517c\u5bb9API\uff0c\u4fbf\u4e8e\u96c6\u6210\u5230\u73b0\u6709\u7cfb\u7edf\u3002\u7528\u6237\u53ef\u4ee5\u901a\u8fc7\u4ee5\u4e0b\u547d\u4ee4\u542f\u52a8API\u670d\u52a1\u5668\uff1a <code>bash<br \/>\naphrodite run --api-keys \"your-api-key\" meta-llama\/Meta-Llama-3.1-8B-Instruct<br \/>\n<\/code><\/li>\n<\/ul>\n<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>Aphrodite Engine\u662fPygmalionAI\u7684\u5b98\u65b9\u540e\u7aef\u5f15\u64ce\uff0c\u65e8\u5728\u4e3aPygmalionAI\u7f51\u7ad9\u63d0\u4f9b\u63a8\u7406\u7aef\u70b9\uff0c\u5e76\u652f\u6301Hugging Face\u517c\u5bb9\u6a21\u578b\u7684\u5feb\u901f\u90e8\u7f72\u3002\u8be5\u5f15\u64ce\u5229\u7528vLLM\u7684Paged Attention\u6280\u672f\uff0c\u5b9e\u73b0\u4e86\u9ad8\u6548\u7684K\/&#8230;<\/p>\n","protected":false},"author":1,"featured_media":61272,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[232],"class_list":["post-13656","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tool","tag-bendebushukaiyuanba"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/13656","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=13656"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/13656\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media\/61272"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media?parent=13656"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/categories?post=13656"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/tags?post=13656"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}