{"id":20791,"date":"2025-02-11T14:46:26","date_gmt":"2025-02-11T06:46:26","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=20791"},"modified":"2025-02-11T14:47:36","modified_gmt":"2025-02-11T06:47:36","slug":"ktransformers","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/ja\/ktransformers\/","title":{"rendered":"KTransformers\uff1a\u5927\u6a21\u578b\u63a8\u7406\u6027\u80fd\u5f15\u64ce\uff1a\u6781\u81f4\u52a0\u901f\uff0c\u7075\u6d3b\u8d4b\u80fd"},"content":{"rendered":"<p>KTransformers\uff1a\u4e13\u4e3a\u7a81\u7834\u5927\u6a21\u578b\u63a8\u7406\u74f6\u9888\u800c\u751f\u7684\u9ad8\u6027\u80fd Python \u6846\u67b6\u3002\u00a0\u5b83\u4e0d\u4ec5\u4ec5\u662f\u4e00\u4e2a\u7b80\u5355\u7684\u6a21\u578b\u8fd0\u884c\u5de5\u5177\uff0c\u66f4\u662f\u4e00\u5957\u00a0\u6781\u81f4\u7684\u6027\u80fd\u4f18\u5316\u5f15\u64ce\u00a0\u548c\u00a0\u7075\u6d3b\u7684\u63a5\u53e3\u8d4b\u80fd\u5e73\u53f0\u3002 KTransformers \u81f4\u529b\u4e8e\u4ece\u5e95\u5c42\u63d0\u5347\u5927\u6a21\u578b\u63a8\u7406\u6548\u7387\uff0c\u901a\u8fc7\u5148\u8fdb\u7684\u5185\u6838\u4f18\u5316\u3001\u5f3a\u5927\u7684\u5e76\u884c\u7b56\u7565 (\u591aGPU\u3001\u7a00\u758f\u6ce8\u610f\u529b)\u00a0\u7b49\u6838\u5fc3\u6280\u672f\uff0c\u663e\u8457\u52a0\u901f\u6a21\u578b\u63a8\u7406\u901f\u5ea6\uff0c\u964d\u4f4e\u786c\u4ef6\u95e8\u69db\u3002<\/p>\n<p>\u8d85\u8d8a\u7b80\u5355\u7684\u6a21\u578b\u8fd0\u884c\uff0cKTransformers \u63d0\u4f9b\u7684\u662f\u5168\u65b9\u4f4d\u7684\u6027\u80fd\u63d0\u5347\u548c\u5e94\u7528\u62d3\u5c55\u80fd\u529b\u3002\u00a0\u7528\u6237\u4e0d\u4ec5\u53ef\u4ee5\u83b7\u5f97\u5ab2\u7f8e\u539f\u751f Transformers \u7684\u517c\u5bb9\u63a5\u53e3\uff0c\u65e0\u7f1d\u8fc1\u79fb\u73b0\u6709\u9879\u76ee\uff0c\u66f4\u80fd\u8f7b\u677e\u6784\u5efa\u7b26\u5408 OpenAI \u548c <a href=\"https:\/\/www.kdjingpai.com\/ja\/ollama\/\">Ollama<\/a> \u6807\u51c6\u7684 RESTful API \u670d\u52a1\uff0c\u5feb\u901f\u96c6\u6210\u5230\u5404\u7c7b\u5e94\u7528\u4e2d\u3002 \u540c\u65f6\uff0c\u6211\u4eec\u8fd8\u63d0\u4f9b\u4e86\u5f00\u7bb1\u5373\u7528\u7684 <a href=\"https:\/\/www.kdjingpai.com\/ja\/chatgpt-6\/\">ChatGPT<\/a> \u98ce\u683c\u7f51\u9875\u754c\u9762\uff0c\u65b9\u4fbf\u7528\u6237\u5feb\u901f\u4f53\u9a8c\u548c\u6d4b\u8bd5\u6a21\u578b\u6548\u679c\uff0c\u65e0\u9700\u7e41\u7410\u914d\u7f6e\u3002<\/p>\n<p>KTransformers \u4e13\u4e3a\u5bf9\u6027\u80fd\u6709\u66f4\u9ad8\u8981\u6c42\u7684\u7528\u6237\u8bbe\u8ba1\u3002\u00a0\u65e0\u8bba\u4f60\u662f\u8ffd\u6c42\u6781\u81f4\u63a8\u7406\u901f\u5ea6\u7684\u5f00\u53d1\u8005\uff0c\u9700\u8981\u9ad8\u6548\u90e8\u7f72\u5927\u6a21\u578b\u5e94\u7528\u7684\u5de5\u7a0b\u5e08\uff0c\u8fd8\u662f\u5e0c\u671b\u5728\u672c\u5730\u8f7b\u677e\u4f53\u9a8c\u9ad8\u6027\u80fd\u5927\u6a21\u578b\u7684\u7528\u6237\uff0cKTransformers \u90fd\u80fd\u4e3a\u4f60\u63d0\u4f9b\u5f3a\u5927\u7684\u652f\u6301\uff0c\u52a9\u4f60\u5145\u5206\u91ca\u653e\u5927\u6a21\u578b\u7684\u6f5c\u80fd\uff0c\u9a71\u52a8\u521b\u65b0\u5e94\u7528\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter wp-image-20793\" title=\"KTransformers\uff1a\u5927\u6a21\u578b\u63a8\u7406\u6027\u80fd\u5f15\u64ce\uff1a\u6781\u81f4\u52a0\u901f\uff0c\u7075\u6d3b\u8d4b\u80fd-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/6f4a6276cc218e0.png\" alt=\"KTransformers\uff1a\u5927\u6a21\u578b\u63a8\u7406\u6027\u80fd\u5f15\u64ce\uff1a\u6781\u81f4\u52a0\u901f\uff0c\u7075\u6d3b\u8d4b\u80fd-1\" width=\"1002\" height=\"594\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/6f4a6276cc218e0.png 1603w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/6f4a6276cc218e0-768x456.png 768w\" sizes=\"auto, (max-width: 1002px) 100vw, 1002px\" \/><\/p>\n<p>&nbsp;<\/p>\n<p><strong>\u6838\u5fc3\u4f18\u52bf\uff1a<\/strong><\/p>\n<ul>\n<li><strong>\u6781\u81f4\u6027\u80fd\uff1a<\/strong>\u00a0\u5185\u6838\u7ea7\u4f18\u5316\u548c\u5e76\u884c\u7b56\u7565\u5e26\u6765<strong>\u6570\u91cf\u7ea7\u7684\u63a8\u7406\u901f\u5ea6\u63d0\u5347<\/strong>\u3002<\/li>\n<li><strong>\u7075\u6d3b\u63a5\u53e3\uff1a<\/strong>\u00a0\u63d0\u4f9b Transformers \u517c\u5bb9\u63a5\u53e3\u3001RESTful API \u548c\u7f51\u9875\u754c\u9762\uff0c<strong>\u6ee1\u8db3\u4e0d\u540c\u5e94\u7528\u573a\u666f\u9700\u6c42<\/strong>\u3002<\/li>\n<li><strong>\u5e7f\u6cdb\u517c\u5bb9\uff1a<\/strong>\u00a0\u652f\u6301\u591a GPU\u3001\u591a\u79cd CPU \u67b6\u6784\u548c\u591a\u79cd\u4e3b\u6d41\u5927\u6a21\u578b\uff0c<strong>\u9002\u5e94\u591a\u6837\u5316\u786c\u4ef6\u548c\u6a21\u578b\u9009\u62e9<\/strong>\u3002<\/li>\n<li><strong>\u6613\u7528\u6027\u4e0e\u53ef\u5b9a\u5236\u6027\u5e76\u5b58\uff1a<\/strong>\u00a0\u65e2\u6709<strong>\u5f00\u7bb1\u5373\u7528\u7684\u4fbf\u6377\u6027<\/strong>\uff0c\u53c8\u63d0\u4f9b<strong>\u4e30\u5bcc\u7684\u914d\u7f6e\u9009\u9879<\/strong>\uff0c\u6ee1\u8db3\u9ad8\u7ea7\u7528\u6237\u7684\u6df1\u5ea6\u4f18\u5316\u9700\u6c42\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li><strong>\u9ad8\u6027\u80fd Transformers \u517c\u5bb9\u63a5\u53e3<\/strong>\u00a0\uff1a \u63d0\u4f9b\u4e0e Transformers \u5e93\u5b8c\u5168\u517c\u5bb9\u7684\u63a5\u53e3\uff0c<strong>\u96f6\u6210\u672c\u8fc1\u79fb\u73b0\u6709\u9879\u76ee\uff0c\u5373\u523b\u4eab\u53d7\u6027\u80fd\u63d0\u5347<\/strong>\u3002<\/li>\n<li><strong>\u7075\u6d3b\u6613\u7528\u7684 RESTful API \u670d\u52a1<\/strong>\u00a0\uff1a \u9075\u5faa OpenAI \u548c Ollama \u6807\u51c6\uff0c<strong>\u5feb\u901f\u6784\u5efa\u53ef\u6269\u5c55\u7684 API \u670d\u52a1<\/strong>\uff0c\u8f7b\u677e\u96c6\u6210\u5230\u5404\u79cd\u5e94\u7528\u548c\u5e73\u53f0\u3002<\/li>\n<li><strong>\u5f00\u7bb1\u5373\u7528\u7684 ChatGPT \u98ce\u683c\u7f51\u9875\u754c\u9762<\/strong>\u00a0\uff1a \u63d0\u4f9b\u53cb\u597d\u7684\u4ea4\u4e92\u5f0f\u754c\u9762\uff0c<strong>\u96f6\u4ee3\u7801\u5feb\u901f\u4f53\u9a8c\u548c\u6d4b\u8bd5\u6a21\u578b\u6027\u80fd<\/strong>\uff0c\u65b9\u4fbf\u6f14\u793a\u548c\u9a8c\u8bc1\u3002<\/li>\n<li><strong>\u591aGPU \u5e76\u884c\u8ba1\u7b97\u5f15\u64ce<\/strong>\u00a0\uff1a\u00a0<strong>\u5145\u5206\u91ca\u653e\u591a GPU \u7b97\u529b<\/strong>\uff0c\u7ebf\u6027\u63d0\u5347\u63a8\u7406\u901f\u5ea6\uff0c\u5927\u5e45\u7f29\u77ed\u54cd\u5e94\u65f6\u95f4\u3002<\/li>\n<li><strong>\u5185\u6838\u7ea7\u6df1\u5ea6\u6027\u80fd\u4f18\u5316<\/strong>\u00a0\uff1a \u91c7\u7528\u5148\u8fdb\u5185\u6838\u4f18\u5316\u6280\u672f\uff0c<strong>\u4ece\u5e95\u5c42\u6316\u6398\u786c\u4ef6\u6f5c\u529b<\/strong>\uff0c\u5b9e\u73b0\u6a21\u578b\u63a8\u7406\u6027\u80fd\u7684\u8d28\u7684\u98de\u8dc3\u3002<\/li>\n<li><strong>\u667a\u80fd\u7a00\u758f\u6ce8\u610f\u529b\u6846\u67b6<\/strong>\u00a0\uff1a \u652f\u6301\u5757\u7a00\u758f\u6ce8\u610f\u529b\u673a\u5236\uff0c<strong>\u5927\u5e45\u964d\u4f4e\u5185\u5b58\u5360\u7528<\/strong>\uff0c\u5e76\u652f\u6301 CPU \u9ad8\u6548\u89e3\u7801\uff0c<strong>\u7a81\u7834\u786c\u4ef6\u74f6\u9888<\/strong>\u3002<\/li>\n<li><strong>\u5e7f\u6cdb\u7684\u5927\u6a21\u578b\u751f\u6001\u652f\u6301<\/strong>\u00a0\uff1a \u517c\u5bb9 InternLM\u3001DeepSeek-Coder \u7b49<strong>\u591a\u79cd\u4e3b\u6d41\u5927\u6a21\u578b<\/strong>\u00a0(\u6301\u7eed\u6269\u5c55\u4e2d)\uff0c<strong>\u7075\u6d3b\u9009\u62e9\u6700\u4f73\u6a21\u578b\u65b9\u6848<\/strong>\u3002<\/li>\n<li><strong>\u8f7b\u91cf\u7ea7\u672c\u5730\u9ad8\u6027\u80fd\u63a8\u7406<\/strong>\u00a0\uff1a \u65e0\u9700\u6602\u8d35\u4e13\u4e1a\u786c\u4ef6\uff0c<strong>\u5728\u666e\u901a\u684c\u9762\u73af\u5883\u5373\u53ef\u5b9e\u73b0\u5353\u8d8a\u63a8\u7406\u6027\u80fd<\/strong>\uff0c\u964d\u4f4e\u4f7f\u7528\u95e8\u69db\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<h3>\u5b89\u88c5<\/h3>\n<ol>\n<li><strong>\u514b\u9686\u4ed3\u5e93<\/strong>\u00a0\uff1a<\/li>\n<\/ol>\n<pre><code>git clone https:\/\/github.com\/kvcache-ai\/ktransformers.git\r\ncd ktransformers\r\n<\/code><\/pre>\n<ol start=\"2\">\n<li><strong>\u5b89\u88c5\u4f9d\u8d56<\/strong>\u00a0\uff1a<\/li>\n<\/ol>\n<pre><code>pip install -r requirements-local_chat.txt\r\n<\/code><\/pre>\n<ol start=\"3\">\n<li><strong>\u5b89\u88c5KTransformers<\/strong>\u00a0\uff1a<\/li>\n<\/ol>\n<pre><code>python setup.py install\r\n<\/code><\/pre>\n<h3>\u5feb\u901f\u4e0a\u624b<\/h3>\n<ol>\n<li><strong>\u52a0\u8f7d\u6a21\u578b<\/strong>\u00a0\uff1a<\/li>\n<\/ol>\n<pre><code>from ktransformers import KTransformers\r\nmodel = KTransformers(model_name=\"your_model_name\")\r\n<\/code><\/pre>\n<ol start=\"2\">\n<li><strong>\u63a8\u7406\u793a\u4f8b<\/strong>\u00a0\uff1a<\/li>\n<\/ol>\n<pre><code>input_text = \"\u4f60\u597d\uff0cKTransformers\uff01\"\r\noutput = model.infer(input_text)\r\nprint(output)\r\n<\/code><\/pre>\n<ol start=\"3\">\n<li><strong>\u4f7f\u7528RESTful API<\/strong>\u00a0\uff1a \u542f\u52a8API\u670d\u52a1\uff1a<\/li>\n<\/ol>\n<pre><code>python -m ktransformers.api\r\n<\/code><\/pre>\n<p>\u53d1\u9001\u8bf7\u6c42\uff1a<\/p>\n<pre><code>curl -X POST \"http:\/\/localhost:8000\/infer\" -d '{\"text\": \"\u4f60\u597d\uff0cKTransformers\uff01\"}'\r\n<\/code><\/pre>\n<h3>\u9ad8\u7ea7\u529f\u80fd<\/h3>\n<ul>\n<li><strong>\u591aGPU\u652f\u6301<\/strong>\u00a0\uff1a\u7f16\u8f91\u9879\u76ee\u6839\u76ee\u5f55\u4e0b\u7684\u914d\u7f6e\u6587\u4ef6\u00a0<code>config.yaml<\/code>\uff0c\u6307\u5b9a\u591aGPU\u8bbe\u7f6e\uff0c\u63d0\u5347\u63a8\u7406\u901f\u5ea6\u3002<\/li>\n<li><strong>\u7a00\u758f\u6ce8\u610f\u529b<\/strong>\u00a0\uff1a\u5728\u9879\u76ee\u6839\u76ee\u5f55\u4e0b\u7684\u914d\u7f6e\u6587\u4ef6\u00a0<code>config.yaml<\/code>\u00a0\u4e2d\u6dfb\u52a0\u7a00\u758f\u6ce8\u610f\u529b\u914d\u7f6e\uff0c\u4f18\u5316\u5185\u5b58\u4f7f\u7528\uff0c\u5c24\u5176\u9002\u7528\u4e8e\u8d44\u6e90\u53d7\u9650\u7684\u73af\u5883\u3002<\/li>\n<li><strong>\u672c\u5730\u63a8\u7406<\/strong>\u00a0\uff1a\u5728\u9879\u76ee\u6839\u76ee\u5f55\u4e0b\u7684\u914d\u7f6e\u6587\u4ef6\u00a0<code>config.yaml<\/code>\u00a0\u4e2d\u6307\u5b9a\u5185\u5b58\u548c\u663e\u5b58\u53c2\u6570\uff0c\u5728\u672c\u5730\u684c\u9762\u73af\u5883\u4e0b\u8fdb\u884c\u9ad8\u6548\u63a8\u7406\uff0c\u652f\u630124GB VRAM\u548c150GB DRAM\u3002<\/li>\n<\/ul>\n<h3>\u914d\u7f6e\u8be6\u89e3<\/h3>\n<ol>\n<li><strong>\u914d\u7f6e\u591aGPU<\/strong>\u00a0\uff1a \u7f16\u8f91\u914d\u7f6e\u6587\u4ef6\u00a0<code>config.yaml<\/code>\uff1a<\/li>\n<\/ol>\n<pre><code>gpu:\r\n- id: 0 # GPU \u8bbe\u5907\u7d22\u5f15 0\r\n- id: 1 # GPU \u8bbe\u5907\u7d22\u5f15 1\r\n<\/code><\/pre>\n<ol start=\"2\">\n<li><strong>\u542f\u7528\u7a00\u758f\u6ce8\u610f\u529b<\/strong>\u00a0\uff1a \u5728\u914d\u7f6e\u6587\u4ef6\u4e2d\u6dfb\u52a0\uff1a<\/li>\n<\/ol>\n<pre><code>attention:\r\ntype: sparse\r\n<\/code><\/pre>\n<ol start=\"3\">\n<li><strong>\u672c\u5730\u63a8\u7406\u8bbe\u7f6e<\/strong>\u00a0\uff1a \u5728\u914d\u7f6e\u6587\u4ef6\u4e2d\u6307\u5b9a\u5185\u5b58\u548c\u663e\u5b58\u53c2\u6570\uff1a<\/li>\n<\/ol>\n<pre><code>memory:\r\nvram: 24GB  # \u663e\u5b58\u9650\u5236 (GB)\uff0c\u6839\u636e\u5b9e\u9645\u60c5\u51b5\u8c03\u6574\r\ndram: 150GB # \u5185\u5b58\u9650\u5236 (GB)\uff0c\u6839\u636e\u5b9e\u9645\u60c5\u51b5\u8c03\u6574<\/code><\/pre>\n","protected":false},"excerpt":{"rendered":"<p>KTransformers\uff1a\u4e13\u4e3a\u7a81\u7834\u5927\u6a21\u578b\u63a8\u7406\u74f6\u9888\u800c\u751f\u7684\u9ad8\u6027\u80fd Python \u6846\u67b6\u3002\u00a0\u5b83\u4e0d\u4ec5\u4ec5\u662f\u4e00\u4e2a\u7b80\u5355\u7684\u6a21\u578b\u8fd0\u884c\u5de5\u5177\uff0c\u66f4\u662f\u4e00\u5957\u00a0\u6781\u81f4\u7684\u6027\u80fd\u4f18\u5316\u5f15\u64ce\u00a0\u548c\u00a0\u7075\u6d3b\u7684\u63a5\u53e3\u8d4b\u80fd\u5e73\u53f0\u3002 KTransformers \u81f4\u529b\u4e8e\u4ece\u5e95\u5c42\u63d0\u5347\u5927\u6a21\u578b\u63a8\u7406\u6548\u7387\uff0c\u901a\u8fc7&#8230;<\/p>\n","protected":false},"author":1,"featured_media":61807,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[230],"class_list":["post-20791","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tool","tag-aikaiyuanxiangmu"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts\/20791","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=20791"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts\/20791\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/media\/61807"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/media?parent=20791"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/categories?post=20791"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/tags?post=20791"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}