{"id":2549,"date":"2025-07-18T15:29:10","date_gmt":"2025-07-18T07:29:10","guid":{"rendered":"https:\/\/sanlangcode.com\/?p=2549"},"modified":"2025-07-18T15:29:10","modified_gmt":"2025-07-18T07:29:10","slug":"%f0%9f%a6%be-%e5%a4%a7%e6%a8%a1%e5%9e%8b%e6%90%ac%e5%ae%b6%e8%ae%b0%ef%bc%9a%e4%bb%8e%e7%a1%ac%e7%9b%98%e5%88%b0-gpu-cpu-%e7%9a%84%e5%a5%87%e5%b9%bb%e6%bc%82%e6%b5%81","status":"publish","type":"post","link":"https:\/\/sanlangcode.com\/index.php\/2025\/07\/18\/%f0%9f%a6%be-%e5%a4%a7%e6%a8%a1%e5%9e%8b%e6%90%ac%e5%ae%b6%e8%ae%b0%ef%bc%9a%e4%bb%8e%e7%a1%ac%e7%9b%98%e5%88%b0-gpu-cpu-%e7%9a%84%e5%a5%87%e5%b9%bb%e6%bc%82%e6%b5%81\/","title":{"rendered":"\ud83e\uddbe \u5927\u6a21\u578b\u201c\u642c\u5bb6\u8bb0\u201d\uff1a\u4ece\u786c\u76d8\u5230 GPU\/CPU \u7684\u5947\u5e7b\u6f02\u6d41"},"content":{"rendered":"\n<p><\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201c\u4f60\u4ee5\u4e3a\u6a21\u578b\u53ea\u662f\u4e2a\u6587\u4ef6\uff1f\u5176\u5b9e\u5b83\u662f\u4e2a\u2018\u5de8\u65e0\u9738\u2019\uff01\u201d<br>\u2014\u2014 \u4e00\u4f4d\u88ab CUDA out of memory \u652f\u914d\u8fc7\u7684\u5f00\u53d1\u8005<\/p>\n<\/blockquote>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u76ee\u5f55<\/h2>\n\n\n\n<ol class=\"wp-block-list\">\n<li><a href=\"#\u5927\u6a21\u578b\u7684\u642c\u5bb6\u8def\u7ebf\u56fe\">\u5927\u6a21\u578b\u7684\u201c\u642c\u5bb6\u201d\u8def\u7ebf\u56fe<\/a><\/li>\n\n\n\n<li><a href=\"#\u642c\u8fdb-gpu-\u7684\u90a3\u4e9b\u4e8b\u513f\">\u642c\u8fdb GPU \u7684\u90a3\u4e9b\u4e8b\u513f<\/a>\n<ul class=\"wp-block-list\">\n<li><a href=\"#gpu-\u5e38\u89c1\u95ee\u9898\">\u5e38\u89c1\u95ee\u9898<\/a><\/li>\n\n\n\n<li><a href=\"#gpu-\u52a0\u8f7d\u4ee3\u7801\u793a\u4f8b\">\u4ee3\u7801\u793a\u4f8b<\/a><\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><a href=\"#\u642c\u8fdb-cpu-\u7684\u90a3\u4e9b\u4e8b\u513f\">\u642c\u8fdb CPU \u7684\u90a3\u4e9b\u4e8b\u513f<\/a>\n<ul class=\"wp-block-list\">\n<li><a href=\"#cpu-\u5e38\u89c1\u95ee\u9898\">\u5e38\u89c1\u95ee\u9898<\/a><\/li>\n\n\n\n<li><a href=\"#cpu-\u52a0\u8f7d\u4ee3\u7801\u793a\u4f8b\">\u4ee3\u7801\u793a\u4f8b<\/a><\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><a href=\"#\u5e38\u89c1\u642c\u5bb6\u4e8b\u6545\u5904\u7406\u6307\u5357\">\u5e38\u89c1\u201c\u642c\u5bb6\u201d\u4e8b\u6545\u5904\u7406\u6307\u5357<\/a><\/li>\n\n\n\n<li><a href=\"#\u7ed3\u8bed\">\u7ed3\u8bed<\/a><\/li>\n<\/ol>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u5927\u6a21\u578b\u7684\u201c\u642c\u5bb6\u201d\u8def\u7ebf\u56fe<\/h2>\n\n\n\n<p>\u8ba9\u6211\u4eec\u7528\u4e00\u5f20\u201c\u642c\u5bb6\u201d\u6d41\u7a0b\u56fe\u6765\u5f62\u8c61\u7406\u89e3\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>flowchart TD\n    A&#91;\u786c\u76d8\u4e0a\u7684\u6a21\u578b\u6743\u91cd\u6587\u4ef6] --&gt; B&#91;\u52a0\u8f7d\u5230CPU\u5185\u5b58]\n    B --&gt; C{\u53bb\u54ea\uff1f}\n    C -- GPU\u6709\u94b1\u6709\u95f2 --&gt; D&#91;\u642c\u5230GPU\u663e\u5b58]\n    C -- GPU\u592a\u7a77\/\u592a\u5fd9 --&gt; E&#91;\u7559\u5728CPU\u5185\u5b58]\n    D &amp; E --&gt; F&#91;\u5f00\u59cb\u63a8\u7406\/\u8bad\u7ec3]<\/code><\/pre>\n\n\n\n<p>\u7b80\u800c\u8a00\u4e4b\uff0c\u6a21\u578b\u7684\u201c\u642c\u5bb6\u201d\u5206\u4e3a\u4e24\u6b65\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u5148\u4ece\u786c\u76d8\u642c\u5230 CPU \u5185\u5b58\u3002<\/li>\n\n\n\n<li>\u518d\u51b3\u5b9a\u662f\u642c\u5230 GPU \u663e\u5b58\uff0c\u8fd8\u662f\u7ee7\u7eed\u7559\u5728 CPU\u3002<\/li>\n<\/ol>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u642c\u8fdb GPU \u7684\u90a3\u4e9b\u4e8b\u513f<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">\u4e3a\u4ec0\u4e48\u8981\u642c\u8fdb GPU\uff1f<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>GPU \u662f\u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u808c\u8089\u7537\u201d\uff0c\u5e76\u884c\u8ba1\u7b97\u80fd\u529b\u7206\u8868\uff0c\u63a8\u7406\/\u8bad\u7ec3\u901f\u5ea6\u98de\u5feb\u3002<\/li>\n\n\n\n<li>\u4f46\u808c\u8089\u7537\u7684\u201c\u623f\u95f4\u201d\uff08\u663e\u5b58\uff09\u5f88\u8d35\uff0c\u7a7a\u95f4\u6709\u9650\u3002\uff08\u5fc5\u987b\u5b8c\u6574\u8fde\u7eed\u7a7a\u95f4\uff09<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">GPU \u5e38\u89c1\u95ee\u9898<\/h3>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u663e\u5b58\u4e0d\u591f\uff0c\u642c\u4e0d\u8fdb\u53bb\uff01<\/strong><\/li>\n<\/ol>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u62a5\u9519\uff1a<code>CUDA out of memory<\/code><\/li>\n\n\n\n<li>\u89e3\u51b3\uff1a\u91cf\u5316\u3001\u5206\u5e03\u5f0f\u3001\u5206\u5c42\u52a0\u8f7d\u3001\u4e70\u65b0\u5361\uff08\u6709\u94b1\u771f\u597d\uff09<\/li>\n<\/ul>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u591a\u5361\u642c\u5bb6\uff0c\u8c01\u4f4f\u54ea\uff1f<\/strong><\/li>\n<\/ol>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u9700\u8981\u7528 <code>device_map=\"auto\"<\/code> \u6216\u624b\u52a8\u5206\u914d<\/li>\n\n\n\n<li>\u8fd8\u53ef\u4ee5\u7528 DeepSpeed\u3001Accelerate \u7b49\u201c\u642c\u5bb6\u516c\u53f8\u201d<\/li>\n<\/ul>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u6a21\u578b\u592a\u5927\uff0cGPU\u592a\u5c0f\uff0c\u600e\u4e48\u529e\uff1f<\/strong><\/li>\n<\/ol>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u91cf\u5316\uff08int8\/int4\uff09\uff0c\u6a21\u578b\u201c\u7626\u8eab\u201d<\/li>\n\n\n\n<li>\u90e8\u5206\u6743\u91cd\u653eCPU\uff0c\u63a8\u7406\u65f6\u201c\u4e34\u65f6\u642c\u8fd0\u201d<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">GPU \u52a0\u8f7d\u4ee3\u7801\u793a\u4f8b<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code>from transformers import AutoModelForCausalLM, AutoTokenizer\n\n# 1. \u52a0\u8f7d\u5206\u8bcd\u5668\ntokenizer = AutoTokenizer.from_pretrained(\"your-model-path\")\n\n# 2. \u52a0\u8f7d\u6a21\u578b\u5230GPU\nmodel = AutoModelForCausalLM.from_pretrained(\n    \"your-model-path\",\n    device_map=\"cuda\",  # \u6216 \"auto\" \u81ea\u52a8\u5206\u914d\u591a\u5361\n    torch_dtype=\"auto\"  # \u81ea\u52a8\u9009\u62e9\u7cbe\u5ea6\n)<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">\u91cf\u5316\u52a0\u8f7d\uff08\u8ba9\u6a21\u578b\u201c\u51cf\u80a5\u201d\uff09<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code>from transformers import BitsAndBytesConfig\n\nbnb_config = BitsAndBytesConfig(load_in_4bit=True)\nmodel = AutoModelForCausalLM.from_pretrained(\n    \"your-model-path\",\n    device_map=\"cuda\",\n    quantization_config=bnb_config\n)<\/code><\/pre>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u642c\u8fdb CPU \u7684\u90a3\u4e9b\u4e8b\u513f<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">\u4e3a\u4ec0\u4e48\u8981\u642c\u8fdb CPU\uff1f<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6ca1\u6709 GPU\uff1f\u6ca1\u5173\u7cfb\uff0cCPU \u4e5f\u80fd\u201c\u6536\u7559\u201d\u6a21\u578b\u3002<\/li>\n\n\n\n<li>\u9002\u5408\u5c0f\u6a21\u578b\u3001\u6d4b\u8bd5\u3001\u5f00\u53d1\u73af\u5883\u3002<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">CPU \u5e38\u89c1\u95ee\u9898<\/h3>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u901f\u5ea6\u611f\u4eba\uff0c\u50cf\u8717\u725b\u642c\u5bb6<\/strong><\/li>\n<\/ol>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u63a8\u7406\/\u8bad\u7ec3\u901f\u5ea6\u8fdc\u6162\u4e8e GPU<\/li>\n\n\n\n<li>\u9002\u5408\u201c\u4f5b\u7cfb\u201d\u7528\u6237<\/li>\n<\/ul>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u5185\u5b58\u4e5f\u6709\u9650\uff0c\u522b\u592a\u5927\u610f<\/strong><\/li>\n<\/ol>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5927\u6a21\u578b\u7167\u6837\u80fd\u628a\u5185\u5b58\u5403\u5149<\/li>\n\n\n\n<li>\u62a5\u9519\uff1a<code>Killed<\/code> \u6216 <code>MemoryError<\/code><\/li>\n<\/ul>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u591a\u7ebf\u7a0b\/\u591a\u8fdb\u7a0b\u51b2\u7a81<\/strong><\/li>\n<\/ol>\n\n\n\n<ul class=\"wp-block-list\">\n<li>CPU \u63a8\u7406\u65f6\uff0c\u7ebf\u7a0b\u6570\u8bbe\u7f6e\u4e0d\u5f53\u4f1a\u5bfc\u81f4\u6548\u7387\u4f4e\u4e0b\u6216\u5d29\u6e83<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">CPU \u52a0\u8f7d\u4ee3\u7801\u793a\u4f8b<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code>from transformers import AutoModelForCausalLM, AutoTokenizer\n\ntokenizer = AutoTokenizer.from_pretrained(\"your-model-path\")\nmodel = AutoModelForCausalLM.from_pretrained(\n    \"your-model-path\",\n    device_map=\"cpu\"  # \u660e\u786e\u6307\u5b9a\u7528CPU\n)<\/code><\/pre>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u5e38\u89c1\u201c\u642c\u5bb6\u201d\u4e8b\u6545\u5904\u7406\u6307\u5357<\/h2>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>\u95ee\u9898\u73b0\u8c61<\/th><th>\u53ef\u80fd\u539f\u56e0<\/th><th>\u89e3\u51b3\u529e\u6cd5<\/th><\/tr><\/thead><tbody><tr><td>CUDA out of memory<\/td><td>\u663e\u5b58\u4e0d\u8db3<\/td><td>\u91cf\u5316\u3001\u5206\u5e03\u5f0f\u3001\u51cf\u5c0f batch\u3001\u4e70\u65b0\u5361<\/td><\/tr><tr><td>Killed\/MemoryError<\/td><td>CPU \u5185\u5b58\u4e0d\u8db3<\/td><td>\u6362\u5927\u5185\u5b58\u3001\u51cf\u5c0f\u6a21\u578b\u3001\u5206\u6279\u5904\u7406<\/td><\/tr><tr><td>\u52a0\u8f7d\u6162\/\u63a8\u7406\u6162<\/td><td>CPU \u6027\u80fd\u6709\u9650<\/td><td>\u6362 GPU\u3001\u51cf\u5c11\u5e76\u53d1\u3001\u4f18\u5316\u4ee3\u7801<\/td><\/tr><tr><td>\u6743\u91cd\u6587\u4ef6\u635f\u574f\/\u4e0d\u517c\u5bb9<\/td><td>\u4e0b\u8f7d\u4e0d\u5b8c\u6574\/\u7248\u672c\u4e0d\u5339\u914d<\/td><td>\u91cd\u65b0\u4e0b\u8f7d\u3001\u68c0\u67e5 transformers \u7248\u672c<\/td><\/tr><tr><td>device_map \u62a5\u9519<\/td><td>\u5206\u914d\u65b9\u5f0f\u4e0d\u5bf9<\/td><td>\u7528 &#8220;auto&#8221; \u6216\u624b\u52a8\u5206\u914d<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u7ed3\u8bed<\/h2>\n\n\n\n<p>\u5927\u6a21\u578b\u201c\u642c\u5bb6\u201d\u5176\u5b9e\u5c31\u662f\u4e00\u573a\u201c\u5185\u5b58\u4e89\u593a\u6218\u201d\uff01<br>GPU \u662f\u8c6a\u5b85\uff0c\u4f4f\u5f97\u8212\u670d\u4f46\u623f\u79df\u8d35\uff08\u663e\u5b58\u6709\u9650\uff09\uff1bCPU \u662f\u8001\u65e7\u516c\u5bd3\uff0c\u4fbf\u5b9c\u4f46\u62e5\u6324\uff08\u6162\u4e14\u6613\u7206\u5185\u5b58\uff09\u3002<br>\u60f3\u8ba9\u6a21\u578b\u5b89\u5bb6\uff0c\u8bb0\u5f97\u63d0\u524d\u91cf\u597d\u201c\u5bb6\u5177\u5c3a\u5bf8\u201d\uff08\u6a21\u578b\u5927\u5c0f\uff09\uff0c\u9009\u597d\u201c\u623f\u5b50\u201d\uff08\u786c\u4ef6\uff09\uff0c\u522b\u8ba9\u6a21\u578b\u201c\u6d41\u6d6a\u8857\u5934\u201d\uff01<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201c\u6a21\u578b\u52a0\u8f7d\u6210\u529f\uff0c\u63a8\u7406\u5982\u98de\uff0c\u8001\u677f\u52a0\u85aa\uff01\u201d<br>\u2014\u2014 \u4f60\uff0c\u672a\u6765\u7684 AI \u5de5\u7a0b\u5e08<\/p>\n<\/blockquote>\n","protected":false},"excerpt":{"rendered":"<p>\u201c\u4f60\u4ee5\u4e3a\u6a21\u578b\u53ea\u662f\u4e2a\u6587\u4ef6\uff1f\u5176\u5b9e\u5b83\u662f&#46;&#46;&#46;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[47,46,74,4],"tags":[],"class_list":["post-2549","post","type-post","status-publish","format-standard","hentry","category-cuda","category-pytorch","category-rag","category-4"],"_links":{"self":[{"href":"https:\/\/sanlangcode.com\/index.php\/wp-json\/wp\/v2\/posts\/2549","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/sanlangcode.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/sanlangcode.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/sanlangcode.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/sanlangcode.com\/index.php\/wp-json\/wp\/v2\/comments?post=2549"}],"version-history":[{"count":1,"href":"https:\/\/sanlangcode.com\/index.php\/wp-json\/wp\/v2\/posts\/2549\/revisions"}],"predecessor-version":[{"id":2550,"href":"https:\/\/sanlangcode.com\/index.php\/wp-json\/wp\/v2\/posts\/2549\/revisions\/2550"}],"wp:attachment":[{"href":"https:\/\/sanlangcode.com\/index.php\/wp-json\/wp\/v2\/media?parent=2549"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/sanlangcode.com\/index.php\/wp-json\/wp\/v2\/categories?post=2549"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/sanlangcode.com\/index.php\/wp-json\/wp\/v2\/tags?post=2549"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}