{"id":5263,"date":"2024-08-14T18:24:44","date_gmt":"2024-08-14T10:24:44","guid":{"rendered":"https:\/\/blog.coderfan.org\/?p=5263"},"modified":"2024-08-14T18:28:54","modified_gmt":"2024-08-14T10:28:54","slug":"deploy-llama-cpp-in-android-system","status":"publish","type":"post","link":"https:\/\/blog.coderfan.org\/en\/deploy-llama-cpp-in-android-system.html","title":{"rendered":"Android\u7cfb\u7edf\u4e2d\u901a\u8fc7llama.cpp\u90e8\u7f72\u5927\u6a21\u578b"},"content":{"rendered":"<div class='booster-block booster-read-block'>\n                <div class=\"twp-read-time\">\n                \t<i class=\"booster-icon twp-clock\"><\/i> <span>Read Time:<\/span>3 Minute, 49 Second                <\/div>\n\n            <\/div>\n<p class=\"has-text-align-justify\">\u65f6\u81f3\u4eca\u65e5\uff0c\u5927\u6a21\u578b\u5df2\u7ecf\u8fdb\u5165\u5230\u5f88\u591a\u4eba\u7684\u751f\u6d3b\u4e2d\uff0c\u5982GPT4,\u901a\u4e49\u5343\u95ee\u7b49\u3002\u5927\u6a21\u578b\u7684\u67b6\u6784\u4e5f\u4ece\u4e91\u7aef\u5f00\u59cb\u5f80\u7aef\u4e91\u878d\u5408\uff0c\u6700\u7ec8\u5230\u8fbeAll-On-Device\u7684\u67b6\u6784\uff0c\u4e5f\u5c31\u662f\u8d8a\u6765\u8d8a\u591a\u88ab\u63d0\u53ca\u7684EdgeAI\u3002\u800cAndroid\u8bbe\u5907\u4f5c\u4e3aEdge Device\u4e2d\u5360\u636e\u4e3b\u5bfc\u7684\u90e8\u5206\uff0c\u4e5f\u5b58\u5728\u5c06\u5927\u6a21\u578b\u90e8\u7f72\u5230\u5b89\u5353\u8bbe\u5907\u7684\u9700\u6c42\u3002\u4eca\u5929\u8fd9\u7bc7\u6587\u7ae0\u5c06\u4e3a\u5927\u5bb6\u4ecb\u7ecd\u5982\u4f55\u5728\u6211\u4eec\u7684Android\u8bbe\u5907\u4e2d\u901a\u8fc7llama.cpp\u90e8\u7f72\u5927\u6a21\u578b\u3002<\/p>\n\n\n\n<p class=\"has-text-align-justify\">llama.cpp\u662f\u4e00\u5957\u5f00\u6e90\u6846\u67b6\uff0c\u53ef\u4ee5\u8ba9\u4f60\u5728\u81ea\u5df1\u7684\u7535\u8111\u4e0a\u8fd0\u884c\u5927\u578b\u8bed\u8a00\u6a21\u578b\u3002\u6211\u4eec\u53ef\u4ee5\u5728Github\u4e0a\u627e\u5230\u5176\u4ee3\u7801\u4ed3\u5e93\u3002llama.cpp\u4ecd\u65e7\u5904\u4e8e\u79ef\u6781\u5f00\u53d1\u7684\u72b6\u6001\uff0c\u793e\u533a\u76f8\u5f53\u6d3b\u8dc3\uff0c\u5f53\u6211\u4eec\u9047\u5230\u95ee\u9898\u65f6\u53ef\u4ee5\u4ece\u793e\u533a\u4e2d\u83b7\u53d6\u5230\u76f8\u5f53\u591a\u7684\u5e2e\u52a9\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u7f16\u8bd1\u6784\u5efa<\/h3>\n\n\n\n<p class=\"has-text-align-justify\">llama.cpp\u662f\u5b8c\u5168\u7531C\/C++\u7f16\u5199\u7684\u5927\u6a21\u578b\u90e8\u7f72\u5de5\u5177\uff0c\u8fd9\u4e5f\u610f\u5473\u7740\u6211\u4eec\u53ef\u4ee5\u901a\u8fc7CMake\/NDK-build\u5de5\u5177\u76f4\u63a5\u5c06\u5176\u96c6\u6210\u5230Android\u7cfb\u7edf\u4e2d\u3002\u8fd9\u91cc\u6211\u4eec\u7f16\u8bd1\u65f6\u7684\u7248\u672c\u4fe1\u606f\u5982\u4e0b\uff1a<\/p>\n\n\n\n<pre class=\"io-enlighter-pre\"><code class=\"gl\" data-enlighter-language=\"php\" data-enlighter-linenumbers=\"true\" data-enlighter-lineoffset=\"\" data-enlighter-highlight=\"\">Commit\uff1a97bdd26eee11fe109dec00de75690ceef61c03f2\nTag\uff1ab3400<\/code><\/pre>\n\n\n\n<p>\u7f16\u8bd1\u73af\u5883\u4e3aUbuntu 22.04.4 LTS\u3002<\/p>\n\n\n\n<p>1.\u4e0b\u8f7d\u6e90\u7801<\/p>\n\n\n\n<pre class=\"io-enlighter-pre\"><code class=\"gl\" data-enlighter-language=\"php\" data-enlighter-linenumbers=\"true\" data-enlighter-lineoffset=\"\" data-enlighter-highlight=\"\">git clone https:\/\/github.com\/ggerganov\/llama.cpp<\/code><\/pre>\n\n\n\n<p>2.\u914d\u7f6eNDK\u73af\u5883<\/p>\n\n\n\n<p class=\"has-text-align-justify\">\u6839\u636e\u6211\u4eec\u7684\u8fd0\u884c\u73af\u5883\uff0c\u4e0b\u8f7d\u5408\u9002\u7684NDK\u7248\u672c\uff0c\u5177\u4f53\u7248\u672c\u7684\u4e0b\u8f7d\u53ef\u4ee5\u4ece\u8be5\u94fe\u63a5\u4e2d\u67e5\u627e\uff1a<\/p>\n\n\n\n<pre class=\"io-enlighter-pre\"><code class=\"gl\" data-enlighter-language=\"php\" data-enlighter-linenumbers=\"true\" data-enlighter-lineoffset=\"\" data-enlighter-highlight=\"\">https:\/\/github.com\/android\/ndk\/releases<\/code><\/pre>\n\n\n\n<p class=\"has-text-align-justify\">\u5728\u6211\u7684\u8bd5\u9a8c\u4e2d\uff0c\u6211\u4f7f\u7528\u7684NDK\u7248\u672c\u4e3ar25c\u7248\u672c\u3002<\/p>\n\n\n\n<p>3.\u914d\u7f6e\u7f16\u8bd1<\/p>\n\n\n\n<pre class=\"io-enlighter-pre\"><code class=\"gl\" data-enlighter-language=\"php\" data-enlighter-linenumbers=\"true\" data-enlighter-lineoffset=\"\" data-enlighter-highlight=\"\">$ cd llama.cpp\n$ mkdir build-android\n$ cd build-android\n$ export NDK=&lt;your_ndk_directory>\n$ cmake -DCMAKE_TOOLCHAIN_FILE=$NDK\/build\/cmake\/android.toolchain.cmake -DANDROID_ABI=arm64-v8a -DANDROID_PLATFORM=android-${PLATFORM_API} -DCMAKE_C_FLAGS=-march=armv8.4a+dotprod ..\n$ make <\/code><\/pre>\n\n\n\n<p class=\"has-text-align-justify\">\u7f16\u8bd1\u5b8c\u6210\u540e\uff0c\u6211\u4eec\u53ef\u4ee5\u5728build\u76ee\u5f55\u4e0b\u627e\u5230\u751f\u6210\u7684\u4e8c\u8fdb\u5236\u4ea7\u7269\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u5177\u4f53\u4f7f\u7528<\/h3>\n\n\n\n<p class=\"has-text-align-justify\">llama.cpp\u9879\u76ee\u4e2d\u63d0\u4f9b\u4e86CLI\u547d\u4ee4\u884c\u5de5\u5177\uff0c\u53ef\u4ee5\u8ba9\u4f7f\u7528\u8005\u5feb\u901f\u4e0a\u624b\u3002\u5728\u7f16\u8bd1\u5b8c\u6210\u4e4b\u540e\u6211\u4eec\u9700\u8981\u5c06\u76f8\u5173\u4e8c\u8fdb\u5236\u4ea7\u7269\u79fb\u5230\u6211\u4eec\u7684Android\u73af\u5883\u4e2d\u8fd0\u884c\u3002\u6d89\u53ca\u5230\u7684\u4e8c\u8fdb\u5236\u4ea7\u7269\u5305\u62ec\uff1a<\/p>\n\n\n\n<p class=\"has-text-align-justify\">llama-cli\uff1aCLI\u547d\u4ee4\u884c\u5de5\u5177\uff0c\u9700\u8981\u63a8\u9001\u5230\/system\/bin\/\u76ee\u5f55\u4e0b<\/p>\n\n\n\n<p class=\"has-text-align-justify\">libllama.so\uff1a\u4f9d\u8d56\u5e93,\u9700\u8981\u63a8\u9001\u5230\/system\/lib64\uff0864\u4f4d\uff09\u76ee\u5f55\u4e0b<\/p>\n\n\n\n<p class=\"has-text-align-justify\">libggml.so:\u4f9d\u8d56\u5e93\uff0c\u9700\u8981\u63a8\u9001\u5230\/system\/lib64\uff0864\u4f4d\uff09\u76ee\u5f55\u4e0b<\/p>\n\n\n\n<p class=\"has-text-align-justify\">\u4e8b\u5b9e\u4e0a\u9664\u4e86\u4e0a\u8ff0\u6bd4\u8f83\u5173\u952e\u7684\u4e8c\u8fdb\u5236\u4ea7\u7269\uff0c\u8fd8\u6709\u4e00\u4e9bAndroid\u7cfb\u7edf\u63d0\u4f9b\u7684\u57fa\u7840\u4f9d\u8d56\u5e93\uff0c\u5305\u62eclibc.so\u3001libm.so\u3001libdl.so\uff0c\u800c\u8fd9\u4e9b\u662fAndroid\u7cfb\u7edf\u81ea\u5e26\u7684\uff0c\u4e0d\u9700\u8981\u6211\u4eec\u81ea\u884c\u7f16\u8bd1\u3002<\/p>\n\n\n\n<p class=\"has-text-align-justify\">\u5f53\u6211\u4eec\u5728Android\u7cfb\u7edf\u4e2d\u4f7f\u7528\u65f6\uff0c\u8fd8\u9700\u8981\u624b\u52a8\u5c06\u6a21\u578b\u6570\u636e\u5bfc\u5165Android\u7cfb\u7edf\u3002\u5173\u4e8e\u5177\u4f53\u7684\u6a21\u578b\u6211\u4eec\u53ef\u4ee5\u5230huggingface\u4e2d\u8fdb\u884c\u4e0b\u8f7d\uff0c\u9700\u8981\u8bf4\u660e\u7684\u662f\u5e76\u4e0d\u662f\u6240\u6709\u7684\u6a21\u578bllama.cpp\u90fd\u652f\u6301\uff0c\u8fd9\u91cc\u6211\u4eec\u53ef\u4ee5llama.cpp\u9879\u76eeREADME\u4e2d\u770b\u5230\u5176\u6240\u652f\u6301\u7684\u4e00\u4e9b\u4e3b\u6d41\u6a21\u578b\uff1aLLaMa,Falcon,Phi,Gemma,Grok\u7b49\u3002<\/p>\n\n\n\n<p class=\"has-text-align-justify\">\u8fd9\u91cc\u6211\u4eec\u4e0b\u8f7dllama 7B\u6a21\u578b\u76844bit\u91cf\u5316\u7248\u672c\uff0c\u5176\u6570\u636e\u5927\u5c0f\u4e3a3.8G\uff0c\u5982\u4e0b\u56fe\u6240\u793a\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"865\" height=\"729\" src=\"https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image.png\" alt=\"\" class=\"wp-image-5264\" srcset=\"https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image.png 865w, https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-300x253.png 300w, https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-768x647.png 768w, https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-14x12.png 14w\" sizes=\"auto, (max-width: 865px) 100vw, 865px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-justify\">\u4e0b\u8f7d\u4ee5\u540e\u6211\u4eec\u901a\u8fc7adb push\u5230Android\u8bbe\u5907\u7684\/data\/temp\/\u76ee\u5f55\u4e0b\uff0c\u53c2\u8003\u5b98\u65b9\u4f7f\u7528\u793a\u4f8b\uff0c\u6211\u4eec\u52a0\u8f7d\u6a21\u578b\u5e76\u4ee5\u4ea4\u4e92\u6a21\u5f0f\u8fdb\u884c\u6d4b\u8bd5\uff0c\u5982\u4e0b\u6240\u793a\uff1a<\/p>\n\n\n\n<pre class=\"io-enlighter-pre\"><code class=\"gl\" data-enlighter-language=\"php\" data-enlighter-linenumbers=\"true\" data-enlighter-lineoffset=\"\" data-enlighter-highlight=\"\">llama-cli -m \/data\/temp\/llama-2-7b-chat.Q4KS.gguf -p \u201cYou are a helpful assistant\u201d -cnv<\/code><\/pre>\n\n\n\n<p class=\"has-text-align-justify\">\u8fd9\u91cc\u7684\u51e0\u4e2a\u547d\u4ee4\u53c2\u6570\u9700\u8981\u4f5c\u4e00\u4e0b\u8bf4\u660e\uff1a<\/p>\n\n\n\n<p>-m\uff1a\u6307\u5b9aModel<\/p>\n\n\n\n<p>-p\uff1a\u6307\u5b9aPrompt<\/p>\n\n\n\n<p>-cnv\uff1a\u6307\u5b9a\u4e3aconversation\u5bf9\u8bdd\u6a21\u5f0f<\/p>\n\n\n\n<p>\u6307\u4ee4\u6267\u884c\u8fc7\u7a0b\u5982\u4e0b\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"865\" height=\"246\" src=\"https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-1.png\" alt=\"\" class=\"wp-image-5265\" style=\"width:670px;height:auto\" srcset=\"https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-1.png 865w, https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-1-300x85.png 300w, https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-1-768x218.png 768w, https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-1-18x5.png 18w\" sizes=\"auto, (max-width: 865px) 100vw, 865px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-justify\">\u5f53\u6211\u4eec\u770b\u5230>\u51fa\u73b0\u65f6\uff0c\u5c31\u610f\u5473\u7740\u6211\u4eec\u8fdb\u5165\u4e86\u4ea4\u4e92\u6a21\u5f0f\uff0c\u6b64\u65f6\u6211\u4eec\u5c31\u53ef\u4ee5\u50cf\u4e0eGPT4\u5bf9\u8bdd\u90a3\u6837\u4e0e\u5176\u8fdb\u884c\u5bf9\u8bdd\uff0c\u5982\u4e0b\u6240\u793a\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"865\" height=\"532\" src=\"https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-2.png\" alt=\"\" class=\"wp-image-5266\" srcset=\"https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-2.png 865w, https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-2-300x185.png 300w, https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-2-768x472.png 768w, https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-2-18x12.png 18w\" sizes=\"auto, (max-width: 865px) 100vw, 865px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-justify\">\u5f53\u6211\u4eec\u60f3\u7ed3\u675f\u5bf9\u8bdd\u65f6\uff0c\u53ea\u9700\u8981\u901a\u8fc7Ctrl+C\u8fdb\u884c\u7ec8\u7aef\u7ed3\u679c\uff0c\u6b64\u65f6\u6211\u4eec\u8fd8\u53ef\u4ee5\u770b\u5230\u4e00\u4e9b\u5927\u6a21\u578b\u63a8\u7406\u7684\u53c2\u6570\u6307\u6807\uff1a<\/p>\n\n\n\n<pre class=\"io-enlighter-pre\"><code class=\"gl\" data-enlighter-language=\"php\" data-enlighter-linenumbers=\"true\" data-enlighter-lineoffset=\"\" data-enlighter-highlight=\"\">llama_print_timings:        load time =    8891.31 ms\nllama_print_timings:      sample time =      33.95 ms \/   102 runs   (    0.33 ms per token,  3004.15 tokens per second)\nllama_print_timings: prompt eval time =   27400.86 ms \/    51 tokens (  537.27 ms per token,     1.86 tokens per second)\nllama_print_timings:        eval time =   30497.56 ms \/    99 runs   (  308.06 ms per token,     3.25 tokens per second)\nllama_print_timings:       total time =  528586.44 ms \/   150 tokens<\/code><\/pre>\n\n\n\n<p class=\"has-text-align-justify\">\u7b2c\u4e00\u884c\u6253\u5370\u7684\u5185\u5bb9\u4e3a\u6a21\u578b\u52a0\u8f7d\u65f6\u95f4\uff0c\u5728\u8fd9\u91cc\u6a21\u578b\u52a0\u8f7d\u82b1\u8d398891ms\uff1b\u901a\u5e38\u800c\u8a00\u6a21\u578b\u7684\u52a0\u8f7d\u65f6\u95f4\u4e0e\u6a21\u578b\u5927\u5c0f\u6210\u7ebf\u6027\u76f8\u5173\uff1b<\/p>\n\n\n\n<p class=\"has-text-align-justify\">\u7b2c\u4e8c\u884c\u6253\u5370\u7684\u4fe1\u606f\u4e3a\u6a21\u578b\u751f\u6210Token\u7684\u6027\u80fd\u6307\u6807\uff0c\u8fd9\u91cc\u6211\u4eec\u53ef\u4ee5\u770b\u5230Token\u7684\u751f\u6210\u901f\u7387\u4e3a3000 tokens\/s\uff1b<\/p>\n\n\n\n<p class=\"has-text-align-justify\">\u7b2c\u4e09\u884c\u6253\u5370\u7684\u4fe1\u606f\u4e3a\u6a21\u578b\u5904\u7406Prompt\u65f6\u7684\u6027\u80fd\u6307\u6807;\u5355\u4e2aToken\u9700\u8981\u82b1\u8d39537.27ms\u7684\u65f6\u95f4\u53bb\u5904\u7406\uff1b\u6bcf\u79d2\u949f\u53ef\u4ee5\u5904\u74061.86\u4e2aToken\uff1b<\/p>\n\n\n\n<p class=\"has-text-align-justify\">\u7b2c\u56db\u884c\u6253\u5370\u7684\u4fe1\u606f\u4e3a\u6a21\u578b\u8bc4\u4f30\u7684\u6027\u80fd\u6307\u6807\uff0c\u5176\u8868\u660e\u6bcf\u79d2\u949f\u80fd\u591f\u751f\u62103.25\u4e2aToken\uff0c\u8fd9\u610f\u5473\u7740\u5176\u751f\u6210\u6587\u672c\u7684\u901f\u7387\u662f\u5927\u4e8e\u5904\u7406Prompt\u65f6\u7684\u901f\u7387\u7684\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u786c\u4ef6\u52a0\u901f<\/h3>\n\n\n\n<p class=\"has-text-align-justify\">llama.cpp\u9879\u76ee\u539f\u672c\u662f\u7528\u4e8e\u5728\u7eafCPU\u73af\u5883\u4e0b\u5b9e\u73b0\u5927\u6a21\u578b\u7684\u63a8\u7406\uff0c\u8fd9\u5bf9\u8bbe\u5907\u7684CPU\u6d88\u8017\u662f\u5f88\u5927\u7684\uff0c\u5982\u4e0b\u6240\u793a\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"865\" height=\"138\" src=\"https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-3.png\" alt=\"\" class=\"wp-image-5268\" srcset=\"https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-3.png 865w, https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-3-300x48.png 300w, https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-3-768x123.png 768w, https:\/\/blog.coderfan.org\/wp-content\/uploads\/2024\/08\/image-3-18x3.png 18w\" sizes=\"auto, (max-width: 865px) 100vw, 865px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-justify\">\u4ee5\u6211\u7684Android\u865a\u62df\u673a\u4e3a\u4f8b\uff0c\u6211\u4eec\u4e3a\u5176\u542f\u7528\u4e868\u6838\u5fc3\uff0c\u5f53llama-cli\u8fdb\u884c\u63a8\u7406\u65f6\uff0c\u5176CPU\u5360\u7528\u9ad8\u8fbe787%\uff0c\u8fd9\u4e2a\u6570\u636e\u5bf9\u4e8eAndroid\u8bbe\u5907\u662f\u4e0d\u53ef\u63a5\u53d7\u7684\u3002<\/p>\n\n\n\n<p class=\"has-text-align-justify\">\u968f\u7740\u9879\u76ee\u7684\u53d1\u5c55\uff0cllama.cpp\u4e5f\u5f00\u59cb\u652f\u6301GPU\u7b49\u786c\u4ef6\u52a0\u901f\u3002\u5728Android\u7cfb\u7edf\u4e2d\uff0c\u5219\u652f\u6301\u901a\u8fc7Vulkan API\u6765\u5b9e\u73b0GPU\u52a0\u901f\u3002\u9057\u61be\u7684\u662f\uff0c\u5b98\u65b9\u4e00\u76f4\u6ca1\u6709\u63d0\u4f9b\u5b8c\u5584\u7684\u6587\u6863\u6765\u6307\u5bfc\u6784\u5efa\uff0c\u8fd9\u91cc\u5c06\u7ed3\u5408\u6211\u7684\u4e2a\u4eba\u5b9e\u8df5\u6765\u6559\u5927\u5bb6\u5982\u4f55\u901a\u8fc7Vulkan\u652f\u6301GPU\u786c\u4ef6\u52a0\u901f\u3002<\/p>\n\n\n\n<p>\u9996\u5148\u6211\u4eec\u9700\u8981\u5b89\u88c5Vulkan SDK,\u8fd9\u91cc\u4ee5Ubuntu 22\u4e3a\u4f8b\uff1a<\/p>\n\n\n\n<pre class=\"io-enlighter-pre\"><code class=\"gl\" data-enlighter-language=\"php\" data-enlighter-linenumbers=\"true\" data-enlighter-lineoffset=\"\" data-enlighter-highlight=\"\">wget -qO - https:\/\/packages.lunarg.com\/lunarg-signing-key-pub.asc | apt-key add -\nwget -qO \/etc\/apt\/sources.list.d\/lunarg-vulkan-jammy.list https:\/\/packages.lunarg.com\/vulkan\/lunarg-vulkan-jammy.list\napt update -y\napt-get install -y vulkan-sdk\n# \u786e\u8ba4Vulkan SDK\u662f\u5426\u5b89\u88c5\nvulkaninfo<\/code><\/pre>\n\n\n\n<p class=\"has-text-align-justify\">\u5f53\u6211\u4eec\u914d\u7f6e\u597dVulkan SDK\u540e\uff0c\u8fd8\u9700\u8981\u4e0b\u8f7dAndroid NDK\uff0c\u5b89\u5353NDK\u7684\u4e0b\u8f7d\u914d\u7f6e\u6b64\u5904\u4e0d\u8d58\u8ff0\uff0c\u53ef\u4ee5\u76f4\u63a5\u4eceGoogle\u5b98\u65b9\u7f51\u7ad9\u4e0b\u8f7d\uff0c\u4e5f\u53ef\u4ee5\u901a\u8fc7\u547d\u4ee4\u884c\u5de5\u5177sdkManager\u8fdb\u884c\u4e0b\u8f7d\u914d\u7f6e\u3002<\/p>\n\n\n\n<p class=\"has-text-align-justify\">\u8fd9\u91cc\u6211\u4e0b\u8f7d\u4f7f\u7528\u7684NDK\u7248\u672c\u4e3aNDK r25\u7248\u672c\uff0c\u4e0b\u8f7d\u5b8c\u540e\u8bf7\u8bbe\u7f6eNDK\u7684\u73af\u5883\u53d8\u91cf\uff0c\u4ee5\u4fbf\u540e\u7eed\u4f7f\u7528\u3002\u4e4b\u540e\u6211\u4eec\u5f00\u59cb\u7f16\u8bd1\uff1a<\/p>\n\n\n\n<pre class=\"io-enlighter-pre\"><code class=\"gl\" data-enlighter-language=\"php\" data-enlighter-linenumbers=\"true\" data-enlighter-lineoffset=\"\" data-enlighter-highlight=\"\">mkdir build-android &amp;&amp; cd build-android &amp;&amp; cmake -DCMAKE_TOOLCHAIN_FILE=$NDK\/build\/cmake\/android.toolchain.cmake -DANDROID_ABI=arm64-v8a -DANDROID_PLATFORM=latest -DCMAKE_C_FLAGS=-march=armv8.4a+dotprod  -DGGML_VULKAN=1  .. &amp;&amp; make -j4<\/code><\/pre>\n\n\n\n<p>\u5728\u7f16\u8bd1\u8fc7\u7a0b\u4e2d\uff0c\u6211\u4eec\u4f1a\u9047\u5230\u4ee5\u4e0b\u95ee\u9898\uff1a<\/p>\n\n\n\n<p class=\"has-text-align-justify\">1. vulkan-shaders-gen: not found\uff0c\u8fd9\u91cc\u7684\u89e3\u51b3\u529e\u6cd5\u662f\u5148\u7f16\u8bd1\u652f\u6301\u672c\u5730Host\u67b6\u6784\u7684vulkan\u540e\u7aef\uff0c\u6b64\u65f6\u4f1a\u5728build\/bin\/\u76ee\u5f55\u4e0b\u751f\u6210vulkan-shaders-gen\u6587\u4ef6\uff0c\u8be5\u6587\u4ef6\u4f1a\u7528\u4e8e\u751f\u6210Vulkan shader\u3002\u8fd9\u91cc\u6211\u4eec\u9700\u8981\u624b\u52a8\u5c06\u8be5\u6267\u884c\u6587\u4ef6\u7684\u8def\u5f84\u6dfb\u52a0\u5230PATH\u73af\u5883\u53d8\u91cf\uff0c\u4ee5\u4fbf\u540e\u7eedCMake\u5728\u4ea4\u53c9\u7f16\u8bd1Android\u7cfb\u7edf\u7684\u4ea7\u7269\u65f6\u80fd\u591f\u627e\u5230\u8be5\u7a0b\u5e8f\uff1b<\/p>\n\n\n\n<pre class=\"io-enlighter-pre\"><code class=\"gl\" data-enlighter-language=\"php\" data-enlighter-linenumbers=\"true\" data-enlighter-lineoffset=\"\" data-enlighter-highlight=\"\">1.#\u9996\u5148\u5f00\u542fVulkan\u7f16\u8bd1\u914d\u7f6e\uff0c\u6ce8\u610f\u662f\u4e3aHost\u67b6\u6784\u8fdb\u884c\u7f16\u8bd1\u800c\u975e\u4ea4\u53c9\u7f16\u8bd1\ncmake -B build -DCMAKE_BUILD_TYPE=Release -DGGML_VULKAN=1 &amp;&amp; cmake --build build\n2.#\u7f16\u8bd1\u5b8c\u6210\u540e\uff0c\u8bbe\u7f6e\u73af\u5883\u53d8\u91cf\u4ee5\u4fbf\u5bfc\u5165vulkan-shader-gen,LLAMA_CPP_HOME\u6307\u4ee3llama.cpp\u7684\u5de5\u7a0b\u9879\u76ee\nexoprt PATH=$PATH:${LLAMA_CPP_HOME}\/build\/bin\n3.\u7f16\u8bd1Android\nmkdir build-android &amp;&amp; cd build-android &amp;&amp; cmake -DCMAKE_TOOLCHAIN_FILE=$NDK\/build\/cmake\/android.toolchain.cmake -DANDROID_ABI=arm64-v8a -DANDROID_PLATFORM=latest -DCMAKE_C_FLAGS=-march=armv8.4a+dotprod  -DGGML_VULKAN=1  .. &amp;&amp; make -j4<\/code><\/pre>\n\n\n\n<p class=\"has-text-align-justify\">2.llama.cpp\/ggml\/src\/ggml-vulkan.cpp:7:10: fatal error: &#8216;vulkan\/vulkan.hpp&#8217; file not found\uff0c\u8fd9\u91cc\u6211\u4eec\u9700\u8981\u624b\u52a8\u66f4\u65b0vulkan\u76f8\u5173\u7684\u5934\u6587\u4ef6\uff0c\u56e0\u4e3aAndroid NDK\u4e2d\u7684Vulkan SDK\u4e2d\u662f\u6bd4\u8f83\u6ede\u540e\u7684\uff0c\u56e0\u6b64\u4f1a\u62a5\u5934\u6587\u4ef6\u627e\u4e0d\u5230\u76f8\u5173\u7684\u95ee\u9898\u3002\u8fd9\u5c31\u9700\u8981\u6211\u4eec\u624b\u52a8\u66f4\u65b0NDK\u4e2d\u7684Vulkan\u5934\u6587\u4ef6\uff1a<\/p>\n\n\n\n<pre class=\"io-enlighter-pre\"><code class=\"gl\" data-enlighter-language=\"php\" data-enlighter-linenumbers=\"true\" data-enlighter-lineoffset=\"\" data-enlighter-highlight=\"\">1.\u4e0b\u8f7dVulkan Headers\nhttps:\/\/github.com\/KhronosGroup\/Vulkan-Headers\/releases\/tag\/vulkan-sdk-1.3.290.0\n2.\u5c06Vulkan Headers\u66f4\u65b0\u5230NDK\u5185\n${ANDROID_NDK}\/toolchains\/llvm\/prebuilt\/linux-x86_64\/sysroot\/usr\/include\/vk_video\n${ANDROID_NDK}\/toolchains\/llvm\/prebuilt\/linux-x86_64\/sysroot\/usr\/include\/vulkan<\/code><\/pre>\n\n\n\n<p>\u91cd\u65b0\u7f16\u8bd1\u3002<\/p>\n\n\n\n<p class=\"has-text-align-justify\">3.\u5c06llama-cli,libggml.so,libllama.so\u63a8\u9001\u5230Android\u7cfb\u7edf,\u6216\u8005\u5c06libggml.so,libllama.so\u76f8\u5173\u5e93\u901a\u8fc7JNI\u65b9\u5f0f\u96c6\u6210\u5230Android\u7cfb\u7edf\u4e2d\u62a5\u9519libomp.so\uff0c\u8fd9\u662f\u56e0\u4e3alibomp.so\u5e76\u975eAndroid\u7cfb\u7edf\u7684\u5185\u7f6e\u5e93\uff0c\u9700\u8981\u6211\u4eec\u624b\u52a8\u63a8\u5165\u5230\u7cfb\u7edf\u6216\u8005\u96c6\u6210\u5230\u5e94\u7528\u4e2d\uff0c\u6211\u4eec\u53ef\u4ee5\u5728NDK\u4e2d\u627e\u5230\u8be5\u5e93\u3002<\/p>\n\n\n\n<pre class=\"io-enlighter-pre\"><code class=\"gl\" data-enlighter-language=\"php\" data-enlighter-linenumbers=\"true\" data-enlighter-lineoffset=\"\" data-enlighter-highlight=\"\">.\/toolchains\/llvm\/prebuilt\/linux-x86_64\/lib\/clang\/18\/lib\/linux\/i386\/libomp.so\n.\/toolchains\/llvm\/prebuilt\/linux-x86_64\/lib\/clang\/18\/lib\/linux\/riscv64\/libomp.so\n.\/toolchains\/llvm\/prebuilt\/linux-x86_64\/lib\/clang\/18\/lib\/linux\/arm\/libomp.so\n.\/toolchains\/llvm\/prebuilt\/linux-x86_64\/lib\/clang\/18\/lib\/linux\/x86_64\/libomp.so\n.\/toolchains\/llvm\/prebuilt\/linux-x86_64\/lib\/clang\/18\/lib\/linux\/aarch64\/libomp.so<\/code><\/pre>\n\n\n\n<p>\u8fd9\u91cc\u9700\u8981\u627e\u5230\u5bf9\u5e94\u67b6\u6784\u7684\u5e93\u3002<\/p>\n\n\n\n<p>\u4ee5\u4e0a\u662f\u4e00\u4e9b\u5e38\u89c1\u7684\u95ee\u9898\uff0c\u5f53\u6211\u4eec\u89e3\u51b3\u4ee5\u4e0a\u95ee\u9898\u540e\u5c31\u53ef\u4ee5\u5f00\u542fGPU\u52a0\u901f\u3002<\/p>\n\n\n\n<p class=\"has-text-align-justify\">\u5982\u679c\u6211\u4eec\u4f7f\u7528llama-cli,\u53ef\u4ee5\u901a\u8fc7-ngl\u53c2\u6570\u5f00\u542fGPU\u52a0\u901f,\u5982\u679c\u662f\u901a\u8fc7JNI\u7684\u65b9\u5f0f\uff0c\u5219\u9700\u8981\u914d\u7f6en_gpu_layers\u53c2\u6570\u3002<\/p>\n\n\n\n<p>\u5728\u5b9e\u8df5\u8fc7\u7a0b\u4e2d\uff0c\u53d1\u73b0\u9ad8\u901aAdreno\u7cfb\u5217\u7684GPU\u4e0a\u8fd0\u884c\u4f1a\u5b58\u5728\u95ee\u9898\uff0c\u76f8\u5173\u62a5\u9519\u5982\u4e0b\uff1a<\/p>\n\n\n\n<pre class=\"io-enlighter-pre\"><code class=\"gl\" data-enlighter-language=\"php\" data-enlighter-linenumbers=\"true\" data-enlighter-lineoffset=\"\" data-enlighter-highlight=\"\">08-10 16:06:09.099 30852 30926 I AdrenoVK-0: Failed to link shaders.\n08-10 16:06:09.099 30852 30926 I AdrenoVK-0: Pipeline create failed\n08-10 16:06:09.108 30852 30926 E LLama-android: llama_model_load: error loading model: vk::Device::createComputePipeline: ErrorUnknown\n08-10 16:06:09.108 30852 30926 E LLama-android: llama_load_model_from_file: failed to load model<\/code><\/pre>\n        <div class=\"booster-block booster-reactions-block\">\n            <div class=\"twp-reactions-icons\">\n                \n                <div class=\"twp-reacts-wrap\">\n                    <a react-data=\"be-react-1\" post-id=\"5263\" class=\"be-face-icons un-reacted\" href=\"javascript:void(0)\">\n                        <img decoding=\"async\" src=\"https:\/\/blog.coderfan.org\/wp-content\/plugins\/booster-extension\/\/assets\/icon\/happy.svg\" alt=\"Happy\">\n                    <\/a>\n                    <div class=\"twp-reaction-title\">\n                        Happy                    <\/div>\n                    <div class=\"twp-count-percent\">\n                                                    <span style=\"display: none;\" class=\"twp-react-count\">0<\/span>\n                        \n                                                <span class=\"twp-react-percent\"><span>0<\/span> %<\/span>\n                                            <\/div>\n                <\/div>\n\n                <div class=\"twp-reacts-wrap\">\n                    <a react-data=\"be-react-2\" post-id=\"5263\" class=\"be-face-icons un-reacted\" href=\"javascript:void(0)\">\n                        <img decoding=\"async\" src=\"https:\/\/blog.coderfan.org\/wp-content\/plugins\/booster-extension\/\/assets\/icon\/sad.svg\" alt=\"Sad\">\n                    <\/a>\n                    <div class=\"twp-reaction-title\">\n                        Sad                    <\/div>\n                    <div class=\"twp-count-percent\">\n                                                    <span style=\"display: none;\" class=\"twp-react-count\">0<\/span>\n                                                                        <span class=\"twp-react-percent\"><span>0<\/span> %<\/span>\n                                            <\/div>\n                <\/div>\n\n                <div class=\"twp-reacts-wrap\">\n                    <a react-data=\"be-react-3\" post-id=\"5263\" class=\"be-face-icons un-reacted\" href=\"javascript:void(0)\">\n                        <img decoding=\"async\" src=\"https:\/\/blog.coderfan.org\/wp-content\/plugins\/booster-extension\/\/assets\/icon\/excited.svg\" alt=\"Excited\">\n                    <\/a>\n                    <div class=\"twp-reaction-title\">\n                        Excited                    <\/div>\n                    <div class=\"twp-count-percent\">\n                                                    <span style=\"display: none;\" class=\"twp-react-count\">0<\/span>\n                                                                        <span class=\"twp-react-percent\"><span>0<\/span> %<\/span>\n                                            <\/div>\n                <\/div>\n\n                <div class=\"twp-reacts-wrap\">\n                    <a react-data=\"be-react-6\" post-id=\"5263\" class=\"be-face-icons un-reacted\" href=\"javascript:void(0)\">\n                        <img decoding=\"async\" src=\"https:\/\/blog.coderfan.org\/wp-content\/plugins\/booster-extension\/\/assets\/icon\/sleepy.svg\" alt=\"Sleepy\">\n                    <\/a>\n                    <div class=\"twp-reaction-title\">\n                        Sleepy                    <\/div>\n                    <div class=\"twp-count-percent\">\n                                                    <span style=\"display: none;\" class=\"twp-react-count\">0<\/span>\n                        \n                                                <span class=\"twp-react-percent\"><span>0<\/span> %<\/span>\n                                            <\/div>\n                <\/div>\n\n                <div class=\"twp-reacts-wrap\">\n                    <a react-data=\"be-react-4\" post-id=\"5263\" class=\"be-face-icons un-reacted\" href=\"javascript:void(0)\">\n                        <img decoding=\"async\" src=\"https:\/\/blog.coderfan.org\/wp-content\/plugins\/booster-extension\/\/assets\/icon\/angry.svg\" alt=\"Angry\">\n                    <\/a>\n                    <div class=\"twp-reaction-title\">Angry<\/div>\n                    <div class=\"twp-count-percent\">\n                                                    <span style=\"display: none;\" class=\"twp-react-count\">0<\/span>\n                                                                        <span class=\"twp-react-percent\"><span>0<\/span> %<\/span>\n                        \n                    <\/div>\n                <\/div>\n\n                <div class=\"twp-reacts-wrap\">\n                    <a react-data=\"be-react-5\" post-id=\"5263\" class=\"be-face-icons un-reacted\" href=\"javascript:void(0)\">\n                        <img decoding=\"async\" src=\"https:\/\/blog.coderfan.org\/wp-content\/plugins\/booster-extension\/\/assets\/icon\/surprise.svg\" alt=\"Surprise\">\n                    <\/a>\n                    <div class=\"twp-reaction-title\">Surprise<\/div>\n                    <div class=\"twp-count-percent\">\n                                                    <span style=\"display: none;\" class=\"twp-react-count\">0<\/span>\n                                                                        <span class=\"twp-react-percent\"><span>0<\/span> %<\/span>\n                                            <\/div>\n                <\/div>\n\n            <\/div>\n        <\/div>","protected":false},"excerpt":{"rendered":"<p>\u5728Android\u7cfb\u7edf\u4e2d\u90e8\u7f72llama.cpp\u7684\u6d41\u7a0b\u4e0e\u95ee\u9898\u8bb0\u5f55<\/p>","protected":false},"author":1,"featured_media":4483,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_editorskit_title_hidden":false,"_editorskit_reading_time":0,"_editorskit_is_block_options_detached":false,"_editorskit_block_options_position":"{}","om_disable_all_campaigns":false,"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"footnotes":""},"categories":[18],"tags":[341,343,342],"class_list":["post-5263","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-operating-system","tag-ai","tag-llama-cpp","tag-llm"],"blocksy_meta":[],"aioseo_notices":[],"featured_image_urls":{"full":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android.jpg",512,269,false],"thumbnail":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android-150x150.jpg",150,150,true],"medium":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android-300x158.jpg",300,158,true],"medium_large":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android.jpg",512,269,false],"large":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android.jpg",512,269,false],"1536x1536":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android.jpg",512,269,false],"2048x2048":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android.jpg",512,269,false],"trp-custom-language-flag":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android-18x9.jpg",18,9,true]},"post_excerpt_stackable":"<p>\u5728Android\u7cfb\u7edf\u4e2d\u90e8\u7f72llama.cpp\u7684\u6d41\u7a0b\u4e0e\u95ee\u9898\u8bb0\u5f55<\/p>\n","category_list":"<a href=\"https:\/\/blog.coderfan.org\/en\/category\/operating-system\" rel=\"category tag\">\u64cd\u4f5c\u7cfb\u7edf<\/a>","author_info":{"name":"FranzKafka95","url":"https:\/\/blog.coderfan.org\/en\/author\/yushenglonely95"},"comments_num":"0 comments","featured_image_urls_v2":{"full":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android.jpg",512,269,false],"thumbnail":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android-150x150.jpg",150,150,true],"medium":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android-300x158.jpg",300,158,true],"medium_large":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android.jpg",512,269,false],"large":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android.jpg",512,269,false],"1536x1536":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android.jpg",512,269,false],"2048x2048":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android.jpg",512,269,false],"trp-custom-language-flag":["https:\/\/blog.coderfan.org\/wp-content\/uploads\/2023\/04\/android-18x9.jpg",18,9,true]},"post_excerpt_stackable_v2":"<p>\u5728Android\u7cfb\u7edf\u4e2d\u90e8\u7f72llama.cpp\u7684\u6d41\u7a0b\u4e0e\u95ee\u9898\u8bb0\u5f55<\/p>\n","category_list_v2":"<a href=\"https:\/\/blog.coderfan.org\/en\/category\/operating-system\" rel=\"category tag\">\u64cd\u4f5c\u7cfb\u7edf<\/a>","author_info_v2":{"name":"FranzKafka95","url":"https:\/\/blog.coderfan.org\/en\/author\/yushenglonely95"},"comments_num_v2":"0 comments","_links":{"self":[{"href":"https:\/\/blog.coderfan.org\/en\/wp-json\/wp\/v2\/posts\/5263","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/blog.coderfan.org\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blog.coderfan.org\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blog.coderfan.org\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/blog.coderfan.org\/en\/wp-json\/wp\/v2\/comments?post=5263"}],"version-history":[{"count":5,"href":"https:\/\/blog.coderfan.org\/en\/wp-json\/wp\/v2\/posts\/5263\/revisions"}],"predecessor-version":[{"id":5273,"href":"https:\/\/blog.coderfan.org\/en\/wp-json\/wp\/v2\/posts\/5263\/revisions\/5273"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/blog.coderfan.org\/en\/wp-json\/wp\/v2\/media\/4483"}],"wp:attachment":[{"href":"https:\/\/blog.coderfan.org\/en\/wp-json\/wp\/v2\/media?parent=5263"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.coderfan.org\/en\/wp-json\/wp\/v2\/categories?post=5263"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.coderfan.org\/en\/wp-json\/wp\/v2\/tags?post=5263"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}