{"id":65459,"date":"2024-04-19T10:55:15","date_gmt":"2024-04-19T02:55:15","guid":{"rendered":"https:\/\/www.ioiotimes.com\/?p=65459"},"modified":"2024-04-19T10:55:16","modified_gmt":"2024-04-19T02:55:16","slug":"%e5%85%a8%e9%9d%a2%e9%96%8b%e6%94%be%ef%bc%9anvidia%e5%8a%a0%e9%80%9fmeta-llama-3%e4%b8%8a%e7%9a%84%e6%8e%a8%e8%ab%96","status":"publish","type":"post","link":"https:\/\/www.ioiotimes.com\/?p=65459","title":{"rendered":"\u5168\u9762\u958b\u653e\uff1aNVIDIA\u52a0\u901fMeta Llama 3\u4e0a\u7684\u63a8\u8ad6"},"content":{"rendered":"\n<h3 class=\"wp-block-heading\">\u63a1\u7528NVIDIA\u6280\u8853\u69cb\u5efa\u7684Meta\u6700\u65b0\u958b\u653e\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u7d93\u904e\u6700\u4f73\u5316\uff0c\u53ef\u5728\u5f9e\u96f2\u7aef\u548c\u8cc7\u6599\u4e2d\u5fc3\u5230\u908a\u7de3\u548c\u500b\u4eba\u96fb\u8166\u7684NVIDIA GPU\u4e0a\u904b\u884c<\/h3>\n\n\n\n<p>NVIDIA \u4eca\u5929\u5ba3\u5e03\u5c0d\u5176\u6240\u6709\u5e73\u53f0\u9032\u884c\u6700\u4f73\u5316\uff0c\u4ee5\u52a0\u901f\u6700\u65b0\u4e00\u4ee3\u5927\u578b\u8a9e\u8a00\u6a21\u578b\uff08<a href=\"https:\/\/www.nvidia.com\/en-us\/glossary\/large-language-models\/\" target=\"_blank\" rel=\"noreferrer noopener\">LLM<\/a>\uff09<a href=\"https:\/\/ai.meta.com\/blog\/meta-llama-3\/\" target=\"_blank\" rel=\"noreferrer noopener\">Meta Llama 3<\/a>\u3002<\/p>\n\n\n\n<p>\u6b64\u958b\u653e\u5f0f\u6a21\u578b\u8207 NVIDIA <a href=\"https:\/\/blogs.nvidia.com\/blog\/what-is-accelerated-computing\/\" target=\"_blank\" rel=\"noreferrer noopener\">\u52a0\u901f\u904b\u7b97<\/a>\u76f8\u7d50\u5408\uff0c\u4f7f\u958b\u767c\u4eba\u54e1\u3001\u7814\u7a76\u4eba\u54e1\u548c\u4f01\u696d\u80fd\u5920\u5728\u5404\u7a2e\u61c9\u7528\u4e2d\u8ca0\u8cac\u4efb\u5730\u9032\u884c\u5275\u65b0\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1280\" height=\"680\" src=\"https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2024\/04\/20240419-nvidia01.jpg\" alt=\"\" class=\"wp-image-65465\" title=\"\" srcset=\"https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2024\/04\/20240419-nvidia01.jpg 1280w, https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2024\/04\/20240419-nvidia01-300x159.jpg 300w, https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2024\/04\/20240419-nvidia01-1024x544.jpg 1024w, https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2024\/04\/20240419-nvidia01-768x408.jpg 768w\" sizes=\"(max-width: 1280px) 100vw, 1280px\" \/><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p> <\/p>\n\n\n\n<p><strong>\u5728NVIDIA AI\u4e0a\u8a13\u7df4<\/strong><\/p>\n\n\n\n<p>Meta \u7684\u5de5\u7a0b\u5e2b\u5011\u4f7f\u7528\u4e86\u4e00\u500b\u642d\u8f09 24,576 \u500b <a href=\"https:\/\/www.nvidia.com\/zh-tw\/data-center\/h100\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA H100 Tensor \u6838\u5fc3 GPU<\/a> \u7684\u96fb\u8166\u53e2\u96c6\uff0c\u8207 <a href=\"https:\/\/www.nvidia.com\/zh-tw\/networking\/quantum2\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA Quantum-2 InfiniBand<\/a> \u7db2\u8def\u9023\u63a5\uff0c\u5c0d Llama 3 \u9032\u884c\u4e86\u8a13\u7df4\u3002\u5728 NVIDIA \u7684\u652f\u63f4\u4e0b\uff0cMeta \u70ba\u5176\u65d7\u8266\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u8abf\u6574\u4e86\u7db2\u8def\u3001\u8edf\u9ad4\u548c\u6a21\u578b\u67b6\u69cb\u3002<\/p>\n\n\n\n<p>\u70ba\u4e86\u9032\u4e00\u6b65\u63a8\u9032<a href=\"https:\/\/www.nvidia.com\/en-us\/glossary\/generative-ai\/\" target=\"_blank\" rel=\"noreferrer noopener\">\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u6167<\/a>\uff08AI\uff09\u767c\u5c55\uff0cMeta <a href=\"https:\/\/engineering.fb.com\/2024\/03\/12\/data-center-engineering\/building-metas-genai-infrastructure\/\" target=\"_blank\" rel=\"noreferrer noopener\">\u6700\u8fd1\u8aaa\u660e<\/a>\u4e86\u5c07\u5176\u57fa\u790e\u8a2d\u65bd\u64f4\u5c55\u5230 35 \u842c\u500b H100 GPU \u7684\u8a08\u5283\u3002<\/p>\n\n\n\n<p><\/p>\n\n\n\n<p><strong>\u8b93Llama 3\u767c\u63ee\u4f5c\u7528<\/strong><\/p>\n\n\n\n<p>\u5728 NVIDIA GPU \u4e0a\u52a0\u901f\u7684 Llama 3 \u7248\u672c\uff0c\u4eca\u5929\u5df2\u7d93\u53ef\u4ee5\u5728\u96f2\u7aef\u3001\u8cc7\u6599\u4e2d\u5fc3\u3001\u908a\u7de3\u548c\u500b\u4eba\u96fb\u8166\u4e0a\u4f7f\u7528\u3002<\/p>\n\n\n\n<p>\u958b\u767c\u8005\u53ef\u4ee5\u5f9e\u700f\u89bd\u5668\u4e0a\u5617\u8a66\u5728 <a href=\"http:\/\/ai.nvidia.com\" target=\"_blank\" rel=\"noreferrer noopener\">ai.nvidia.com<\/a>\u4e0a\u4f7f\u7528 Llama 3\u3002\u5b83\u88ab\u6253\u5305\u70ba <a href=\"https:\/\/developer.nvidia.com\/blog\/nvidia-nim-offers-optimized-inference-microservices-for-deploying-ai-models-at-scale\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA NIM<\/a> \u5fae\u670d\u52d9\uff0c\u5177\u6709\u6a19\u6e96\u7684\u61c9\u7528\u7a0b\u5f0f\u4ecb\u9762\uff0c\u53ef\u4ee5\u5728\u4efb\u4f55\u5730\u65b9\u90e8\u7f72\u3002<\/p>\n\n\n\n<p>\u4f01\u696d\u53ef\u4ee5\u4f7f\u7528 <a href=\"https:\/\/www.nvidia.com\/zh-tw\/ai-data-science\/products\/nemo\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA NeMo<\/a> \u6839\u64da\u5176\u8cc7\u6599\u5fae\u8abf Llama 3\uff0cNVIDIA NeMo \u662f\u4e00\u500b\u9069\u7528\u65bc\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u7684\u958b\u6e90\u6846\u67b6\uff0c\u4e26\u70ba\u5b89\u5168\u3001\u53d7\u5230\u652f\u63f4\u7684 <a href=\"https:\/\/www.nvidia.com\/zh-tw\/data-center\/products\/ai-enterprise\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA AI Enterprise<\/a> \u5e73\u53f0\u7684\u4e00\u90e8\u5206\u3002\u5ba2\u88fd\u5316\u6a21\u578b\u53ef\u4ee5\u4f7f\u7528 <a href=\"https:\/\/developer.nvidia.com\/blog\/nvidia-tensorrt-llm-supercharges-large-language-model-inference-on-nvidia-h100-gpus\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA TensorRT-LLM<\/a> \u9032\u884c\u63a8\u8ad6\u6700\u4f73\u5316\uff0c\u4e26\u4f7f\u7528 <a href=\"https:\/\/www.nvidia.com\/zh-tw\/ai-data-science\/products\/triton-inference-server\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA Triton \u63a8\u8ad6\u4f3a\u670d\u5668<\/a>\u9032\u884c\u90e8\u7f72\u3002<\/p>\n\n\n\n<p><\/p>\n\n\n\n<p><strong>\u5c07<\/strong><strong> Llama 3 <\/strong><strong>\u5e36\u5165\u8a2d\u5099\u548c\u500b\u4eba\u96fb\u8166\u4e0a<\/strong><strong><\/strong><\/p>\n\n\n\n<p>Llama 3 \u4e5f\u53ef\u4ee5\u5728 <a href=\"https:\/\/www.nvidia.com\/zh-tw\/autonomous-machines\/embedded-systems\/jetson-orin\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA Jetson Orin<\/a> \u4e0a\u904b\u884c\uff0c\u7528\u65bc\u6a5f\u5668\u4eba\u548c\u908a\u7de3\u904b\u7b97\u8a2d\u5099\uff0c\u5275\u5efa\u50cf <a href=\"https:\/\/www.jetson-ai-lab.com\/\" target=\"_blank\" rel=\"noreferrer noopener\">Jetson AI \u5be6\u9a57\u5ba4<\/a>\u4e2d\u4e92\u52d5\u5f0f\u4ee3\u7406\u4eba\u4e00\u6a23\u7684\u61c9\u7528\u3002<\/p>\n\n\n\n<p>\u6b64\u5916\uff0c\u9069\u7528\u65bc\u5de5\u4f5c\u7ad9\u548c\u500b\u4eba\u96fb\u8166\u7684 <a href=\"https:\/\/www.nvidia.com\/en-us\/geforce\/rtx\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA RTX<\/a> \u548c <a href=\"https:\/\/www.nvidia.com\/en-us\/geforce\/graphics-cards\/40-series\/\" target=\"_blank\" rel=\"noreferrer noopener\">GeForce RTX<\/a> GPU\uff0c\u53ef\u4ee5\u52a0\u901f Llama 3 \u7684\u63a8\u8ad6\u901f\u5ea6\u3002\u9019\u4e9b\u7cfb\u7d71\u70ba\u958b\u767c\u8005\u63d0\u4f9b\u4e86\u5168\u7403\u8d85\u904e\u4e00\u5104\u500b NVIDIA \u52a0\u901f\u7cfb\u7d71\u7684\u76ee\u6a19\u3002<\/p>\n\n\n\n<p><\/p>\n\n\n\n<p><strong>\u4f7f\u7528<\/strong><strong> Llama 3 <\/strong><strong>\u7372\u5f97\u6700\u4f73\u6548\u80fd<\/strong><strong><\/strong><\/p>\n\n\n\n<p>\u70ba\u804a\u5929\u6a5f\u5668\u4eba\u90e8\u7f72\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u6642\uff0c\u6700\u4f73\u72c0\u614b\u662f\u5728\u4f4e\u5ef6\u9072\u3001\u826f\u597d\u7684\u8b80\u53d6\u901f\u5ea6\u548c\u6700\u4f73\u7684 GPU \u4f7f\u7528\u4e4b\u9593\u53d6\u5f97\u5e73\u8861\uff0c\u4ee5\u964d\u4f4e\u6210\u672c\u3002<\/p>\n\n\n\n<p>\u9019\u6a23\u7684\u670d\u52d9\u9700\u8981\u4ee5\u5927\u7d04\u6bcf\u79d2 10 \u500b\u8a5e\u5143\u3001\u4f7f\u7528\u8005\u95b1\u8b80\u901f\u5ea6\u5169\u500d\u5de6\u53f3\u7684\u901f\u5ea6\u4f86\u50b3\u9001\u8a5e\u5143\uff0c\u9019\u4e9b\u8a5e\u5143\u5927\u81f4\u76f8\u7576\u65bc\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u7684\u8a5e\u5f59\u3002<\/p>\n\n\n\n<p>\u61c9\u7528\u9019\u4e9b\u6307\u6a19\uff0c\u5728\u4f7f\u7528\u5177\u6709 700 \u5104\u500b\u53c3\u6578\u7684 Llama 3 \u7248\u672c\u7684\u521d\u59cb\u6e2c\u8a66\u4e2d\uff0c\u55ae\u4e00 <a href=\"https:\/\/www.nvidia.com\/zh-tw\/data-center\/h200\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA H200 Tensor \u6838\u5fc3 GPU<\/a> \u6bcf\u79d2\u53ef\u7522\u751f\u7d04 3,000 \u500b\u8a5e\u5143\uff0c\u8db3\u4ee5\u540c\u6642\u70ba\u7d04 300 \u500b\u4f7f\u7528\u8005\u63d0\u4f9b\u670d\u52d9\u3002<\/p>\n\n\n\n<p>\u9019\u610f\u5473\u8457\u4e00\u53f0\u914d\u5099 8 \u500b H200 GPU \u7684 <a href=\"https:\/\/www.nvidia.com\/zh-tw\/data-center\/hgx\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA HGX<\/a> \u4f3a\u670d\u5668\u6bcf\u79d2\u53ef\u63d0\u4f9b 24,000 \u500b\u8a5e\u5143\uff0c\u4e26\u900f\u904e\u540c\u6642\u652f\u63f4 2,400 \u591a\u500b\u4f7f\u7528\u8005\u4f86\u9032\u4e00\u6b65\u6700\u4f73\u5316\u6210\u672c\u3002<\/p>\n\n\n\n<p>\u91dd\u5c0d\u908a\u7de3\u8a2d\u5099\uff0c\u5177\u5099 80 \u5104\u500b\u53c3\u6578\u7684 Llama 3 \u7248\u672c\u5728 Jetson AGX Orin \u4e0a\u6bcf\u79d2\u6700\u591a\u53ef\u751f\u621040 \u500b\u8a5e\u5143\uff0c\u5728 Jetson Orin Nano \u4e0a\u5247\u6bcf\u79d2\u53ef\u751f\u6210 15 \u500b\u8a5e\u5143\u3002<\/p>\n\n\n\n<p><\/p>\n\n\n\n<p><strong>\u63a8\u52d5\u793e\u7fa4\u6a21\u578b<\/strong><\/p>\n\n\n\n<p>\u4f5c\u70ba\u7a4d\u6975\u7684\u958b\u6e90\u8ca2\u737b\u8005\uff0cNVIDIA \u81f4\u529b\u65bc\u6700\u4f73\u5316\u793e\u7fa4\u8edf\u9ad4\uff0c\u5e6b\u52a9\u4f7f\u7528\u8005\u61c9\u5c0d\u6700\u68d8\u624b\u7684\u6311\u6230\u3002\u958b\u6e90\u6a21\u578b\u9084\u53ef\u4ee5\u63d0\u9ad8 AI \u7684\u900f\u660e\u5ea6\uff0c\u8b93\u4f7f\u7528\u8005\u5ee3\u6cdb\u5206\u4eab AI \u5b89\u5168\u6027\u548c\u5f48\u6027\u65b9\u9762\u7684\u5de5\u4f5c\u3002<\/p>\n\n\n\n<p>\u8a73\u7d30\u4e86\u89e3 NVIDIA \u7684 AI \u63a8\u8ad6\u5e73\u53f0\uff0c\u5305\u62ec NIM\u3001TensorRT-LLM \u548c Triton \u5982\u4f55\u4f7f\u7528\u50cf\u662f<a href=\"https:\/\/developer.nvidia.com\/blog\/tune-and-deploy-lora-llms-with-nvidia-tensorrt-llm\/\" target=\"_blank\" rel=\"noreferrer noopener\">\u4f4e\u968e\u9069\u61c9<\/a>\uff08low-rank adaptation\uff09\u7b49\u6700\u5148\u9032\u7684\u6280\u8853\u4f86\u52a0\u901f\u6700\u65b0\u7684\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u3002<\/p>\n\n\n\n<p><\/p>\n\n\n\n<p><\/p>\n\n\n\n<p><\/p>\n\n\n\n<h4 class=\"wp-block-heading has-text-align-right has-very-light-gray-to-cyan-bluish-gray-gradient-background has-background\">\ud83d\udfe6<a href=\"https:\/\/reurl.cc\/bV5Mdr\" data-type=\"link\" data-id=\"https:\/\/reurl.cc\/bV5Mdr\" target=\"_blank\" rel=\"noreferrer noopener\">\u597d\u5eb7\u6d3b\u52d5\u5feb\u4f86!!<\/a> \u6700\u5f37Intel Core Ultra\u96fb\u7af6\u638c\u6a5f\u975e\u300cMSI Claw A1M\u300d\u83ab\u5c6c\u5566~<br>\ud83d\udfe6\u770b\u6587\u7ae0\u3001\u62ff\u597d\u79ae\uff01\u300cMSI B760M Gaming WiFi\u300d\u5e73\u50f9\u8d85\u983b\u4e3b\u6a5f\u677f\uff5e<a href=\"https:\/\/reurl.cc\/6vgWZk\" data-type=\"link\" data-id=\"https:\/\/reurl.cc\/6vgWZk\" target=\"_blank\" rel=\"noreferrer noopener\">\u6d3b\u52d5\u62bd\u7368\u5bb6\u5fae\u661f\u9f8d\u60a0\u904a\u5361<\/a><br>\ud83d\udfe6 <strong><a href=\"https:\/\/www.facebook.com\/ioioTIMES\/posts\/pfbid02M9EG6pXAauMJPBjrFyXRSVbg9EujoTphuiFa51u2KNMXXZH3UAWWzzwjWcF2wjS9l\" target=\"_blank\" rel=\"noreferrer noopener\">\u770b\u6587\u7ae0\u5b8c\u6210\u4efb\u52d9\u5c31\u9001\u4f60 \u300cbe quiet! Pure Rock L P+ Pure Wings 3 PWM 140mm\u7121\u6575\u6563\u71b1\u7d44\u5408\u300d<\/a><\/strong><br>\ud83d\udfe6<strong>\u73fe\u5728\u5c31\u52a0\u5165&nbsp;<a href=\"https:\/\/www.facebook.com\/profile.php?id=100086628162118\" target=\"_blank\" rel=\"noreferrer noopener\">ioioTIMES \u81c9\u66f8\u7c89\u7d72\u5718<\/a>&nbsp;\u66f4\u591a\u4e92\u52d5\u3001\u66f4\u591a\u597d\u5eb7\u650f\u62b5\u52a0!!<\/strong><br>\ud83d\udfe6<strong>\u6211\u5011\u6709<a href=\"https:\/\/today.line.me\/tw\/v2\/publisher\/103117\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">LINE TODAY<\/a>\u983b\u9053\u4e86\uff0c\u5feb\u4f86\u8ffd\u8e2a\u6211\u5011\u5427!!&#8211;\u6700\u65b0\u79d1\u6280\u65b0\u805e \u76e1\u5728\u4f60\u624b<\/strong><\/h4>\n","protected":false},"excerpt":{"rendered":"<p>\u63a1\u7528NVIDIA\u6280\u8853\u69cb\u5efa\u7684Meta\u6700<\/p>\n","protected":false},"author":3,"featured_media":65465,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"rank_math_lock_modified_date":false,"footnotes":""},"categories":[13],"tags":[580,9392,9391,71],"class_list":["post-65459","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news","tag-focus","tag-llm","tag-meta-liama-3","tag-nvidia"],"_links":{"self":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/65459"}],"collection":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=65459"}],"version-history":[{"count":5,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/65459\/revisions"}],"predecessor-version":[{"id":65469,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/65459\/revisions\/65469"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/media\/65465"}],"wp:attachment":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=65459"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=65459"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=65459"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}