{"id":7100,"date":"2024-08-05T17:05:30","date_gmt":"2024-08-05T09:05:30","guid":{"rendered":"https:\/\/aict.nkust.edu.tw\/digitrans\/?p=7100"},"modified":"2024-12-13T21:10:28","modified_gmt":"2024-12-13T13:10:28","slug":"%e5%a6%82%e4%bd%95%e5%9c%a8%e9%9b%b2%e7%ab%af%e9%83%a8%e7%bd%b2-apple-%e7%9a%84-dclm-7b","status":"publish","type":"post","link":"https:\/\/aict.nkust.edu.tw\/digitrans\/?p=7100","title":{"rendered":"\u5982\u4f55\u5728\u96f2\u7aef\u90e8\u7f72 Apple \u7684 DCLM-7B"},"content":{"rendered":"\n<p>2024-08-05 | Ayush Kumar<\/p>\n\n\n\n<p>Apple \u767c\u5e03\u4e86 DCLM\uff0c\u9019\u662f\u4e00\u500b 70 \u5104\u53c3\u6578\u7684\u958b\u6e90\u8a9e\u8a00\u6a21\u578b\uff0c\u5728\u958b\u6e90 AI \u9818\u57df\u9081\u51fa\u4e86\u91cd\u5927\u4e00\u6b65\u3002<\/p>\n\n\n\n<p>DCLM-Baseline-7B \u662f\u4e00\u500b\u5728 DCLM-Baseline \u8cc7\u6599\u96c6\u4e0a\u8a13\u7df4\u7684 70 \u5104\u53c3\u6578\u8a9e\u8a00\u6a21\u578b\uff0c\u8a72\u8cc7\u6599\u96c6\u662f\u4f5c\u70ba DataComp for Language Models (DCLM) \u57fa\u6e96\u6e2c\u8a66\u7684\u4e00\u90e8\u5206\u800c\u88fd\u5b9a\u7684\u3002\u8a72\u6a21\u578b\u65e8\u5728\u5c55\u793a\u7cfb\u7d71\u8cc7\u6599\u7ba1\u7406\u6280\u8853\u5728\u63d0\u9ad8\u8a9e\u8a00\u6a21\u578b\u6548\u80fd\u65b9\u9762\u7684\u6709\u6548\u6027\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"model-details\">\u578b\u865f\u8a73\u60c5<\/h3>\n\n\n\n<figure class=\"wp-block-table aligncenter\"><table class=\"has-fixed-layout\"><thead><tr><th>Size<\/th><th>Training Tokens<\/th><th>Layers<\/th><th>Hidden Size<\/th><th>Attention Heads<\/th><th>Context Length<\/th><\/tr><\/thead><tbody><tr><td>7B<\/td><td>2.5T<\/td><td>32<\/td><td>4096<\/td><td>32<\/td><td>2048<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"key-points-of-dclm\">DCLM\u7684\u8981\u9ede\uff1a<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u6a21\u578b\u898f\u683c<\/strong>\uff1a7B \u57fa\u672c\u6a21\u578b\u5728 2.5 \u5146\u500b\u6a19\u8a18\u4e0a\u9032\u884c\u8a13\u7df4\uff0c\u4e3b\u8981\u4f7f\u7528\u5177\u6709 2048 \u500b\u4e0a\u4e0b\u6587\u8996\u7a97\u7684\u82f1\u8a9e\u8cc7\u6599\u3002<\/li>\n\n\n\n<li><strong>\u8a13\u7df4\u8cc7\u6599<\/strong>\uff1a\u7d44\u5408\u4f86\u81ea DCLM-BASELINE\u3001StarCoder \u548c ProofPile2 \u7684\u8cc7\u6599\u96c6\u3002<\/li>\n\n\n\n<li><strong>\u6027\u80fd<\/strong>\uff1a\u6a21\u578b\u7684 MMLU \u5f97\u5206\u70ba 0.6372\uff0c\u5176\u6027\u80fd\u9ad8\u65bc Mistral\uff0c\u4f46\u4f4e\u65bc Llama3\u3002<\/li>\n\n\n\n<li><strong>\u8a31\u53ef\u8b49<\/strong>\uff1a\u6839\u64da\u958b\u653e\u8a31\u53ef\u8b49\u767c\u5e03\uff0c\u7279\u5225\u662f Apple \u7bc4\u4f8b\u7a0b\u5f0f\u78bc\u8a31\u53ef\u8b49\u3002<\/li>\n\n\n\n<li><strong>\u6bd4\u8f03<\/strong>\uff1a\u8207 Mistral \u7b49\u5c01\u9589\u8cc7\u6599\u96c6\u6a21\u578b\u7684\u6548\u80fd\u76f8\u7b26\u3002<\/li>\n\n\n\n<li><strong>\u8a13\u7df4\u6846\u67b6<\/strong>\uff1a\u4f7f\u7528 PyTorch \u548c OpenLM \u6846\u67b6\u958b\u767c\u3002<\/li>\n\n\n\n<li><strong>\u53ef\u7528\u6027<\/strong>\uff1a\u8a72\u6a21\u578b\u53ef\u5728 Hugging Face \u4e0a\u5b58\u53d6\u4e26\u6574\u5408\u5230 Transformers \u4e2d\u3002<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"additional-insights\">\u984d\u5916\u7684\u898b\u89e3\uff1a<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u8cc7\u6599\u7ba1\u7406<\/strong>\uff1a\u8a73\u7d30\u89e3\u91cb\u8cc7\u6599\u7ba1\u7406\u904e\u7a0b\uff0c\u63d0\u4f9b\u6709\u6548\u6cd5\u5b78\u78a9\u58eb\u57f9\u8a13\u7684\u898b\u89e3\u3002<\/li>\n\n\n\n<li><strong>\u8a13\u7df4\u6846\u67b6<\/strong>\uff1a\u5229\u7528DataComp-LM\u6846\u67b6\uff0c\u5c08\u6ce8\u65bc\u900f\u904e\u8cc7\u6599\u96c6\u5be6\u9a57\u6539\u9032\u8a9e\u8a00\u6a21\u578b\u3002<\/li>\n\n\n\n<li><strong>\u57fa\u6e96<\/strong>\uff1a\u4f7f\u7528\u4f86\u81ea Common Crawl \u8cc7\u6599\u96c6\u7684 2.5 \u5146\u500b\u4ee4\u724c\u9032\u884c\u8a13\u7df4\uff0c\u65e8\u5728\u63d0\u9ad8\u6548\u80fd\u3002<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"step-by-step-process-to-deploying-appledclm-7b-in-the-cloud\"><strong>\u5728\u96f2\u7aef\u90e8\u7f72 Apple\/DCLM-7B \u7684\u9010\u6b65\u904e\u7a0b<\/strong><\/h3>\n\n\n\n<p>\u5728\u672c\u6559\u5b78\u4e2d\uff0c\u6211\u5011\u5c07\u4f7f\u7528 NodeShift \u63d0\u4f9b\u7684 GPU \u9a45\u52d5\u7684\u865b\u64ec\u6a5f\u5668\uff1b\u4f46\u662f\uff0c\u60a8\u53ef\u4ee5\u8207\u60a8\u9078\u64c7\u7684\u4efb\u4f55\u5176\u4ed6\u96f2\u7aef\u63d0\u4f9b\u8005\u8907\u88fd\u76f8\u540c\u7684\u6b65\u9a5f\u3002<\/p>\n\n\n\n<p><strong>\u6b65\u9a5f 1\uff1a \u8a3b\u518a\u4e26\u8a2d\u5b9a NodeShift \u96f2\u7aef\u5e33\u6236<\/strong><\/p>\n\n\n\n<p>\u9020\u8a2a NodeShift Cloud \u7db2\u7ad9 (&nbsp;https:\/\/app.nodeshift.com\/&nbsp;) \u4e26\u5efa\u7acb\u5e33\u6236\u3002\u8a3b\u518a\u5f8c\uff0c\u767b\u5165\u60a8\u7684\u5e33\u6236\u3002<\/p>\n\n\n\n<p>\u9075\u5faa\u5e33\u6236\u8a2d\u5b9a\u6d41\u7a0b\u4e26\u63d0\u4f9b\u5fc5\u8981\u7684\u8a73\u7d30\u8cc7\u8a0a\u548c\u8cc7\u8a0a\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p><strong>\u6b65\u9a5f2\uff1a\u5efa\u7acbGPU\u865b\u64ec\u6a5f<\/strong><\/p>\n\n\n\n<p>NodeShift GPU \u63d0\u4f9b\u9748\u6d3b\u4e14\u53ef\u64f4\u5145\u7684\u96a8\u9078\u8cc7\u6e90\uff0c\u4f8b\u5982\u914d\u5099\u5f9e H100 \u5230 A100 \u7b49\u5404\u7a2e GPU \u7684 NodeShift \u865b\u64ec\u6a5f\u5668 (VM)\u3002\u9019\u4e9b\u7531 GPU \u9a45\u52d5\u7684\u865b\u64ec\u6a5f\u5668\u63d0\u4f9b\u589e\u5f37\u7684\u74b0\u5883\u63a7\u5236\uff0c\u53ef\u6839\u64da\u7279\u5b9a\u8981\u6c42\u8abf\u6574 GPU\u3001CPU\u3001RAM \u548c\u5132\u5b58\u7684\u914d\u7f6e\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-1.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p>\u5c0e\u89bd\u81f3\u5de6\u5074\u7684\u9078\u55ae\u3002\u9078\u64c7 GPU VM \u9078\u9805\uff0c\u5728\u5100\u8868\u677f\u4e2d\u5efa\u7acb GPU VM\uff0c\u6309\u4e00\u4e0b\u5efa\u7acb GPU VM \u6309\u9215\uff0c\u7136\u5f8c\u5efa\u7acb\u60a8\u7684\u7b2c\u4e00\u500b\u90e8\u7f72\u3002<\/p>\n\n\n\n<p><strong>\u6b65\u9a5f 3\uff1a\u9078\u64c7\u578b\u865f\u3001\u5340\u57df\u548c\u5b58\u5132<\/strong><\/p>\n\n\n\n<p>\u5728\u300cGPU VM\u300d\u6a19\u7c64\u4e2d\uff0c\u6839\u64da\u60a8\u7684\u9700\u6c42\u4ee5\u53ca\u8981\u555f\u52d5\u6a21\u578b\u7684\u5730\u7406\u5340\u57df\u9078\u64c7 GPU \u578b\u865f\u548c\u5132\u5b58\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-2.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p>\u5728\u672c\u6559\u7a0b\u4e2d\uff0c\u6211\u5011\u4f7f\u7528 1x NVIDIA RTX A6000 \u4f86\u90e8\u7f72 Apple\/DCLM-7B\u3002\u4e4b\u5f8c\uff0c\u9078\u64c7\u5132\u5b58\u91cf\uff08Apple \u7684 DCLM-7B \u81f3\u5c11\u9700\u8981 70 GB \u7684\u5132\u5b58\u7a7a\u9593\u3002<\/p>\n\n\n\n<p><strong>\u6b65\u9a5f4\uff1a\u9078\u64c7\u8eab\u4efd\u9a57\u8b49\u65b9\u6cd5<\/strong><\/p>\n\n\n\n<p>\u6709\u5169\u7a2e\u8eab\u4efd\u9a57\u8b49\u65b9\u6cd5\u53ef\u7528\uff1a\u5bc6\u78bc\u548c SSH \u91d1\u9470\u3002 SSH \u91d1\u9470\u662f\u4e00\u500b\u66f4\u5b89\u5168\u7684\u9078\u9805\uff0c\u70ba\u4e86\u5275\u5efa\u5b83\u5011\uff0c\u8acb\u8a2a\u554f\u6211\u5011\u7684\u5b98\u65b9\u6587\u4ef6\uff1a(&nbsp;https:\/\/docs.nodeshift.com\/gpus\/create-gpu-deployment&nbsp;)<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-3.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p><strong>\u7b2c 5 \u6b65\uff1a\u9078\u64c7\u5f71\u50cf<\/strong><\/p>\n\n\n\n<p>\u63a5\u4e0b\u4f86\uff0c\u60a8\u9700\u8981\u70ba\u60a8\u7684\u865b\u64ec\u6a5f\u5668\u9078\u64c7\u4e00\u500b\u6620\u50cf\u3002\u6211\u5011\u5c07\u5728 NVIDIA Cuda \u865b\u64ec\u6a5f\u5668\u4e0a\u90e8\u7f72 Apple\/DCLM-7B\u3002\u9019\u500b\u5c08\u6709\u7684\u9589\u6e90\u5e73\u884c\u904b\u7b97\u5e73\u53f0\u5c07\u5141\u8a31\u60a8\u5728 GPU VM \u4e0a\u5b89\u88dd Apple\/DCLM-7B\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-4.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p>\u9078\u64c7\u6620\u50cf\u5f8c\uff0c\u6309\u4e00\u4e0b\u300c\u5efa\u7acb\u300d\u6309\u9215\uff0c\u60a8\u7684\u865b\u64ec\u6a5f\u5668\u5c07\u88ab\u90e8\u7f72\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-5.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p><strong>\u6b65\u9a5f6\uff1a\u865b\u64ec\u6a5f\u5668\u90e8\u7f72\u6210\u529f<\/strong><\/p>\n\n\n\n<p>\u60a8\u5c07\u5f97\u5230\u6a5f\u5668\u5df2\u555f\u52d5\u4e26\u6b63\u5728\u904b\u884c\u7684\u8996\u89ba\u78ba\u8a8d\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-6.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p><strong>\u7b2c 7 \u6b65\uff1a\u4f7f\u7528 SSH \u9023\u7dda\u5230 GPU<\/strong><\/p>\n\n\n\n<p>NodeShift GPU \u53ef\u4ee5\u4f7f\u7528 GPU \u5efa\u7acb\u671f\u9593\u63d0\u4f9b\u7684 SSH \u91d1\u9470\u9023\u63a5\u5230\u7d42\u7aef\u4e26\u900f\u904e\u7d42\u7aef\u9032\u884c\u63a7\u5236\u3002<\/p>\n\n\n\n<p>\u6210\u529f\u5efa\u7acb GPU VM \u90e8\u7f72\u4e26\u9054\u5230\u300c\u6b63\u5728\u57f7\u884c\u300d\u72c0\u614b\u5f8c\uff0c\u60a8\u53ef\u4ee5\u5c0e\u89bd\u81f3 GPU \u90e8\u7f72\u57f7\u884c\u500b\u9ad4\u7684\u9801\u9762\u3002\u7136\u5f8c\uff0c\u9ede\u64ca\u53f3\u4e0a\u89d2\u7684\u201c\u9023\u63a5\u201d\u6309\u9215\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-7.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p>\u73fe\u5728\u6253\u958b\u7d42\u7aef\u6a5f\u4e26\u8cbc\u4e0a\u4ee3\u7406 SSH IP\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-8.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p>\u63a5\u4e0b\u4f86\uff0c\u5982\u679c\u60a8\u60f3\u6aa2\u67e5 GPU \u8a73\u7d30\u4fe1\u606f\uff0c\u8acb\u57f7\u884c\u4ee5\u4e0b\u547d\u4ee4\u201cnvidia-smi\u201d\uff1a<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-9.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p><strong>\u7b2c 8 \u6b65\uff1a\u5b89\u88dd Python \u548c Python \u5957\u4ef6<\/strong><\/p>\n\n\n\n<p>\u5b8c\u6210\u4e0a\u8ff0\u6b65\u9a5f\u5f8c\uff0c\u5c31\u53ef\u4ee5\u5275\u5efaPython\u7684\u865b\u64ec\u74b0\u5883\u4e86\u3002\u4e0b\u8f09 Python \u548c<strong>Python \u5957\u4ef6<\/strong>\u3002<\/p>\n\n\n\n<p>\u57f7\u884c\u4ee5\u4e0b\u547d\u4ee4\u4f86\u5b89\u88dd Python \u548c Python \u5957\u4ef6\u3002<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>sudo apt install python3.10 \npip install pandas\npip install transformers\npip install accelerate<\/code><\/pre>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-10.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-11.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p>\u6ce8\u610f\uff1a\u4e0b\u8f09\u6700\u65b0\u7248\u672c\u7684 Python\uff0c\u56e0\u70ba Pandas\u3001Pytorch \u7b49 Python \u5957\u4ef6\u9700\u8981\u6700\u65b0\u7248\u672c\u7684 Python \u624d\u80fd\u5728<strong>Apple\/DCLM-7B \u578b\u865f<\/strong>\u4e0a\u904b\u884c\uff1b\u5982\u679c\u4f60\u4e0b\u8f09\u820a\u7248\u7684Python\uff0c\u5b83\u6703\u5831\u932f\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-12.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p>\u6aa2\u67e5\u4e0b\u9762\u7684\u87a2\u5e55\u622a\u5716\u662f\u5426\u6709\u932f\u8aa4\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-13.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p>\u6ce8\u610f\uff1a\u5b8c\u6210\u6240\u6709\u9019\u4e9b\u6b65\u9a5f\u5f8c\uff0c\u8acb\u6aa2\u67e5\u5305\u62ec Pandas \u5728\u5167\u7684 Python \u5957\u4ef6\u7684\u7248\u672c\uff0c\u770b\u770b\u662f\u5426\u6709\u4efb\u4f55\u932f\u8aa4\u3002<\/p>\n\n\n\n<p><strong>\u6b65\u9a5f 9\uff1a\u5b89\u88dd Apple\/DCLM-7B \u578b\u865f<\/strong><\/p>\n\n\n\n<p>\u73fe\u5728\uff0c\u662f\u6642\u5019\u5f9e Hugging Face \u7db2\u7ad9\u4e0b\u8f09\u6a21\u578b\u4e86\u3002\u9023\u7d50\uff1ahttps :\/\/huggingface.co\/apple\/DCLM-7B<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-14.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p>\u4e4b\u5f8c\uff0c\u6211\u5011\u5c07\u5728cmd\u4e2d\u57f7\u884c\u4ee5\u4e0b\u547d\u4ee4\uff0c\u5b89\u88dd\u5c07\u958b\u59cb\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>pip install git+https:\/\/github.com\/mlfoundations\/open_lm.git<\/code><\/pre>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-15.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p>\u73fe\u5728\uff0c\u6211\u5011\u770b\u5230\u6211\u5011\u7684\u5b89\u88dd\u904e\u7a0b\u5df2\u7d93\u5b8c\u6210\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-16.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p><strong>\u7b2c10\u6b65\uff1a\u904b\u884cApple\/DCLM-7B\u6a21\u578b<\/strong><\/p>\n\n\n\n<p>\u6211\u5011\u6709\u5169\u7a2e\u904b\u884c DCLM 7B \u6a21\u578b\u7684\u9078\u9805\uff1aJupyter Lab \u548c\u7d42\u7aef\u6a5f\u3002<\/p>\n\n\n\n<p>\u5c0d\u65bc Jupyter Lab\uff0c\u6211\u5011\u5fc5\u9808\u5b89\u88dd\u4e00\u500b\u7b46\u8a18\u672c\uff0c\u5c0d\u65bc\u7d42\u7aef\uff0c\u6211\u5011\u5c07\u57f7\u884c Hugging Face \u7db2\u7ad9\u4e0a\u63d0\u4f9b\u7684\u8173\u672c\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-17.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p>\u6211\u5011\u5c07\u900f\u904e Jupyter Lab \u4f86\u5b8c\u6210\u6b64\u4efb\u52d9\u3002\u57f7\u884c\u4ee5\u4e0b\u547d\u4ee4\u5728\u865b\u64ec\u6a5f\u5668\u4e0a\u5b89\u88dd Jupyter Lab\u3002<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>pip install jupyterlab charset_normalizer<\/code><\/pre>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-18.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p>\u7576\u60a8\u57f7\u884c\u6b64\u547d\u4ee4\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>jupyter-lab <\/code><\/pre>\n\n\n\n<p>\u5b83\u5c07\u5728\u60a8\u7684\u700f\u89bd\u5668\u4e2d\u555f\u52d5\u7b46\u8a18\u672c\uff0c\u73fe\u5728\u60a8\u53ef\u4ee5\u8207\u60a8\u7684\u6a21\u578b\u4e92\u52d5\uff1a<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-21.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-20.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/blog.nodeshift.com\/content\/images\/2024\/08\/image-22.png?w=640&#038;ssl=1\" alt=\"\"\/><\/figure>\n<\/div>\n\n\n<p><strong>\u7d50\u8ad6<\/strong><br><strong>Apple DCLM-Baseline-7B<\/strong> \u6a21\u578b\u662f\u4e00\u500b\u64c1\u6709 70 \u5104\u53c3\u6578\u7684\u8a9e\u8a00\u6a21\u578b\uff0c\u5c55\u73fe\u4e86\u7cfb\u7d71\u6027\u8cc7\u6599\u6574\u7406\u5c0d\u8a9e\u8a00\u6a21\u578b\u6548\u80fd\u7684\u5f71\u97ff\u3002\u8a72\u6a21\u578b\u5728 2.5 \u5146\u8a5e\u5143\u4e0a\u9032\u884c\u8a13\u7df4\uff0c\u4e26\u904b\u7528\u4e86\u5148\u9032\u7684\u8cc7\u6599\u6574\u7406\u6280\u8853\uff0c\u5728 <strong>MMLU<\/strong> \u57fa\u6e96\u6e2c\u8a66\u4e2d\u53d6\u5f97\u4e86\u5177\u6709\u7af6\u722d\u529b\u7684\u7d50\u679c\u3002\u8a72\u6a21\u578b\u958b\u653e\u6388\u6b0a\uff0c\u4e26\u53ef\u5728 <strong>Hugging Face<\/strong> \u4e0a\u5b58\u53d6\uff0c\u4f7f\u7528 <strong>PyTorch<\/strong> \u548c <strong>OpenLM<\/strong> \u6846\u67b6\u958b\u767c\u3002\u5c07 <strong>Apple\/DCLM-7B<\/strong> \u90e8\u7f72\u5728\u96f2\u7aef\uff0c\u5c24\u5176\u662f\u4f7f\u7528 <strong>NodeShift<\/strong> \u7684 GPU \u865b\u64ec\u6a5f\u5668\uff0c\u6d89\u53ca\u5f9e\u5e33\u6236\u8a2d\u7f6e\u5230\u5728 <strong>Jupyter Lab<\/strong> \u4e2d\u904b\u884c\u6a21\u578b\u7684\u7c21\u55ae\u6b65\u9a5f\uff0c\u78ba\u4fdd\u4f7f\u7528\u8005\u80fd\u5920\u6709\u6548\u5730\u5229\u7528\u5176\u529f\u80fd\u3002<\/p>\n\n\n\n<p>\u8cc7\u6599\u4f86\u6e90: <a href=\"https:\/\/blog.nodeshift.com\/how-to-deploy-apple-dclm-7b-in-the-cloud-a-comprehensive-guide\/\">https:\/\/blog.nodeshift.com\/how-to-deploy-apple-dclm-7b-in-the-cloud-a-comprehensive-guide\/<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>2024-08-05 | Ayush Kumar Apple \u767c\u5e03\u4e86 DCLM\uff0c\u9019\u662f\u4e00\u500b 70 \u5104\u53c3\u6578\u7684\u958b\u6e90\u8a9e\u8a00\u6a21\u578b\uff0c\u5728\u958b\u6e90 AI \u9818\u57df\u9081\u51fa\u4e86\u91cd\u5927\u4e00\u6b65\u3002 DCLM-Baseline&hellip;<\/p>\n","protected":false},"author":4,"featured_media":7101,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_post_was_ever_published":false},"categories":[579,4],"tags":[40],"class_list":["post-7100","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-579","category-industry-news","tag-40"],"gutentor_comment":0,"jetpack_featured_media_url":"https:\/\/i0.wp.com\/aict.nkust.edu.tw\/digitrans\/wp-content\/uploads\/2024\/10\/%E8%9E%A2%E5%B9%95%E6%93%B7%E5%8F%96%E7%95%AB%E9%9D%A2-2024-10-04-170852.png?fit=1107%2C685&ssl=1","jetpack-related-posts":[],"jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts\/7100","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=7100"}],"version-history":[{"count":2,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts\/7100\/revisions"}],"predecessor-version":[{"id":7125,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts\/7100\/revisions\/7125"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/media\/7101"}],"wp:attachment":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7100"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7100"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7100"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}