{"id":5616,"date":"2024-01-13T21:17:20","date_gmt":"2024-01-13T13:17:20","guid":{"rendered":"https:\/\/aict.nkust.edu.tw\/digitrans\/?p=5616"},"modified":"2024-12-13T20:36:34","modified_gmt":"2024-12-13T12:36:34","slug":"%e5%bf%ab%e9%80%9f%e5%b8%b6%e4%bd%a0%e7%9c%8b-breeze-7b-%e9%80%8f%e9%81%8e-mistral-7b-fine-tune-%e5%87%ba%e4%be%86%e7%9a%84%e7%b9%81%e4%b8%ad%e9%96%8b%e6%ba%90%e6%a8%a1%e5%9e%8b","status":"publish","type":"post","link":"https:\/\/aict.nkust.edu.tw\/digitrans\/?p=5616","title":{"rendered":"[\u5feb\u901f\u5e36\u4f60\u770b] Breeze-7B: \u900f\u904e Mistral-7B Fine-Tune \u51fa\u4f86\u7684\u7e41\u4e2d\u958b\u6e90\u6a21\u578b"},"content":{"rendered":"\n<p>2024-01-13 | SimonLiu<\/p>\n\n\n\n<h1 class=\"wp-block-heading\" id=\"8e66\">I. Breeze \u6a21\u578b\u4ecb\u7d39<\/h1>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"2533\">\u6a21\u578b\u540d\u7a31<\/h2>\n\n\n\n<p id=\"cb7b\">Breeze-7B \u7cfb\u5217\u6a21\u578b<br>\u6b64\u6a21\u578b\u662f\u7531\u806f\u767c\u79d1\u7814\u767c\u5c0f\u7d44\u6240 Fine-Tuning\uff0c\u4e26\u4e14\u958b\u6e90\u51fa\u4f86\u7684 LLM \u6a21\u578b\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"b76c\">\u6a21\u578b\u958b\u6e90\u72c0\u6cc1 \/ License<\/h2>\n\n\n\n<p id=\"2fcc\">\u5f9e HuggingFace \u8cc7\u8a0a\u53ef\u4ee5\u770b\u5230\uff0cApache 2.0 License\uff0c\u662f\u4e00\u500b\u958b\u6e90\u6a21\u578b\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"2b11\">\u53c3\u6578\u91cf<\/h2>\n\n\n\n<p id=\"87dc\">7B<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"be0a\">\u6a21\u578b\u7a2e\u985e<\/h2>\n\n\n\n<p id=\"c704\">Breeze-7B \u9019\u6b21\u958b\u6e90\u51fa\u4f86\u7e3d\u5171\u6709\u4e09\u7a2e\u6a21\u578b\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>MediaTek-Research\/Breeze-7B-Base-v0.1: \u6b64\u6a21\u578b\u662f Breeze-7B \u7684 Base Model\u3002\u5982\u679c\u6709\u9700\u8981\u91dd\u5c0d\u7279\u5b9a\u7684\u9818\u57df\u505a\u8abf\u6574\uff0c\u90a3\u53ef\u4ee5\u5617\u8a66\u4f7f\u7528\u6b64\u6a21\u578b\u9032\u884c\u8655\u7406\u3002<\/li>\n\n\n\n<li>MediaTek-Research\/Breeze-7B-Instruct-v0.1: \u6b64\u6a21\u578b\u662f\u5f9e Breeze-7B-Base-v0.1 \u53bb\u9032\u884c\u8abf\u6574\uff0c\u8b93\u6a21\u578b\u9069\u5408\u7528\u65bc\u5e38\u898b\u4efb\u52d9\uff0c\u4f8b\u5982\u8aaa\u554f\u7b54\u7cfb\u7d71\u5c31\u53ef\u4ee5\u4f7f\u7528\u6b64\u6a21\u578b\u9032\u884c\u4ecb\u63a5\u548c\u4f7f\u7528\u3002<\/li>\n\n\n\n<li>MediaTek-Research\/Breeze-7B-Instruct-64k-v0.1: \u6b64\u6a21\u578b\u662f Breeze-7B-Instruct-v0.1 \u7684\u5fae\u8abf\u7248\u672c\uff0c\u53ef\u5728\u5927\u7d04 9 \u842c\u500b\u4e2d\u6587\u5b57\u7684 64k token \u4e0a\u9032\u884c\u4f7f\u7528\u3002<\/li>\n<\/ol>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"23a6\">\u6a21\u578b\u8868\u73fe<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6839\u64da\u5b98\u65b9\u7684\u8cc7\u6599\u986f\u793a\uff0c\u4ee5 TMMLU+ \u7684\u6e2c\u8a66\u7d50\u679c\u9032\u884c\u6bd4\u8f03\uff0c\u6b64\u6a21\u578b\u80fd\u5920\u5728\u7e41\u9ad4\u4e2d\u6587\u7684\u8a55\u5206\u4e0a\uff0c\u6bd4\u539f\u59cb\u7684 Mistral-7B-v0.1, Taiwan-LLM-7B\/13B-chat, QWen-7B-Chat, and Yi-6B-Chat \u6709\u66f4\u597d\u7684\u8868\u73fe\u3002<\/li>\n\n\n\n<li>\u82e5\u8ddf OpenAI GPT \u6a21\u578b\u76f8\u6bd4\uff0c\u6b64\u6a21\u578b\u76ee\u524d\u5728 TMMLU+ \u8207 GPT-3.5 \u6a21\u578b\u540c\u7b49\u5206\u6578\uff0c\u8b93\u6574\u9ad4\u7684\u9810\u6e2c\u80fd\u529b\u80fd\u5920\u66f4\u63a5\u8fd1 OpenAI \u6240\u63d0\u4f9b\u7684\u6a21\u578b\u3002<\/li>\n<\/ul>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/miro.medium.com\/v2\/resize%3Afit%3A700\/1%2AI3-keAdb2MLLCJkV8riWrg.png?w=640&#038;ssl=1\" alt=\"\"\/><figcaption class=\"wp-element-caption\">\u6a21\u578b\u5728 MMLU, TMMLU+ \u8cc7\u6599\u96c6\u7684\u8868\u73fe\u72c0\u6cc1 (\u900f\u904e ChatGPT \u6574\u7406\u7684\u8868\u683c)<\/figcaption><\/figure>\n<\/div>\n\n\n<ul class=\"wp-block-list\">\n<li>\u81f3\u65bc\u5728\u63a8\u8ad6\u6a21\u578b\u7684\u901f\u5ea6\u4f86\u8aaa\uff0c\u63a8\u8ad6\u6642\u9593\u8d8a\u77ed\u8868\u793a\u6a21\u578b\u8655\u7406\u8f38\u5165\u7684\u901f\u5ea6\u8d8a\u5feb\uff0c\u800c\u6700\u5927\u8f38\u5165\u9577\u5ea6\u5247\u986f\u793a\u6a21\u578b\u80fd\u5920\u8655\u7406\u7684\u5b57\u7b26\u6578\u91cf\u4e0a\u9650\u3002\u9019\u4e9b\u6307\u6a19\u5c0d\u65bc\u8a55\u4f30\u6a21\u578b\u7684\u5be6\u7528\u6027\u548c\u6548\u80fd\u975e\u5e38\u91cd\u8981\u3002\u5728\u4f5c\u8005\u5be6\u6e2c\uff0c\u65bc\u540c\u4e00\u500b\u74b0\u5883\u4e0b\uff0cBreeze-7B \u901f\u5ea6\u53c8\u53ef\u4ee5\u6bd4\u5176\u4ed6\u6a21\u578b\u9084\u8981\u4f86\u5f97\u66f4\u5feb\uff0c\u6700\u5927\u8f38\u5165\u9577\u5ea6\u53c8\u53ef\u4ee5\u6bd4\u8f03\u9577\uff0c\u8b93\u6574\u9ad4\u4f7f\u7528\u4e0a\u7684\u6548\u679c\u66f4\u597d\u3002<\/li>\n<\/ul>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" src=\"https:\/\/i0.wp.com\/miro.medium.com\/v2\/resize%3Afit%3A700\/1%2ARyAeTxqpcb7w-GswdsQq5Q.png?w=640&#038;ssl=1\" alt=\"\"\/><figcaption class=\"wp-element-caption\">\u63a8\u8ad6\u6642\u9593\u8207\u8f38\u5165\u9577\u5ea6\u8868\u683c (\u900f\u904e ChatGPT \u6574\u7406\u7684\u8868\u683c)<\/figcaption><\/figure>\n<\/div>\n\n\n<p id=\"31a9\">\u8a73\u7d30\u6bd4\u8f03\u7d50\u679c\u53ef\u898b HuggingFace \u4e0a\u7684\u8cc7\u8a0a\uff1a<a href=\"https:\/\/huggingface.co\/MediaTek-Research\/Breeze-7B-Instruct-64k-v0.1?source=post_page-----3e2f8e2f6da9--------------------------------#base-model-performance\" target=\"_blank\" rel=\"noreferrer noopener\"><\/a><\/p>\n\n\n\n<h2 class=\"wp-block-heading\">MediaTek-Research\/Breeze-7B-Instruct-64k-v0.1 \u00b7 Hugging Face<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">We&#8217;re on a journey to advance and democratize artificial intelligence through open source and open science.<\/h3>\n\n\n\n<p>huggingface.co<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"0306\">\u76f8\u95dc\u6a21\u578b\u7db2\u8def\u8cc7\u6e90<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>HuggingFace &#8211; Breeze-7B\u5c08\u6848\u8cc7\u8a0a\uff1a<a href=\"https:\/\/huggingface.co\/MediaTek-Research?search_models=Breeze-7B&amp;source=post_page-----3e2f8e2f6da9--------------------------------\" target=\"_blank\" rel=\"noreferrer noopener\"><\/a><\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">MediaTek-Research (MediaTek Research)<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">Org profile for MediaTek Research on Hugging Face, the AI community building the future.<\/h3>\n\n\n\n<p>huggingface.co<a href=\"https:\/\/huggingface.co\/audreyt?search_models=Breeze-7B&amp;source=post_page-----3e2f8e2f6da9--------------------------------\" target=\"_blank\" rel=\"noreferrer noopener\"><\/a><\/p>\n\n\n\n<h1 class=\"wp-block-heading\" id=\"248e\">II. \u6211\u7684\u89c0\u9ede<\/h1>\n\n\n\n<p id=\"822c\">Mistral-7B \u5728\u5f8c\u7e8c\u7684\u958b\u6e90\u6a21\u578b\u4e2d\uff0c\u53ef\u4ee5\u770b\u5230\u80fd\u5920\u8d8a\u4f86\u8d8a\u8ddf OpenAI GPT \u6a21\u578b\u5339\u6575\uff0c\u751a\u81f3\u8d85\u8d8a GPT 3.5 \u6a21\u578b\uff0c\u50cf\u662f Zephyr-7B\u3001Notus-7B \u6a21\u578b\u90fd\u662f\u6709\u9019\u6a23\u7684\u7279\u6027\uff1a<a href=\"https:\/\/www.linkedin.com\/posts\/simonliuyuwei_amazing-how-fast-these-small-models-are-catching-activity-7139791861954613248-x2Vm?utm_source=share&amp;utm_medium=member_desktop&amp;source=post_page-----3e2f8e2f6da9--------------------------------\" target=\"_blank\" rel=\"noreferrer noopener\"><\/a><\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Liu Yu-Wei on LinkedIn: \u7576\u5c0f\u6a21\u578b\u8d8a\u4f86\u8d8a\u5f37\uff0c\u6216\u8a31\u672a\u4f86\u7684\u5c0f\u578b\u88dd\u7f6e\uff0c\u90fd\u6703\u4f48\u6eff\u8a31\u591a\u7684 LLM \u6a21\u578b\uff01 Mistral 7B\u4ecb\u7d39\uff1ahttps:\/\/lnkd.in\/gM76_KWu<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">\u7576\u5c0f\u6a21\u578b\u8d8a\u4f86\u8d8a\u5f37\uff0c\u6216\u8a31\u672a\u4f86\u7684\u5c0f\u578b\u88dd\u7f6e\uff0c\u90fd\u6703\u4f48\u6eff\u8a31\u591a\u7684 LLM \u6a21\u578b\uff01 Mistral 7B\u4ecb\u7d39\uff1ahttps:\/\/lnkd.in\/gM76_KWu<\/h3>\n\n\n\n<p>\u7576\u5c0f\u6a21\u578b\u8d8a\u4f86\u8d8a\u5f37\uff0c\u6216\u8a31\u672a\u4f86\u7684\u5c0f\u578b\u88dd\u7f6e\uff0c\u90fd\u6703\u4f48\u6eff\u8a31\u591a\u7684 LLM \u6a21\u578b\uff01 Mistral 7B\u4ecb\u7d39\uff1ahttps:\/\/lnkd.in\/gM76_KWuwww.linkedin.com<\/p>\n\n\n\n<p id=\"8bda\">\u4e5f\u56e0\u6b64\uff0c\u9019\u4e9b Mistral-7B \u7684\u5f8c\u7e8c Fine-Tune \u6a21\u578b\u4e0a\uff0c\u82e5\u80fd\u5920\u6301\u7e8c\u964d\u4f4e\u8017\u8cbb\u7684\u904b\u7b97\u8cc7\u6e90\uff0c\u4e26\u4e14\u5728\u9810\u6e2c\u8868\u73fe\u6301\u7e8c\u66f4\u597d\u7684\u72c0\u6cc1\u4e0b\uff0c\u90e8\u7f72\u5728 Edge \u7aef\u7684\u786c\u9ad4\u88dd\u7f6e\u7684\u53ef\u80fd\u6027\u5c31\u6703\u8d8a\u4f86\u8d8a\u9ad8\uff0c\u6211\u60f3\u9019\u4e5f\u80fd\u5920\u5e6b\u52a9\u806f\u767c\u79d1\u5718\u968a\u5728\u7814\u7a76\u5982\u4f55\u5c07 LLM \u6a21\u578b\u653e\u5230\u534a\u5c0e\u9ad4\u6676\u7247\u4e2d\u6709\u5f88\u597d\u7684\u6548\u679c\u3002<\/p>\n\n\n\n<p id=\"ccc3\">\u81f3\u65bc\u82e5\u60f3\u5c07\u6b64\u985e\u6a21\u578b\u5c0e\u5165\u5728\u4e00\u822c\u5c08\u6848\u4e0a\uff0c\u56e0\u70ba Breeze-7B \u4ecd\u70ba\u5c0f\u6a21\u578b\uff0c\u5728\u5c08\u6848\u4efb\u52d9\u4e0a\uff0c\u5efa\u8b70\u9084\u662f\u8981\u5b9a\u7fa9\u7c21\u55ae\u660e\u77ad\u7684\u5c08\u6848\u76ee\u6a19\uff0c\u4e26\u4e14\u82e5\u5728\u5c08\u6848\u521d\u671f\u7121\u6cd5\u6709\u8db3\u5920\u786c\u9ad4\u8cc7\u6e90\u4e0b\uff0c\u8981\u6ce8\u610f\u4f7f\u7528\u7684\u8cc7\u6e90\u662f\u5426\u80fd\u5920\u627f\u53d7\u8ca0\u64d4\uff0c\u4f86\u5e6b\u52a9\u5c08\u6848\u80fd\u5920\u5be6\u8e10\u6210\u529f\u3002<\/p>\n\n\n\n<p>\u8cc7\u6599\u4f86\u6e90:<a href=\"https:\/\/blog.infuseai.io\/quick-demo-3-breeze-7b-mediatek-intro-3e2f8e2f6da9\" data-type=\"link\" data-id=\"https:\/\/blog.infuseai.io\/quick-demo-3-breeze-7b-mediatek-intro-3e2f8e2f6da9\">https:\/\/blog.infuseai.io\/quick-demo-3-breeze-7b-mediatek-intro-3e2f8e2f6da9<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>2024-01-13 | SimonLiu I. Breeze \u6a21\u578b\u4ecb\u7d39 \u6a21\u578b\u540d\u7a31 Breeze-7B \u7cfb\u5217\u6a21\u578b\u6b64\u6a21\u578b\u662f\u7531\u806f\u767c\u79d1\u7814\u767c\u5c0f\u7d44\u6240 Fine-Tuning\uff0c\u4e26\u4e14\u958b\u6e90\u51fa\u4f86\u7684 &hellip;<\/p>\n","protected":false},"author":9,"featured_media":5617,"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":[26,40],"class_list":["post-5616","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-579","category-industry-news","tag-ai","tag-40"],"gutentor_comment":0,"jetpack_featured_media_url":"https:\/\/i0.wp.com\/aict.nkust.edu.tw\/digitrans\/wp-content\/uploads\/2024\/03\/1_PCa57hc9AIIazGcyR4QmDA.webp?fit=720%2C406&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\/5616","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\/9"}],"replies":[{"embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=5616"}],"version-history":[{"count":1,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts\/5616\/revisions"}],"predecessor-version":[{"id":5618,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts\/5616\/revisions\/5618"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/media\/5617"}],"wp:attachment":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5616"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5616"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5616"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}