{"id":1005,"date":"2023-08-20T09:30:00","date_gmt":"2023-08-20T00:30:00","guid":{"rendered":"https:\/\/staka.jp\/wordpress\/?p=1005"},"modified":"2023-08-20T09:30:00","modified_gmt":"2023-08-20T00:30:00","slug":"weblab-10b-qwen-7b%e3%81%ae%e6%a9%9f%e6%a2%b0%e7%bf%bb%e8%a8%b3%e6%80%a7%e8%83%bd","status":"publish","type":"post","link":"https:\/\/staka.jp\/wordpress\/?p=1005","title":{"rendered":"Weblab-10B, Qwen-7B\u306e\u6a5f\u68b0\u7ffb\u8a33\u6027\u80fd"},"content":{"rendered":"\n<p>\u6700\u8fd1\u3001\u65e5\u82f1\u5bfe\u5fdc\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\u3084\u65e5\u672c\u8a9e\u304c\u6271\u3048\u308b\u591a\u8a00\u8a9e\u306a\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\u304c\u8907\u6570\u306e\u4f1a\u793e\u30fb\u7814\u7a76\u6a5f\u95a2\u304b\u3089\u516c\u958b\u3055\u308c\u305f\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/rinna.co.jp\/news\/2023\/07\/20230731.html\">rinna\u3001\u65e5\u82f1\u30d0\u30a4\u30ea\u30f3\u30ac\u30eb\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\u3092\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u3067\u516c\u958b\uff5crinna\u682a\u5f0f\u4f1a\u793e<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.reuters.com\/technology\/alibaba-unveils-open-sourced-ai-model-similar-metas-llama-2-2023-08-03\/\">Alibaba rolls out open-sourced AI model to take on Meta&#8217;s Llama 2 | Reuters<\/a>\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/github.com\/QwenLM\/Qwen-7B\/blob\/main\/README_JA.md\">Qwen-7B\/README_JA.md at main \u00b7 QwenLM\/Qwen-7B \u00b7 GitHub<\/a><\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><a href=\"https:\/\/ja.stability.ai\/blog\/japanese-stablelm-alpha\">\u65e5\u672c\u8a9e\u8a00\u8a9e\u30e2\u30c7\u30eb\u300cJapanese StableLM Alpha\u300d\u3092\u30ea\u30ea\u30fc\u30b9\u3057\u307e\u3057\u305f \u2014 Stability AI Japan<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.t.u-tokyo.ac.jp\/press\/pr2023-08-18-001\">\u6771\u4eac\u5927\u5b66\u677e\u5c3e\u7814\u7a76\u5ba4 100\u5104\u30d1\u30e9\u30e1\u30fc\u30bf\u30b5\u30a4\u30ba\u30fb\u65e5\u82f12\u30f6\u56fd\u8a9e\u5bfe\u5fdc\u306e\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\u201cWeblab-10B\u201d\u3092\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u3067\u516c\u958b \u2015\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u306e\u65e5\u672c\u8a9e\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\u3067\u6700\u9ad8\u6c34\u6e96\u2015\uff5c\u30d7\u30ec\u30b9\u30ea\u30ea\u30fc\u30b9 | UTokyo-Eng (u-tokyo.ac.jp)<\/a><\/li>\n<\/ul>\n\n\n\n<p>\u4eca\u307e\u3067\u3068\u540c\u69d8\u3001\u4e0a\u8a18\u306e\u6a5f\u68b0\u7ffb\u8a33\u6027\u80fd\u3092\u691c\u8a3c\u3057\u3066\u307f\u305f\u3002\u6027\u80fd\u8a55\u4fa1\u306b\u4f7f\u7528\u3057\u305f\u30c7\u30fc\u30bf\u306f\u4ee5\u524d\uff08<a href=\"https:\/\/staka.jp\/wordpress\/?p=959\">DAMO PolyLM-13B\u306e\u6a5f\u68b0\u7ffb\u8a33\u6027\u80fd | \u3077\u308b\u30fc\u3075\u304a\u3076\u3053\u3093\u305b\u3077\u3068 (staka.jp)<\/a>\uff09\u3068\u540c\u3058\u3002\u691c\u8a3c\u74b0\u5883\u306f\u74b0\u5883\u306fColab Pro+\u3092\u7528\u3044float16\u3067\u8aad\u307f\u8fbc\u3093\u3060\u3002Prompt\u3082\u4ee5\u524d\u3068\u8fd1\u3044\u3082\u306e\u3092\u4f7f\u3063\u3066\u3044\u308b\u304c\u65e5\u672c\u8a9e\u304c\u4e3b\u4f53\u306eLLM\u3068\u601d\u308f\u308c\u308brinna\u3001StableLM Alpha\u3001Weblab\u306b\u3064\u3044\u3066\u306f\u6307\u793a\u3092\u65e5\u672c\u8a9e\u306b\u3057\u305f\u30d7\u30ed\u30f3\u30d7\u30c8\u3082\u8a66\u3057\u3066\u3044\u308b[1]\u3002<\/p>\n\n\n\n<p>\u4e0a\u8a184\u7a2e\u985e\u306e\u30e2\u30c7\u30eb\u3092\u691c\u8a3c\u3057\u3066\u307f\u305f\u304cJapanese StableLM Alpha\u30b7\u30ea\u30fc\u30ba\u306e\u300cJapanese-StableLM-Base-Alpha-7B\u300d[2]\u300cJapanese-StableLM-Instruct-Alpha-7B\u300d[3]\u3001rinna\u793e\u306e\u65e5\u82f1\u30d0\u30a4\u30ea\u30f3\u30ac\u30eb\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\u300cbilingual-gpt-neox-4b-instruction-sft\u300d[4]\u306f\u5341\u5206\u306a\u54c1\u8cea\u304c\u9054\u6210\u3067\u304d\u306a\u304b\u3063\u305f\u305f\u3081\u8868\u306b\u306f\u5165\u308c\u3066\u3044\u306a\u3044\u3002<\/p>\n\n\n\n<p>\u672c\u4ef6\u306e\u691c\u8a3c\u3067\u306fLLM\u306e\u51fa\u529b\u304b\u3089\u624b\u52d5\u3067\u56de\u7b54\u3092\u62bd\u51fa\u3059\u308b\u3053\u3068\u306f\u884c\u3063\u3066\u3044\u306a\u3044\u3002\u591a\u304f\u306e\u5834\u5408\u30eb\u30fc\u30eb\u30d9\u30fc\u30b9\u3067\u5341\u5206\u3060\u3063\u305f\u4e8b\u3068\u624b\u52d5\u62bd\u51fa\u304c\u5fc5\u8981\u306a\u5834\u5408\u306f\u305d\u3082\u305d\u3082\u306e\u7ffb\u8a33\u54c1\u8cea\u304c\u60aa\u304f\u56de\u7b54\u62bd\u51fa\u306b\u610f\u5473\u304c\u306a\u304b\u3063\u305f\u3002<\/p>\n\n\n\n<figure class=\"wp-block-table is-style-stripes\"><table><tbody><tr><td><strong>\u30e2\u30c7\u30eb\u30fb\u6761\u4ef6<\/strong><\/td><td><strong>zero shot<\/strong><\/td><td><strong>1 shot<\/strong> [5]<\/td><\/tr><tr><td>GPT-3.5 [6]<\/td><td>26.7<\/td><td><strong>37.0<\/strong><\/td><\/tr><tr><td>Qwen-7B-Chat [7]<\/td><td>14.5<\/td><td>23.3<\/td><\/tr><tr><td>Weblab-10B <br>weblab-10b-instruction-sft [8]<\/td><td>11.5<\/td><td><strong>29.6<\/strong><\/td><\/tr><\/tbody><\/table><figcaption class=\"wp-element-caption\">Qwen-7B, Weblab-10B, GPT-3.5-TURBO-16K\u306eBLEU<\/figcaption><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">\u7d50\u679c<\/h2>\n\n\n\n<p>\u4ee5\u524d\u306e\u691c\u8a3c\u3067\u307b\u307c\u540c\u6761\u4ef6\uff061 shot\u306e\u6027\u80fd\u306fLlama-2-7b-chat-hf: 20.1\u3001Llama-2-13b-chat-hf: 33.1\u3001PolyLM-13B: 21.5\u3060\u3063\u305f[9]\u3002\u30d1\u30e9\u30e1\u30fc\u30bf\u6570\u3092\u8003\u3048\u308b\u3068Qwen-7B\u3001Weblab-10B\u3068\u3082\u306b\u76f8\u5fdc\u306e\u6a5f\u68b0\u7ffb\u8a33\u6027\u80fd\u3092\u6301\u3063\u3066\u3044\u305d\u3046\u3067\u3042\u308b\u3002<\/p>\n\n\n\n<p>\u76ee\u691c\u3057\u3066\u307f\u308b\u3068Qwen-7B\u306f\u65e5\u672c\u8a9e\u306e\u8a9e\u5f59\u304c\u4e0d\u8db3\u3057\u3066\u3044\u308b\u5370\u8c61\u3001Weblab-10B\u306fhallucination\u304c\u591a\u3081\u306e\u5370\u8c61\u3092\u53d7\u3051\u305f\u3002\u6a5f\u68b0\u7ffb\u8a33\u306f\u30d1\u30e9\u30e1\u30fc\u30bf\u30b5\u30a4\u30ba\u3067\u6027\u80fd\u304c\u5927\u304d\u304f\u5909\u308f\u308b\u306e\u306713B\u7a0b\u5ea6\u306e\u30b5\u30a4\u30ba\u306b\u5408\u308f\u305b\u305f\u691c\u8a3c\u3092\u884c\u3063\u3066\u307f\u305f\u3044\u306a\u3068\u601d\u308f\u306a\u304f\u306f\u306a\u3044\u3002<\/p>\n\n\n\n<p>\u30e9\u30a4\u30bb\u30f3\u30b9\u306fQwen-7B\u304cLlama 2\u306b\u8fd1\u3044\u72ec\u81ea\u30e9\u30a4\u30bb\u30f3\u30b9\uff08<a href=\"https:\/\/github.com\/QwenLM\/Qwen-7B\/blob\/main\/LICENSE\">Qwen-7B\/LICENSE at main \u00b7 QwenLM\/Qwen-7B \u00b7 GitHub<\/a>\uff09\u3001Weblab-10B\u306fCC BY-NC\u3068\u7814\u7a76\u7528\u306b\u306f\u554f\u984c\u306a\u304f\u4f7f\u3048\u308b\u3082\u306e\u3068\u306a\u3063\u3066\u3044\u308b[10]\u3002<\/p>\n\n\n\n<p>\u69d8\u3005\u306a\u7814\u7a76\u6a5f\u95a2\u304c\u30e2\u30c7\u30eb\u3092\u516c\u958b\u3057\u3066\u304f\u308c\u308b\u3068\u305d\u306e\u5dee\u7570\u304b\u3089\u5206\u304b\u308b\u3053\u3068\u3082\u3042\u308a\u975e\u5e38\u306b\u3042\u308a\u304c\u305f\u3044\u3002<a href=\"https:\/\/github.com\/QwenLM\/Qwen-7B#performance\">Qwen-7B\u306fMMLU\u306f\u3058\u3081\u69d8\u3005\u306a\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306e\u7d50\u679c<\/a>\u3001Weblab-10B\u306f<a href=\"https:\/\/github.com\/Stability-AI\/lm-evaluation-harness\/\">GitHub &#8211; Stability-AI\/lm-evaluation-harness: A framework for few-shot evaluation of autoregressive language models.<\/a>\u3092\u7528\u3044\u305fJGLUE\u3067\u306e\u8a55\u4fa1\u7d50\u679c\u3092\u8f09\u305b\u3066\u304f\u308c\u3066\u3044\u308b\u306e\u304c\u3046\u308c\u3057\u3044\u3002\u69d8\u3005\u306a\u30bf\u30b9\u30af\u306b\u304a\u3051\u308b\u6027\u80fd\u306f\u30d1\u30e9\u30e1\u30fc\u30bf\u30b5\u30a4\u30ba\u3084\u6700\u5927\u30b7\u30fc\u30b1\u30f3\u30b9\u9577\u3088\u308a\u91cd\u8981\u306a\u60c5\u5831\u3067\u53ef\u80fd\u306a\u9650\u308a\u516c\u958b\u3057\u3066\u3082\u3089\u3048\u308b\u3068\u3042\u308a\u304c\u305f\u3044\u306a\u3068\u601d\u3046[11]\u3002<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">\u811a\u6ce8<\/h4>\n\n\n\n<p>[1] \u6b63\u76f4\u3001\u5927\u5e45\u306a\u6027\u80fd\u5411\u4e0a\u3068\u3044\u3046\u52b9\u679c\u306f\u898b\u3089\u308c\u306a\u304b\u3063\u305f\u30021 shot\u3067\u306f\u307b\u307c\u52b9\u679c\u304c\u7121\u304f\u3001zero shot\u3067\u306f\u65e5\u672c\u8a9e\u3067\u6307\u793a\u3092\u4e0e\u3048\u305f\u65b9\u304c\u826f\u3044\u3088\u3046\u306b\u3082\u898b\u3048\u305f\u304c\u305d\u3082\u305d\u3082\u306eBLEU\u304c\u4f4e\u3044\u305f\u3081\u53c2\u8003\u306b\u306f\u306a\u3089\u306a\u3044\u5370\u8c61\u3002\u306a\u304a\u3001Weblab-10B\u306b\u3064\u3044\u3066\u306ffew shot\u90e8\u5206\u3092\u516c\u5f0f\u306e\u300c\\n\\n### \u6307\u793a:\\n{} \\n\\n### \u5fdc\u7b54:\\n{}\u300d\u306b\u5408\u308f\u305b\u308b\u306e\u306f\u52b9\u679c\u304c\u3042\u308a\u305d\u3046\u3060\u3063\u305f\u3002<br>[2] <a href=\"https:\/\/huggingface.co\/stabilityai\/stablelm-base-alpha-7b\">stabilityai\/stablelm-base-alpha-7b \u00b7 Hugging Face<\/a><br>[3] <a href=\"https:\/\/huggingface.co\/stabilityai\/japanese-stablelm-instruct-alpha-7b\">stabilityai\/japanese-stablelm-instruct-alpha-7b \u00b7 Hugging Face<\/a><br>[4] <a href=\"https:\/\/huggingface.co\/rinna\/bilingual-gpt-neox-4b-instruction-sft\">rinna\/bilingual-gpt-neox-4b-instruction-sft \u00b7 Hugging Face<\/a><br>[5] \u4ee5\u524d\u3068\u540c\u3058\u3067OpenICL\u3001TopkRetriever\u306b\u3088\u308a\u53d6\u5f97<br>[6] gpt-3.5-turbo-16k-0613<br>[7] <a href=\"https:\/\/huggingface.co\/Qwen\/Qwen-7B-Chat\">Qwen\/Qwen-7B-Chat \u00b7 Hugging Face<\/a><br>[8] <a href=\"https:\/\/huggingface.co\/matsuo-lab\/weblab-10b-instruction-sft\">matsuo-lab\/weblab-10b-instruction-sft \u00b7 Hugging Face<\/a><br>[9] \u56de\u7b54\u3092\u624b\u52d5\u62bd\u51fa\u3057\u305f\u5834\u5408\u306e\u6027\u80fd\u306fLlama-2-7b-chat-hf: 23.9\u3001Llama-2-13b-chat-hf: 35.1\u3001PolyLM-13B: 25.8<br>[10] Qwen\u306f\u5546\u7528\u5229\u7528\u53ef\u80fd\u305d\u3046\u3060\u304c\u300cIf you are commercially using the Materials, and your product or service has more than 100 million monthly active users\u300d\u3084\u300cYou can not use the Materials or any output therefrom to improve any other large language model (excluding Tongyi Qianwen or derivative works thereof).\u300d\u3068\u306e\u8a18\u8f09\u304c\u3042\u308b\u3002\u307e\u305f\u3001<a href=\"https:\/\/huggingface.co\/Qwen\/Qwen-7B-Chat#%E4%BD%BF%E7%94%A8%E5%8D%8F%E8%AE%AE%EF%BC%88license-agreement%EF%BC%89\">HuggingFace\u306e\u30da\u30fc\u30b8<\/a>\u306b\u306f\u300cIf you have requirements for commercial use, please fill out the form to apply.\u300d\u3068\u3044\u3046\u30d5\u30a9\u30fc\u30e0\u304c\u7528\u610f\u3055\u308c\u3066\u3044\u308b\u3002Weblab-10B\u306fCC BY-<strong>NC<\/strong>\u306a\u306e\u3067<strong>\u5546\u7528\u5229\u7528\u4e0d\u53ef<\/strong>\u3002<br>[11] \u30e2\u30c7\u30eb\u304c\u516c\u958b\u3055\u308c\u3066\u3044\u308c\u3070\u691c\u8a3c\u306f\u3067\u304d\u308b\u304c\u6b63\u3057\u304f\u691c\u8a3c\u3067\u304d\u3066\u3044\u308b\u304b\u306f\u8b0e\u306a\u90e8\u5206\u304c\u3042\u308b\u3002\u7279\u306b\u300c\u305d\u308c\u305e\u308c\u306e\u30e2\u30c7\u30eb\u3067\u6700\u9069\u306a\u30d7\u30ed\u30f3\u30d7\u30c8\u300d\u300cFew shot\u4e8b\u4f8b\u306e\u4e0e\u3048\u65b9\u300d\u306f\u4f7f\u3063\u305f\u30c7\u30fc\u30bf\u306b\u3082\u4f9d\u5b58\u3059\u308b\u308f\u3051\u3067\u958b\u767a\u5143\u306e\u516c\u5f0f\u306a\u60c5\u5831\u304c\u3042\u308b\u3068\u826f\u3044\u306a\u3068\u306f\u601d\u3046\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6700\u8fd1\u3001\u65e5\u82f1\u5bfe\u5fdc\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\u3084\u65e5\u672c\u8a9e\u304c\u6271\u3048\u308b\u591a\u8a00\u8a9e\u306a\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\u304c\u8907\u6570\u306e\u4f1a\u793e\u30fb\u7814\u7a76\u6a5f\u95a2\u304b\u3089\u516c\u958b\u3055\u308c\u305f\u3002 \u4eca\u307e\u3067\u3068\u540c\u69d8\u3001\u4e0a\u8a18\u306e\u6a5f\u68b0\u7ffb\u8a33\u6027\u80fd\u3092\u691c\u8a3c\u3057\u3066\u307f\u305f\u3002\u6027\u80fd\u8a55\u4fa1\u306b\u4f7f\u7528\u3057\u305f\u30c7\u30fc\u30bf\u306f\u4ee5\u524d\uff08DAMO PolyLM-13B &hellip; <a href=\"https:\/\/staka.jp\/wordpress\/?p=1005\" class=\"more-link\"><span class=\"screen-reader-text\">&#8220;Weblab-10B, Qwen-7B\u306e\u6a5f\u68b0\u7ffb\u8a33\u6027\u80fd&#8221; \u306e<\/span>\u7d9a\u304d\u3092\u8aad\u3080<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2,13],"tags":[37,64,73],"class_list":["post-1005","post","type-post","status-publish","format-standard","hentry","category-ai","category-13","tag-gpt-3-5-turbo","tag-qwen-7b","tag-weblab-10b"],"_links":{"self":[{"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/1005","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1005"}],"version-history":[{"count":0,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/1005\/revisions"}],"wp:attachment":[{"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1005"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1005"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1005"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}