{"id":4786,"date":"2024-04-29T05:48:00","date_gmt":"2024-04-28T20:48:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=4786"},"modified":"2024-04-29T05:48:00","modified_gmt":"2024-04-28T20:48:00","slug":"snowflake-arctic-sensenova-5-0-qwen-1-5-110b","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=4786","title":{"rendered":"Phi-3, Snowflake Arctic, SenseNova 5.0, OpenELM, Qwen-1.5 110B"},"content":{"rendered":"\n<p>\u5148\u9031\u3082LLM\u95a2\u9023\u306e\u30cb\u30e5\u30fc\u30b9\u304c\u591a\u304b\u3063\u305f\u3002<\/p>\n\n\n\n<p>Phi-3\u306fMicrsoft\u306b\u3088\u308b\u5c0f\u898f\u6a21\uff08\uff1f\uff09LLM\u30013.8B\u30d1\u30e9\u30e1\u30fc\u30bf\u3068\u6bd4\u8f03\u7684\u5c0f\u3055\u3044\u304c\u6027\u80fd\u304c\u9ad8\u3044\u3068\u4e3b\u5f35\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone&nbsp;<\/strong>[144.9]<br>phi-3-mini\u306f\u30013.3\u5146\u306e\u30c8\u30fc\u30af\u30f3\u3067\u8a13\u7df4\u3055\u308c\u305f38\u5104\u306e\u30d1\u30e9\u30e1\u30fc\u30bf\u8a00\u8a9e\u30e2\u30c7\u30eb\u3067\u3042\u308b\u3002 MMLU\u3067\u306f69%\u3001MT\u30d9\u30f3\u30c1\u3067\u306f8.38\u3067\u3042\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2404.14219v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2404.14219v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Mon, 22 Apr 2024 14:32:33 GMT)<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/huggingface.co\/collections\/microsoft\/phi-3-6626e15e9585a200d2d761e3\">Phi-3 &#8211; a microsoft Collection (huggingface.co)<\/a><\/li>\n<\/ul>\n\n\n\n<p>Snowflake\u304c\u767a\u8868\u3057\u305fSnowflake Arctic\u306f\u7dcf\u30d1\u30e9\u30e1\u30fc\u30bf\u6570480B\u3060\u304c\u3001\u63a8\u8ad6\u6642\u306f17B\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u307f\u30a2\u30af\u30c6\u30a3\u30d6\u306b\u306a\u308bMoE\u69cb\u6210\u3002\u9762\u767d\u3044\u69cb\u6210\u3067\u6027\u80fd\u306fLlama3 70B\u76f8\u5f53\u3092\u4e3b\u5f35\u3001Apache-2\u30e9\u30a4\u30bb\u30f3\u30b9\u3068\u771f\u306b\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u306a\u30e9\u30a4\u30bb\u30f3\u30b9\u3067\u3042\u308b\u70b9\u3082\u7d20\u6674\u3089\u3057\u3044\u3002<br><a href=\"https:\/\/www.snowflake.com\/blog\/arctic-open-efficient-foundation-language-models-snowflake\/?lang=ja\">Snowflake Arctic &#8211; \u30a8\u30f3\u30bf\u30fc\u30d7\u30e9\u30a4\u30baAI\u5411\u3051LLM<\/a><\/p>\n\n\n\n<p>SenseNova\u306fSenseTime\u306b\u3088\u308bLLM\u3067GPT 4 turbo\u8d85\u3048\uff08\u305f\u3060\u3057\u6700\u65b0\u30e2\u30c7\u30eb\u306f\u306a\u3044\uff09\u3092\u4e3b\u5f35\u3002\u30af\u30ed\u30fc\u30ba\u30c9\u306a\u30e2\u30c7\u30eb\u3067\u306f\u3042\u308b\u304c\u6027\u80fd\u7af6\u4e89\u304c\u6fc0\u3057\u304f\u306a\u3063\u3066\u3044\u308b\u3053\u3068\u3092\u793a\u3057\u3066\u3044\u308b\u3002<br><a href=\"https:\/\/www.sensetime.com\/en\/news-detail\/51167731?categoryId=1072&amp;gioIndex=1\">SenseTime launches SenseNova 5.0 with comprehensive updates and the industry-leading &#8220;Cloud-to-Edge&#8221; full-stack large model product matrix-Newsroom-SenseTime<\/a><\/p>\n\n\n\n<p>Apple\u304cLLM\u3092\u516c\u958b\u3057\u305f\u3053\u3068\u3082\u8208\u5473\u6df1\u3044\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>OpenELM: An Efficient Language Model Family with Open-source Training and Inference Framework&nbsp;<\/strong>[26.7]<br>\u79c1\u305f\u3061\u306f\u6700\u5148\u7aef\u306e\u30aa\u30fc\u30d7\u30f3\u8a00\u8a9e\u30e2\u30c7\u30eb\u3067\u3042\u308bOpenELM\u3092\u30ea\u30ea\u30fc\u30b9\u3057\u307e\u3059\u3002 \u30d1\u30e9\u30e1\u30fc\u30bf\u4e88\u7b97\u306f\u7d0410\u5104\u306e\u30d1\u30e9\u30e1\u30fc\u30bf\u3067\u3001OpenELM\u306fOLMo\u306b\u6bd4\u3079\u3066\u7cbe\u5ea6\u304c2.36%\u5411\u4e0a\u3057\u3066\u3044\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2404.14619v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2404.14619v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Mon, 22 Apr 2024 23:12:03 GMT)<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/huggingface.co\/apple\/OpenELM\">apple\/OpenELM \u00b7 Hugging Face<\/a><\/li>\n<\/ul>\n\n\n\n<p>\u305d\u306e\u4ed6\u3001Qwen 1.5\u306e\u5927\u898f\u6a21\u30e2\u30c7\u30eb\u3067\u3042\u308bQwen-1.5\uff08<a href=\"https:\/\/huggingface.co\/Qwen\/Qwen1.5-110B\">Qwen\/Qwen1.5-110B \u00b7 Hugging Face<\/a>\uff09\u304c\u516c\u958b\u3001Nyonic Wonton7B\u304c\u767a\u8868\u306a\u3069LLM\u754c\u9688\u306f\u975e\u5e38\u306b\u6d3b\u6cc1\u3067\u3042\u308b\u3002<\/p>\n\n\n\n<p><a href=\"https:\/\/huggingface.co\/datasets\/HuggingFaceFW\/fineweb\">https:\/\/huggingface.co\/datasets\/HuggingFaceFW\/fineweb<\/a> \u306b\u95a2\u9023\u3057\u305fX\uff08\u65e7twitter\uff09\u3067\u306e\u6295\u7a3f\u3082\u8a71\u984c\u306b\u306a\u3063\u3066\u3044\u305f\u3002<a href=\"https:\/\/twitter.com\/Thom_Wolf\/status\/1782691683517305226\">X\u30e6\u30fc\u30b6\u30fc\u306eThomas Wolf\u3055\u3093: \u300cThis take on the FineWeb release is one of the most interesting feedback and also a reason FineWeb is very different from even larger datasets like RedPajama-V2 (which is double its size!) Surprisingly, the size of the dataset of 15T tokens is not very important, what is much\u2026\u300d \/ X (twitter.com)<\/a> \u300cBefore I dive more in this let me give you an example of unintuitive behavior. Between 2022 and 2023 the &#8220;LLM quality&#8221; of Common Crawl dropped significantly as in &#8220;training a LLM on the crawls btw 2022-2023 will give you lower performances on a set of evals&#8221;. What happened? Well it turns out the Common Crawl team has been filtering more strongly domains with adult content. Not really the cause you&#8217;d be intuitively thinking about, right?\u300d\u306f\u975e\u5e38\u306b\u8208\u5473\u6df1\u3044\u3002<br><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Nyonic Technical Report&nbsp;<\/strong>[20.8]<br>Wonton 7B\u30e2\u30c7\u30eb\u306f\u3001\u591a\u8a00\u8a9e\u304a\u3088\u3073\u82f1\u8a9e\u306e\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3067\u7af6\u5408\u6027\u80fd\u3092\u793a\u3057\u305f\u3002 \u30e2\u30c7\u30eb\u306e\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u306f\u3001\u30ed\u30fc\u30bf\u30ea\u30fc\u4f4d\u7f6e\u57cb\u3081\u8fbc\u307f(Rotary Positional Embeddings)\u3001QK-LayerNorm(QK-LayerNorm)\u3001\u7279\u5225\u306b\u88fd\u4f5c\u3055\u308c\u305f\u591a\u8a00\u8a9e\u30c8\u30fc\u30af\u30f3\u30fc\u30b6(multilingual tokenizer)\u306a\u3069\u306e\u6700\u5148\u7aef\u6280\u8853\u3067\u5f37\u5316\u3055\u308c\u3066\u3044\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2404.15702v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2404.15702v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Wed, 24 Apr 2024 07:38:44 GMT)<\/li>\n\n\n\n<li><a href=\"https:\/\/github.com\/nyonicai\/nyonic-public\">GitHub &#8211; nyonicai\/nyonic-public: Reference implementation of models from Nyonic Model Factory<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>\u5148\u9031\u3082LLM\u95a2\u9023\u306e\u30cb\u30e5\u30fc\u30b9\u304c\u591a\u304b\u3063\u305f\u3002 Phi-3\u306fMicrsoft\u306b\u3088\u308b\u5c0f\u898f\u6a21\uff08\uff1f\uff09LLM\u30013.8B\u30d1\u30e9\u30e1\u30fc\u30bf\u3068\u6bd4\u8f03\u7684\u5c0f\u3055\u3044\u304c\u6027\u80fd\u304c\u9ad8\u3044\u3068\u4e3b\u5f35\u3002 Snowflake\u304c\u767a\u8868\u3057\u305fSnowflake Arctic\u306f\u7dcf\u30d1\u30e9\u30e1\u30fc &hellip; <a href=\"https:\/\/devneko.jp\/wordpress\/?p=4786\" class=\"more-link\"><span class=\"screen-reader-text\">&#8220;Phi-3, Snowflake Arctic, SenseNova 5.0, OpenELM, Qwen-1.5 110B&#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],"tags":[223,293],"class_list":["post-4786","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-llm","tag-oss"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4786","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=4786"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4786\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4786"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4786"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4786"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}