{"id":5270,"date":"2024-08-13T04:52:00","date_gmt":"2024-08-12T19:52:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=5270"},"modified":"2024-08-13T04:52:00","modified_gmt":"2024-08-12T19:52:00","slug":"self-taught-evaluators","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=5270","title":{"rendered":"Self-Taught Evaluators\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Self-Taught Evaluators&nbsp;<\/strong>[77.9]<br>\u672c\u7a3f\u3067\u306f,\u4eba\u5de5\u7684\u306a\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u30c7\u30fc\u30bf\u306e\u307f\u3092\u7528\u3044\u3066,\u4eba\u9593\u306e\u30a2\u30ce\u30c6\u30fc\u30b7\u30e7\u30f3\u3092\u4f7f\u308f\u305a\u306b\u5373\u8208\u3067\u8a3c\u660e\u3059\u308b\u3053\u3068\u3092\u76ee\u7684\u3068\u3057\u305f\u30a2\u30d7\u30ed\u30fc\u30c1\u3092\u63d0\u6848\u3059\u308b\u3002 \u6211\u3005\u306e\u81ea\u5df1\u5b66\u7fd2\u8a55\u4fa1\u5668\u306f\u3001RewardBench\u4e0a\u306775.4\u304b\u308988.3\u307e\u3067\u306e\u5f37\u3044LDM\u3092\u6539\u5584\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2408.02666v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2408.02666v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Mon, 05 Aug 2024 17:57:02 GMT)<\/li>\n\n\n\n<li>\u57fa\u672c\u7684\u306b\u7c21\u5358\u3067\u306f\u306a\u3044\u8a55\u4fa1\u3092\u884c\u3046\u30e2\u30c7\u30eb\u3092\u5408\u6210\u30c7\u30fc\u30bf\u7d4c\u7531\u3067\u4f5c\u308b\u7814\u7a76\u3002<\/li>\n\n\n\n<li>\u300cOur Self-Taught evaluator with iterative training over these synthetic preferences greatly boosts the accuracy of a strong seed LLM (Llama3-70B-Instruct) as an evaluator, from 75.4 to 88.7 on RewardBench, a new state-ofthe-art for generative LLM-as-a-Judge methods.\u300d\u3068\u306e\u3053\u3068\u3002\u901a\u5e38\u306e\u30e2\u30c7\u30eb\u69cb\u7bc9\u306b\u304a\u3051\u308b\u5408\u6210\u30c7\u30fc\u30bf\u306e\u6709\u52b9\u6027\u3092\u9451\u307f\u308b\u3068\u9a5a\u304d\u3067\u306f\u306a\u3044\u3082\u306e\u306e\u3001\uff08\u3053\u308c\u4ee5\u5916\u3067\u3082\u6700\u8fd1\u306eself-\u306a\u3093\u3068\u304b\u306a\u5831\u544a\u3092\u898b\u308b\u3068\uff09\u3053\u306e\u65b9\u5411\u6027\u3092\u7a81\u304d\u8a70\u3081\u308b\u3068AGI\u306b\u30fb\u30fb\u30fb\u3068\u3044\u3046\u6c17\u304c\u3057\u306a\u3044\u3067\u3082\u306a\u3044\u3002<\/li>\n<\/ul>\n\n\n\n<p>Self taught\u3060\u3068STaR\uff08Self-Taught Reasoner\uff09\u95a2\u9023\u3082\u9762\u767d\u3044\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Lean-STaR: Learning to Interleave Thinking and Proving\u00a0<\/strong>[53.9]<br>\u8a3c\u660e\u306e\u5404\u30b9\u30c6\u30c3\u30d7\u306b\u5148\u7acb\u3063\u3066,\u975e\u516c\u5f0f\u306a\u601d\u8003\u3092\u751f\u6210\u3059\u308b\u305f\u3081\u306b,\u8a00\u8a9e\u30e2\u30c7\u30eb\u3092\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3059\u308b\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3042\u308bLean-STaR\u3092\u7d39\u4ecb\u3057\u307e\u3059\u3002 Lean-STaR\u306f\u3001Lean\u5b9a\u7406\u8a3c\u660e\u74b0\u5883\u5185\u306eminiF2F-test\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3067\u6700\u5148\u7aef\u306e\u7d50\u679c\u3092\u9054\u6210\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2407.10040v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2407.10040v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Sun, 14 Jul 2024 01:43:07 GMT)<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking\u00a0<\/strong>[34.6]<br>\u81ea\u5df1\u5b66\u7fd2\u578b\u63a8\u8ad6\u5668\u306e\u4e00\u822c\u5316\u3067\u3042\u308bQuiet-STaR\u306b\u3064\u3044\u3066\u8ff0\u3079\u308b\u3002 LM\u306f\u3001\u5c06\u6765\u306e\u30c6\u30ad\u30b9\u30c8\u3092\u8aac\u660e\u3059\u308b\u305f\u3081\u306b\u5404\u30c8\u30fc\u30af\u30f3\u3067\u5408\u7406\u6027\u3092\u751f\u6210\u3059\u308b\u3053\u3068\u3092\u5b66\u3076\u3002 GSM8K\u3068CommonsenseQA\u3067\u306f\u30bc\u30ed\u30b7\u30e7\u30c3\u30c8\u306e\u6539\u5584\u304c\u898b\u3089\u308c\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2403.09629v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2403.09629v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 14 Mar 2024 17:58:16 GMT)<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>STaR: Bootstrapping Reasoning With Reasoning\u00a0<\/strong>[39.5]<br>\u81ea\u5df1\u5b66\u7fd2\u63a8\u8ad6(Slf-Taught Reason:STaR)\u306f\u5358\u7d14\u306a\u30eb\u30fc\u30d7\u306b\u4f9d\u5b58\u3057\u3001\u591a\u304f\u306e\u7591\u554f\u306b\u7b54\u3048\u308b\u305f\u3081\u306b\u7406\u6027\u3092\u751f\u6210\u3059\u308b\u3002 \u7d50\u679c,STaR\u306f\u6700\u7d42\u56de\u7b54\u3092\u76f4\u63a5\u4e88\u6e2c\u3059\u308b\u30e2\u30c7\u30eb\u3068\u6bd4\u8f03\u3057\u3066,\u8907\u6570\u306e\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u8457\u3057\u304f\u5411\u4e0a\u3055\u305b\u308b\u3053\u3068\u304c\u308f\u304b\u3063\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2203.14465v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2203.14465v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Mon, 28 Mar 2022 03:12:15 GMT)<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Self taught\u3060\u3068STaR\uff08Self-Taught Reasoner\uff09\u95a2\u9023\u3082\u9762\u767d\u3044\u3002<\/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":[223,224,356,390,647],"class_list":["post-5270","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-llm","tag-llm-as-a-judge","tag-self-x","tag-synthetic-data","tag-647"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5270","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=5270"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5270\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5270"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5270"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5270"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}