{"id":6692,"date":"2025-05-06T05:06:00","date_gmt":"2025-05-05T20:06:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6692"},"modified":"2025-05-06T05:06:00","modified_gmt":"2025-05-05T20:06:00","slug":"deepcritic-deliberate-critique-with-large-language-models","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6692","title":{"rendered":"DeepCritic: Deliberate Critique with Large Language Models"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>DeepCritic: Deliberate Critique with Large Language Models\u00a0<\/strong>[77.6]<br>\u6211\u3005\u306f,Large Language Models(LLMs)\u306e\u6570\u5b66\u6279\u5224\u80fd\u529b\u306e\u7814\u7a76\u3068\u5411\u4e0a\u306b\u7126\u70b9\u3092\u3042\u3066\u308b\u3002 Qwen2.5-7B-Instruct\u3092\u30d9\u30fc\u30b9\u3068\u3057\u305f\u6279\u5224\u30e2\u30c7\u30eb\u3092\u958b\u767a\u3057\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2505.00662v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2505.00662v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 01 May 2025 17:03:17 GMT)<\/li>\n\n\n\n<li>Deep\u306a\u6279\u8a55\u3092\u884c\u3046\u30e2\u30c7\u30eb\u306e\u63d0\u6848\u3002\u300cIn Stage 1, we first utilize Qwen2.5-72B-Instruct to generate an initial step-wise critique for each step in the solution, followed by an in-depth critique of the initial critique.\u300d\u3001\u300cIn Stage 2, we perform RL to the SFT model on either existing human-annotated data or auto-labeled data via Monte Carlo sampling-based correctness estimation, to further stimulate the critique ability of the critic.\u300d\u306e2\u30b9\u30c6\u30fc\u30b8\u69cb\u6210\u3067\u69cb\u7bc9\u3002Critic\u30e2\u30c7\u30eb\u306f\u4ed6\u306e\u30e2\u30c7\u30eb\u51fa\u529b\u306e\u4fee\u6b63\u306b\u3082\u6709\u52b9\u306a\u3053\u3068\u304c\u77e5\u3089\u308c\u3066\u3044\u308b\u304c\u300cour 7B critique model is also capable of supervising and correcting the outputs of a 72B generator, demonstrating a potential of weak-to-strong supervision\u300d\u306f\u8208\u5473\u6df1\u3044\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/RUCBM\/DeepCritic\">GitHub &#8211; RUCBM\/DeepCritic: Official repository for paper &#8220;DeepCritic: Deliberate Critique with Large Language Models&#8221;<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[84,390],"class_list":["post-6692","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-critic","tag-synthetic-data"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6692","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=6692"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6692\/revisions"}],"predecessor-version":[{"id":6701,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6692\/revisions\/6701"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6692"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6692"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6692"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}