{"id":28245,"date":"2025-02-07T10:36:47","date_gmt":"2025-02-07T02:36:47","guid":{"rendered":"https:\/\/www.1ai.net\/?p=28245"},"modified":"2025-02-07T10:36:47","modified_gmt":"2025-02-07T02:36:47","slug":"%e9%98%bf%e9%87%8c%e4%ba%91%e7%a1%ae%e8%ae%a4%ef%bc%9a%e6%9d%8e%e9%a3%9e%e9%a3%9e%e5%9b%a2%e9%98%9f-s1-%e6%a8%a1%e5%9e%8b%e5%9f%ba%e4%ba%8e-qwen2-5-32b-instruct-%e6%a8%a1%e5%9e%8b%e8%ae%ad%e7%bb%83","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/28245.html","title":{"rendered":"AliCloud Confirms: Feifei Li's Team s1 Model Trained on Qwen2.5-32B-Instruct Model"},"content":{"rendered":"<p><a href=\"https:\/\/www.1ai.net\/en\/28158.html\/\">Li Feifei Research Team<\/a>by<strong>Less than $50.<\/strong>'s cloud computing costs trained an AI inference model called s1 that performed similarly to cutting-edge inference models like OpenAI's o1 and DeepSeek's R1 in tests of mathematical and coding ability.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/02\/b44fec88j00sr8s8u0030d000v900b8p.jpg\" alt=\"For Less Than $50 to Train, Researchers Build an Inference Model That Rivals OpenAI o1\" \/><\/p>\n<p>However, soon, the s1 model was said to be \"not trained from scratch\", and its base model is \"Ali Tongyi Qianqian (Qwen) model\". In response, Sina Tech asked<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e9%98%bf%e9%87%8c%e4%ba%91\" title=\"_Other Organiser\" target=\"_blank\" >Alibaba Cloud<\/a>I'm going to have to ask for confirmation.<strong>AliCloud confirmed the news<\/strong>.<\/p>\n<p>AliCloud responded, \"Using the Ali Tongyi Thousand Questions Qwen2.5-32B-Instruct open-source model as a base, they trained the new model s1-32B in 26 minutes of supervised fine-tuning on 16 H100 GPUs, achieving mathematical and coding capabilities comparable to cutting-edge inference models such as OpenAI's o1 and DeepSeek's R1 comparable results, even outperforming o1-preview by 27% on competition math problems.\"<\/p>\n<p>As previously reported by 1AI, the s1 team revealed that<strong>They created the AI model through \"distillation\" technology.<\/strong>The technology aims to extract the \"reasoning\" ability of an AI model by training it to learn the answers of another AI model.<\/p>\n<p>s1 of the paper shows that it is possible to use a method called supervised fine-tuning (SFT) that<strong>Inference models can be distilled using relatively small datasets<\/strong>In SFT. In SFT, AI models are explicitly instructed to mimic certain behaviors in the dataset.SFT is more cost-effective than the large-scale reinforcement learning approach that DeepSeek uses to train its R1 models.<\/p>\n<p>s1 is based on a small, off-the-shelf, free AI model provided by Qwen, Alibaba's Chinese AI lab. To train s1, the<strong>The researchers created a dataset of just 1,000 carefully curated questions<\/strong>The answers to these questions, as well as the \"thinking\" process behind each of the answers given in the Google Gemini 2.0 Flash Thinking Experimental.<\/p>","protected":false},"excerpt":{"rendered":"<p>Feifei Li's research team trained an AI inference model called s1 for less than $50 in the cloud, which performed similarly to cutting-edge inference models such as OpenAI's o1 and DeepSeek's R1 in tests of mathematical and coding ability. However, it was soon revealed that the s1 model was \"not trained from scratch\" and that its base model was the \"Ali Tongyi Qwen\" model. In this regard, Sina Technology asked AliCloud for confirmation, AliCloud confirmed the news. AliCloud responded: \"They used the Ali Tongyi Qwen2.5-32B-Instruct open-source model as a base, and supervised the fine-tuning on 16 H100 GPUs for 26 minutes to train a new model!<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[146],"tags":[3586,334],"collection":[],"class_list":["post-28245","post","type-post","status-publish","format-standard","hentry","category-news","tag-3586","tag-334"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/28245","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=28245"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/28245\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=28245"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=28245"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=28245"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=28245"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}