{"id":30191,"date":"2025-04-09T10:27:54","date_gmt":"2025-04-09T02:27:54","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=30191"},"modified":"2025-04-09T10:27:54","modified_gmt":"2025-04-09T02:27:54","slug":"ai-scientist-v2","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/de\/ai-scientist-v2\/","title":{"rendered":"AI-Scientist-v2\uff1a\u81ea\u4e3b\u5b8c\u6210\u79d1\u7814\u7814\u7a76\u4e0e\u8bba\u6587\u64b0\u5199"},"content":{"rendered":"<p>AI-Scientist-v2 \u662f\u7531\u65e5\u672c SakanaAI \u516c\u53f8\u5f00\u53d1\u7684\u667a\u80fd\u7cfb\u7edf\uff0c\u65e8\u5728\u8ba9\u673a\u5668\u4ece\u5934\u5230\u5c3e\u81ea\u52a8\u5b8c\u6210\u79d1\u5b66\u7814\u7a76\u3002\u5b83\u80fd\u63d0\u51fa\u7814\u7a76\u60f3\u6cd5\u3001\u8bbe\u8ba1\u5b9e\u9a8c\u3001\u8fd0\u884c\u4ee3\u7801\u3001\u5206\u6790\u6570\u636e\uff0c\u6700\u540e\u5199\u51fa\u79d1\u5b66\u8bba\u6587\u30022025 \u5e74 4 \u6708\uff0c\u8fd9\u4e2a\u5de5\u5177\u5728 GitHub \u4e0a\u5f00\u6e90\uff0c\u5347\u7ea7\u4e86\u7b2c\u4e00\u7248\uff0c\u52a0\u5165\u4e86 Agentic Tree Search \u6280\u672f\uff0c\u8ba9\u63a2\u7d22\u66f4\u667a\u80fd\u3002\u5b83\u751f\u6210\u7684\u7b2c\u4e00\u7bc7\u5b8c\u5168\u7531 AI \u64b0\u5199\u7684\u8bba\u6587\u5df2\u901a\u8fc7 ICLR 2025 \u7814\u8ba8\u4f1a\u7684\u540c\u884c\u8bc4\u5ba1\u3002AI-Scientist-v2 \u4e0d\u4f9d\u8d56\u4eba\u7c7b\u6a21\u677f\uff0c\u9002\u7528\u4e8e\u591a\u79cd\u673a\u5668\u5b66\u4e60\u9886\u57df\uff0c\u9002\u5408\u7814\u7a76\u4eba\u5458\u548c\u5f00\u53d1\u8005\u4f7f\u7528\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-30192\" title=\"AI-Scientist-v2\uff1a\u81ea\u4e3b\u5b8c\u6210\u79d1\u7814\u7814\u7a76\u4e0e\u8bba\u6587\u64b0\u5199-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/04\/07dd797f1868978.png\" alt=\"AI-Scientist-v2\uff1a\u81ea\u4e3b\u5b8c\u6210\u79d1\u7814\u7814\u7a76\u4e0e\u8bba\u6587\u64b0\u5199-1\" width=\"747\" height=\"339\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/04\/07dd797f1868978.png 747w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/04\/07dd797f1868978-18x8.png 18w\" sizes=\"auto, (max-width: 747px) 100vw, 747px\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li><strong>\u63d0\u51fa\u7814\u7a76\u60f3\u6cd5<\/strong>\uff1a\u6839\u636e\u8f93\u5165\u65b9\u5411\uff0c\u81ea\u52a8\u751f\u6210\u53ef\u884c\u7684\u79d1\u7814\u70b9\u5b50\u3002<\/li>\n<li><strong>\u7f16\u5199\u5b9e\u9a8c\u4ee3\u7801<\/strong>\uff1a\u751f\u6210\u8fd0\u884c\u5b9e\u9a8c\u6240\u9700\u7684\u4ee3\u7801\uff0c\u652f\u6301\u8c03\u6574\u548c\u4f18\u5316\u3002<\/li>\n<li><strong>\u6267\u884c\u5b9e\u9a8c\u4e0e\u5206\u6790<\/strong>\uff1a\u81ea\u52a8\u8fd0\u884c\u4ee3\u7801\uff0c\u6536\u96c6\u6570\u636e\uff0c\u751f\u6210\u56fe\u8868\u3002<\/li>\n<li><strong>\u64b0\u5199\u79d1\u5b66\u8bba\u6587<\/strong>\uff1a\u6839\u636e\u5b9e\u9a8c\u7ed3\u679c\uff0c\u8f93\u51fa\u683c\u5f0f\u89c4\u8303\u7684\u8bba\u6587\u3002<\/li>\n<li><strong>\u667a\u80fd\u8def\u5f84\u4f18\u5316<\/strong>\uff1a\u901a\u8fc7 Agentic Tree Search\uff0c\u63a2\u7d22\u6700\u4f73\u7814\u7a76\u65b9\u6848\u3002<\/li>\n<li><strong>\u6587\u732e\u641c\u7d22\u652f\u6301<\/strong>\uff1a\u53ef\u9009\u63a5\u5165 Semantic Scholar API\uff0c\u68c0\u67e5\u65b0\u9896\u6027\u5e76\u6dfb\u52a0\u5f15\u7528\u3002<\/li>\n<li><strong>\u5f00\u6e90\u4ee3\u7801<\/strong>\uff1a\u63d0\u4f9b\u5b8c\u6574\u4ee3\u7801\uff0c\u7528\u6237\u53ef\u81ea\u7531\u4fee\u6539\u548c\u6269\u5c55\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<p>AI-Scientist-v2 \u9700\u8981\u4e00\u5b9a\u7684\u6280\u672f\u57fa\u7840\uff0c\u4f46\u914d\u7f6e\u597d\u540e\u80fd\u5927\u5e45\u7b80\u5316\u79d1\u7814\u5de5\u4f5c\u3002\u4ee5\u4e0b\u662f\u8be6\u7ec6\u6b65\u9aa4\uff0c\u5e2e\u52a9\u7528\u6237\u5feb\u901f\u4e0a\u624b\u3002<\/p>\n<h3>\u5b89\u88c5\u6d41\u7a0b<\/h3>\n<ol>\n<li><strong>\u51c6\u5907\u73af\u5883<\/strong>\n<ul>\n<li>\u9700\u8981 Linux \u7cfb\u7edf\u548c NVIDIA GPU\uff0c\u652f\u6301 CUDA \u548c PyTorch\u3002<\/li>\n<li>\u521b\u5efa Python 3.11 \u73af\u5883\uff1a\n<pre><code>conda create -n ai_scientist python=3.11\r\nconda activate ai_scientist\r\n<\/code><\/pre>\n<\/li>\n<li>\u5b89\u88c5 PyTorch \u548c CUDA\uff1a\n<pre><code>conda install pytorch torchvision torchaudio pytorch-cuda=12.4 -c pytorch -c nvidia\r\n<\/code><\/pre>\n<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u5b89\u88c5\u4f9d\u8d56<\/strong>\n<ul>\n<li>\u4e0b\u8f7d\u4ee3\u7801\uff1a\n<pre><code>git clone https:\/\/github.com\/SakanaAI\/AI-Scientist-v2.git\r\ncd AI-Scientist-v2\r\n<\/code><\/pre>\n<\/li>\n<li>\u5b89\u88c5\u5176\u4ed6\u5de5\u5177\uff1a\n<pre><code>conda install anaconda::poppler  # \u5904\u7406 PDF\r\nconda install conda-forge::chktex  # \u68c0\u67e5\u8bba\u6587\u683c\u5f0f\r\npip install -r requirements.txt\r\n<\/code><\/pre>\n<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u914d\u7f6e API<\/strong>\n<ul>\n<li>\u8bbe\u7f6e\u5927\u6a21\u578b API \u5bc6\u94a5\uff08\u5982 OpenAI\uff09\uff1a\n<pre><code>export OPENAI_API_KEY='\u4f60\u7684\u5bc6\u94a5'\r\n<\/code><\/pre>\n<\/li>\n<li>\u5982\u679c\u7528 <a href=\"https:\/\/www.kdjingpai.com\/ja\/claudeanquanfubai\/\">Claude<\/a> \u6a21\u578b\uff0c\u5b89\u88c5\u989d\u5916\u652f\u6301\uff1a\n<pre><code>pip install anthropic[bedrock]\r\n<\/code><\/pre>\n<p>\u8bbe\u7f6e AWS \u5bc6\u94a5\u548c\u533a\u57df\uff1a<\/p>\n<pre><code>export AWS_ACCESS_KEY_ID='\u4f60\u7684ID'\r\nexport AWS_SECRET_ACCESS_KEY='\u4f60\u7684\u5bc6\u94a5'\r\nexport AWS_REGION_NAME='us-west-2'\r\n<\/code><\/pre>\n<\/li>\n<li>\u53ef\u9009\u914d\u7f6e Semantic Scholar API\uff1a\n<pre><code>export S2_API_KEY='\u4f60\u7684\u5bc6\u94a5'\r\n<\/code><\/pre>\n<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u6d4b\u8bd5\u73af\u5883<\/strong>\n<ul>\n<li>\u68c0\u67e5 GPU \u662f\u5426\u53ef\u7528\uff1a\n<pre><code>python -c \"import torch; print(torch.cuda.is_available())\"\r\n<\/code><\/pre>\n<\/li>\n<li>\u8f93\u51fa\u00a0<code>True<\/code>\u00a0\u8868\u793a\u5b89\u88c5\u6210\u529f\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n<h3>\u5982\u4f55\u4f7f\u7528\u4e3b\u8981\u529f\u80fd<\/h3>\n<h4>1. \u751f\u6210\u7814\u7a76\u60f3\u6cd5<\/h4>\n<ul>\n<li>\u8fdb\u5165\u4ee3\u7801\u76ee\u5f55\uff0c\u8fd0\u884c\uff1a<\/li>\n<\/ul>\n<pre><code>python launch_scientist_bfts.py --load_ideas \"ai_scientist\/ideas\/i_cant_believe_its_not_better.json\" --model_writeup \"claude-3-5-sonnet-20240620\"\r\n<\/code><\/pre>\n<ul>\n<li>\u7cfb\u7edf\u4f1a\u751f\u6210\u4e00\u4e2a JSON \u6587\u4ef6\uff0c\u5305\u542b\u7814\u7a76\u6807\u9898\u548c\u63cf\u8ff0\u3002<\/li>\n<\/ul>\n<h4>2. \u8fd0\u884c\u5b9e\u9a8c<\/h4>\n<ul>\n<li>\u60f3\u6cd5\u751f\u6210\u540e\uff0c\u7cfb\u7edf\u4f1a\u521b\u5efa\u5b9e\u9a8c\u4ee3\u7801\uff08\u5982\u00a0<code>experiment.py<\/code>\uff09\u3002<\/li>\n<li>\u6267\u884c\u5b9e\u9a8c\uff1a<\/li>\n<\/ul>\n<pre><code>python experiment.py\r\n<\/code><\/pre>\n<ul>\n<li>\u7ed3\u679c\u4fdd\u5b58\u5728\u00a0<code>experiments<\/code>\u00a0\u6587\u4ef6\u5939\u4e0b\u7684\u65e5\u5fd7\u4e2d\uff0c\u5305\u62ec\u6570\u636e\u548c\u56fe\u8868\u3002<\/li>\n<\/ul>\n<h4>3. \u64b0\u5199\u8bba\u6587<\/h4>\n<ul>\n<li>\u5b9e\u9a8c\u5b8c\u6210\u540e\uff0c\u751f\u6210\u8bba\u6587\uff1a<\/li>\n<\/ul>\n<pre><code>python launch_scientist_bfts.py --load_code --add_dataset_ref --model_writeup \"o1-preview-2024-09-12\" --model_citation \"gpt-4o-2024-11-20\"\r\n<\/code><\/pre>\n<ul>\n<li>\u8f93\u51fa LaTeX \u6587\u4ef6\uff0c\u5b58\u653e\u5728\u00a0<code>experiments\/timestamp_ideaname\/latex<\/code>\u00a0\u6587\u4ef6\u5939\u3002\u7528 LaTeX \u7f16\u8f91\u5668\u7f16\u8bd1\u5373\u53ef\u67e5\u770b\u3002<\/li>\n<\/ul>\n<h4>4. \u4f7f\u7528 Agentic Tree Search<\/h4>\n<ul>\n<li>\u8fd9\u662f v2 \u7684\u6838\u5fc3\u529f\u80fd\uff0c\u80fd\u4f18\u5316\u7814\u7a76\u8def\u5f84\u3002<\/li>\n<li>\u8fd0\u884c\u65f6\u52a0\u5165\u53c2\u6570\uff1a<\/li>\n<\/ul>\n<pre><code>python launch_scientist_bfts.py --load_ideas \"ai_scientist\/ideas\/i_cant_believe_its_not_better.json\" --tree-search\r\n<\/code><\/pre>\n<ul>\n<li>\u751f\u6210\u00a0<code>unified_tree_viz.html<\/code>\uff0c\u7528\u6d4f\u89c8\u5668\u6253\u5f00\u53ef\u67e5\u770b\u641c\u7d22\u8fc7\u7a0b\u3002<\/li>\n<\/ul>\n<h4>5. \u914d\u7f6e\u6811\u641c\u7d22\u53c2\u6570<\/h4>\n<ul>\n<li>\u7f16\u8f91\u00a0<code>bfts_config.yaml<\/code>\u00a0\u6587\u4ef6\uff1a<\/li>\n<li><code>num_workers<\/code>\uff1a\u5e76\u884c\u5904\u7406\u7684\u8282\u70b9\u6570\uff0c\u5982 3\u3002<\/li>\n<li><code>steps<\/code>\uff1a\u6700\u5927\u63a2\u7d22\u8282\u70b9\u6570\uff0c\u5982 21\u3002<\/li>\n<li><code>num_drafts<\/code>\uff1a\u521d\u59cb\u7814\u7a76\u65b9\u5411\u6570\u91cf\u3002<\/li>\n<li><code>max_debug_depth<\/code>\uff1a\u8c03\u8bd5\u5c1d\u8bd5\u6b21\u6570\u3002<\/li>\n<\/ul>\n<h3>\u6ce8\u610f\u4e8b\u9879<\/h3>\n<ul>\n<li><strong>\u5b89\u5168\u6027<\/strong>\uff1a\u4ee3\u7801\u4f1a\u6267\u884c AI \u5199\u7684\u7a0b\u5e8f\uff0c\u53ef\u80fd\u8c03\u7528\u5371\u9669\u5305\u6216\u8054\u7f51\uff0c\u5efa\u8bae\u7528 Docker \u8fd0\u884c\u3002<\/li>\n<li><strong>\u6210\u672c<\/strong>\uff1a\u6bcf\u6b21\u5b9e\u9a8c\u7ea6 15-20 \u7f8e\u5143\uff0c\u8bba\u6587\u64b0\u5199\u52a0 5 \u7f8e\u5143\u3002<\/li>\n<li><strong>\u6210\u529f\u7387<\/strong>\uff1av2 \u63a2\u7d22\u6027\u5f3a\uff0c\u6210\u529f\u7387\u4f4e\u4e8e v1\uff0c\u9002\u5408\u5f00\u653e\u6027\u7814\u7a76\u3002<\/li>\n<li><strong>\u5185\u5b58\u95ee\u9898<\/strong>\uff1a\u82e5\u63d0\u793a \u201cCUDA Out of Memory\u201d\uff0c\u5728 JSON \u6587\u4ef6\u4e2d\u6362\u5c0f\u6a21\u578b\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u4e9b\u6b65\u9aa4\u80fd\u8ba9\u4f60\u5b8c\u6574\u4f53\u9a8c AI-Scientist-v2 \u7684\u79d1\u7814\u81ea\u52a8\u5316\u529f\u80fd\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>\u5e94\u7528\u573a\u666f<\/h2>\n<ol>\n<li><strong>\u5b66\u672f\u7814\u7a76<\/strong><br \/>\n\u7814\u7a76\u4eba\u5458\u7528\u5b83\u9a8c\u8bc1\u65b0\u7b97\u6cd5\uff0c\u751f\u6210\u8bba\u6587\u521d\u7a3f\uff0c\u8282\u7701\u65f6\u95f4\u3002<\/li>\n<li><strong>\u6559\u80b2\u5b66\u4e60<\/strong><br \/>\n\u5b66\u751f\u7528\u5b83\u6a21\u62df\u79d1\u7814\uff0c\u751f\u6210\u62a5\u544a\uff0c\u5b66\u4e60\u5b9e\u9a8c\u8bbe\u8ba1\u3002<\/li>\n<li><strong>\u6280\u672f\u521b\u65b0<\/strong><br \/>\n\u5f00\u53d1\u8005\u7528\u5b83\u6d4b\u8bd5\u65b0\u60f3\u6cd5\uff0c\u5feb\u901f\u751f\u6210\u4ee3\u7801\u539f\u578b\u3002<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<h2>QA<\/h2>\n<ol>\n<li><strong>\u652f\u6301\u54ea\u4e9b\u6a21\u578b\uff1f<\/strong><br \/>\n\u652f\u6301 Claude 3.5 Sonnet\u3001GPT-4o\u3001o1-preview \u7b49\uff0c\u89c1\u00a0<code>llm.py<\/code>\u00a0\u6587\u4ef6\u3002<\/li>\n<li><strong>\u5b9e\u9a8c\u6210\u672c\u591a\u5c11\uff1f<\/strong><br \/>\n\u7528 Claude 3.5 \u6bcf\u6b21\u7ea6 15-20 \u7f8e\u5143\uff0c\u5199\u4f5c\u52a0 5 \u7f8e\u5143\u3002<\/li>\n<li><strong>\u8bba\u6587\u751f\u6210\u5931\u8d25\u600e\u4e48\u529e\uff1f<\/strong><br \/>\n\u6210\u529f\u7387\u56e0\u6a21\u578b\u548c\u60f3\u6cd5\u590d\u6742\u5ea6\u800c\u5f02\uff0c\u53ef\u8c03\u6574\u53c2\u6570\u6216\u6362\u6a21\u578b\u91cd\u8bd5\u3002<\/li>\n<li><strong>\u5982\u4f55\u6dfb\u52a0\u65b0\u7814\u7a76\u65b9\u5411\uff1f<\/strong><br \/>\n\u5728\u00a0<code>ai_scientist\/ideas\/<\/code>\u00a0\u76ee\u5f55\u4e0b\u65b0\u589e JSON \u6587\u4ef6\uff0c\u53c2\u8003\u793a\u4f8b\u4fee\u6539\u3002<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>AI-Scientist-v2 \u662f\u7531\u65e5\u672c SakanaAI \u516c\u53f8\u5f00\u53d1\u7684\u667a\u80fd\u7cfb\u7edf\uff0c\u65e8\u5728\u8ba9\u673a\u5668\u4ece\u5934\u5230\u5c3e\u81ea\u52a8\u5b8c\u6210\u79d1\u5b66\u7814\u7a76\u3002\u5b83\u80fd\u63d0\u51fa\u7814\u7a76\u60f3\u6cd5\u3001\u8bbe\u8ba1\u5b9e\u9a8c\u3001\u8fd0\u884c\u4ee3\u7801\u3001\u5206\u6790\u6570\u636e\uff0c\u6700\u540e\u5199\u51fa\u79d1\u5b66\u8bba\u6587\u30022025 \u5e74 4 \u6708\uff0c\u8fd9\u4e2a\u5de5\u5177\u5728 GitHub \u4e0a\u5f00\u6e90\uff0c\u5347&#8230;<\/p>\n","protected":false},"author":1,"featured_media":62215,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[389,205],"class_list":["post-30191","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tool","tag-shengchengshenduyanjiuao","tag-lunwen"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/30191","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/comments?post=30191"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/30191\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media\/62215"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media?parent=30191"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/categories?post=30191"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/tags?post=30191"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}