{"id":9642,"date":"2026-06-26T04:31:16","date_gmt":"2026-06-25T20:31:16","guid":{"rendered":"https:\/\/infernews.com\/blog\/cf-world\/"},"modified":"2026-06-26T04:31:16","modified_gmt":"2026-06-25T20:31:16","slug":"cf-world","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/cf-world\/","title":{"rendered":"CF-World \u8a55\u6e2c\uff1a\u63ed\u7a7f\u6587\u751f\u5716\u6a21\u578b\u7684\u300c\u6b78\u7d0d\u706b\u96de\u300d\u76f2\u9ede"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/06\/pasted-c75aecc786b5.jpg\" alt=\"Repository image for jylei16\/CF-World\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">CF-World \u662f\u4e00\u500b\u5c08\u9580\u91dd\u5c0d\u6587\u751f\u5716\uff08text-to-image, T2I\uff09\u6a21\u578b\u7684\u57fa\u6e96\u6e2c\u8a66\u8207\u7814\u7a76\u539f\u578b\uff0c\u7528\u4ee5\u5224\u65b7\u6a21\u578b\u5728\u9762\u5c0d\u9055\u53cd\u5e38\u8b58\u7684\u6307\u4ee4\u6642\uff0c\u5230\u5e95\u662f\u5728\u63a8\u7406\uff0c\u9084\u662f\u50c5\u50c5\u8907\u88fd\u8a13\u7df4\u6578\u64da\u4e2d\u7684\u9ad8\u983b\u6a21\u5f0f\u3002\u73fe\u6709\u7684 T2I \u6a21\u578b\u5728\u65e5\u5e38\u8a9e\u5883\u4e0b\u8868\u73fe\u51fa\u8272\uff0c\u4f46\u53ea\u8981\u7269\u7406\u6cd5\u5247\u88ab\u523b\u610f\u6539\u5beb\uff0c\u4f8b\u5982\u8981\u6c42\u5b83\u5011\u751f\u6210\u300c\u91cd\u529b\u53cd\u8f49\u300d\u6216\u300c\u5149\u7dda\u53cd\u5411\u6298\u5c04\u300d\u7684\u756b\u9762\uff0c\u4fbf\u6703\u51fa\u73fe\u660e\u986f\u5d29\u6f70\u3002CF-World \u63a1\u7528\u4e09\u5c64\u905e\u9032\u8a2d\u8a08\u4f86\u66b4\u9732\u9019\u7a2e\u843d\u5dee\uff1aL1 \u70ba\u4e8b\u5be6\u751f\u6210\uff0c\u8981\u6c42\u6a21\u578b\u6309\u771f\u5be6\u4e16\u754c\u77e5\u8b58\u4f5c\u756b\uff1bL2 \u70ba\u986f\u5f0f\u53cd\u4e8b\u5be6\uff08Explicit Counterfactual\uff09\uff0c\u540c\u6642\u63d0\u4f9b\u53cd\u4e8b\u5be6\u524d\u63d0\u8207\u6307\u5b9a\u7684\u8996\u89ba\u7d50\u679c\uff0c\u6e2c\u8a66\u6a21\u578b\u80fd\u5426\u4f9d\u6307\u4ee4\u8abf\u6574\uff1bL3 \u70ba\u96b1\u5f0f\u53cd\u4e8b\u5be6\uff08Implicit Counterfactual\uff09\uff0c\u53ea\u7d66\u51fa\u53cd\u4e8b\u5be6\u689d\u4ef6\uff0c\u8981\u6c42\u6a21\u578b\u81ea\u884c\u63a8\u5c0e\u61c9\u6709\u7684\u8996\u89ba\u5448\u73fe\uff0c\u5f9e\u800c\u8003\u9a57\u771f\u6b63\u7684\u56e0\u679c\u63a8\u6f14\u80fd\u529b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u70ba\u4e86\u91cf\u5316\u9019\u7a2e\u843d\u5dee\uff0c\u9805\u76ee\u5f15\u5165\u5169\u9805\u6307\u6a19\uff1aPRR\uff08Prior Resistance Rate\uff0c\u5148\u9a57\u62b5\u6297\u7387\uff09\u8861\u91cf\u6a21\u578b\u64fa\u812b\u65e2\u5b9a\u8996\u89ba\u6163\u6027\u7684\u80fd\u529b\uff0cRRR\uff08Reasoning Retention Rate\uff0c\u63a8\u7406\u4fdd\u7559\u7387\uff09\u5247\u6aa2\u9a57\u6a21\u578b\u5728\u591a\u6b65\u9a5f\u6307\u4ee4\u4e0b\u80fd\u5426\u7dad\u6301\u908f\u8f2f\u9023\u8cab\u6027\u3002\u5132\u5b58\u5eab\u9084\u5305\u542b\u56e0\u679c\u89e3\u8026\uff08Causal Decoupling\uff09\u3001\u5c6c\u6027\u89e3\u8026\uff08Attribute Decoupling\uff09\u8207\u53bb\u7bc4\u5f0f\u5316\uff08De-nominalization, De-norm\uff09\u4e09\u689d\u5c08\u9580\u8a55\u6e2c\u7dda\uff0c\u5354\u52a9\u7814\u7a76\u8005\u5340\u5206\u5931\u6557\u7a76\u7adf\u6e90\u81ea\u56e0\u679c\u8b8a\u91cf\u7121\u6cd5\u5206\u96e2\uff0c\u9084\u662f\u6e90\u81ea\u8a9e\u8a00\u5148\u9a57\u7684\u300c\u6982\u5ff5\u9396\u5b9a\u300d\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5728\u8a55\u4f30\u5c0d\u8c61\u65b9\u9762\uff0cCF-World \u6db5\u84cb FLUX.2-dev\u3001Qwen-image\u3001Nano Banana \u7b49\u8fd1\u671f\u6a21\u578b\uff0c\u7d50\u679c\u986f\u793a L1 \u8868\u73fe\u826f\u597d\u7684\u6a21\u578b\u5728 L3 \u5834\u666f\u4e2d\u666e\u904d\u51fa\u73fe\u4e00\u81f4\u6027\u6025\u5287\u4e0b\u964d\uff0c\u8aaa\u660e\u9ad8\u7dad\u7d71\u8a08\u5148\u9a57\u6b63\u5728\u58d3\u5236\u771f\u6b63\u7684\u56e0\u679c\u63a8\u7406\u3002\u4ee3\u78bc\u5eab\u7d50\u69cb\u6e05\u6670\uff1aeval_questions \u6536\u9304\u9810\u5148\u751f\u6210\u7684\u8a55\u6e2c\u984c\u76ee\uff0cprompt \u5b58\u653e\u57fa\u790e\u63d0\u793a\u8a5e\u8207\u53cd\u4e8b\u5be6\u898f\u5247\uff0cscripts \u5247\u6db5\u84cb\u984c\u76ee\u751f\u6210\u53ca\u57fa\u65bc VLM \u7684\u81ea\u52d5\u8a55\u5206\uff08\u652f\u63f4 Gemini \u8207 Qwen3-VL\uff09\u3002\u5c0d\u5f9e\u4e8b\u591a\u6a21\u614b\u6a21\u578b\u8a55\u6e2c\u3001\u8996\u89ba\u63a8\u7406\u7814\u7a76\u6216\u95dc\u5fc3\u6a21\u578b\u5b89\u5168\u908a\u754c\u7684\u5718\u968a\u800c\u8a00\uff0c\u9019\u500b\u57fa\u6e96\u63d0\u4f9b\u4e86\u4e00\u500b\u53ef\u91cd\u73fe\u4e14\u5206\u5c64\u7d30\u7dfb\u7684\u6e2c\u8a66\u5e73\u53f0\uff0c\u6709\u52a9\u65bc\u5b9a\u4f4d\u300c\u6a21\u578b\u7a76\u7adf\u5361\u5728\u54ea\u500b\u74b0\u7bc0\u300d\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"repository-structure\">\ud83d\udcc2 Repository Structure<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">The repository is organized into prompts, pre-generated evaluation questions, and execution scripts:<\/p>\n\n\n\n<div class=\"codecopy-container\">\n                <div class=\"codecopy-header\">\n                    \n                    <button class=\"codecopy-btn\" data-target=\"codecopy-6a3dbc696320d\" title=\"Copy code to clipboard\">\n                <svg width=\"16\" height=\"16\" viewBox=\"0 0 24 24\" fill=\"none\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\">\n                    <path d=\"M16 1H4C2.9 1 2 1.9 2 3V17H4V3H16V1ZM19 5H8C6.9 5 6 5.9 6 7V21C6 22.1 6.9 23 8 23H19C20.1 23 21 22.1 21 21V7C21 5.9 20.1 5 19 5ZM19 21H8V7H19V21Z\" fill=\"currentColor\"\/>\n                <\/svg>\n                <span class=\"codecopy-text\">Copy<\/span>\n            <\/button>\n                <\/div>\n                <pre class=\"line-numbers\" id=\"codecopy-6a3dbc696320d\"><code class=\"language-text\" class=\"language-text\">\u251c\u2500\u2500 eval_questions\/        # Pre-generated evaluation questions (categorized by discipline)\n\u2502   \u251c\u2500\u2500 physics\/           # Physics sub-disciplines (Astronomy, Mechanics, etc.)\n\u2502   \u2514\u2500\u2500 ...\n\u251c\u2500\u2500 prompt\/                # Raw base prompts and counterfactual rules\n\u2502   \u251c\u2500\u2500 physics\/\n\u2502   \u2514\u2500\u2500 ...\n\u2514\u2500\u2500 scripts\/               # Core execution scripts\n    \u251c\u2500\u2500 generate_eval\/     # Scripts to generate evaluation questions\n    \u2502   \u251c\u2500\u2500 gemini.py      # Generates standard CF-World questions via Gemini\n    \u2502   \u2514\u2500\u2500 rule_decouple.py # Generates questions for the Causal Decoupling experiment\n    \u2514\u2500\u2500 score\/             # Automated VLM-based scoring scripts\n        \u251c\u2500\u2500 gemini.py      # Standard multi-dimensional scoring using Gemini\n        \u251c\u2500\u2500 qwen3vl-235b.py# Standard multi-dimensional scoring using Qwen3-VL\n        \u251c\u2500\u2500 rule_decouple.py # Scoring for the Causal Decoupling experiment\n        \u251c\u2500\u2500 attribute_decouple.py # Scoring for the Attribute Decoupling experiment\n        \u2514\u2500\u2500 denorm.py      # Scoring for the De-nominalization (De-norm) experiment\n<\/code><\/pre>\n            <\/div>\n\n\n<hr \/>\n\n\n<p class=\"wp-block-paragraph\"><strong>GitHub\uff1a<\/strong> <a href=\"https:\/\/github.com\/jylei16\/CF-World\" rel=\"noopener noreferrer\">https:\/\/github.com\/jylei16\/CF-World<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9805\u76ee\u4e3b\u9801\uff1a<\/strong> <a href=\"https:\/\/jylei16.github.io\/CF-World.github.io\/\" rel=\"noopener noreferrer\">https:\/\/jylei16.github.io\/CF-World.github.io\/<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Paper\uff1a<\/strong> <a href=\"https:\/\/arxiv.org\/pdf\/2606.24548\" rel=\"noopener noreferrer\">https:\/\/arxiv.org\/pdf\/2606.24548<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>CF-World \u4ee5\u4e09\u5c64\u53cd\u4e8b\u5be6\u6846\u67b6\uff0c\u6e2c\u8a66 T2I \u6a21\u578b\u662f\u5426\u771f\u6b63\u7406\u89e3\u56e0\u679c\uff0c\u800c\u975e\u50c5\u9760\u8a18\u61b6\u7d71\u8a08\u95dc\u806f\u3002<\/p>\n","protected":false},"author":8,"featured_media":9641,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,171,163,164,30,129,170,161,197,200],"tags":[],"class_list":["post-9642","post","type-post","status-publish","format-standard","hentry","category-133","category-171","category-163","category-164","category-image","category-txt2img","category-170","category-161","category-framework","category-200"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9642","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=9642"}],"version-history":[{"count":1,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9642\/revisions"}],"predecessor-version":[{"id":9644,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9642\/revisions\/9644"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/9641"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=9642"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=9642"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=9642"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}