{"id":1535,"date":"2025-09-19T12:08:16","date_gmt":"2025-09-19T04:08:16","guid":{"rendered":"http:\/\/www.applicationsofllm.com\/?p=1535"},"modified":"2025-09-19T12:08:16","modified_gmt":"2025-09-19T04:08:16","slug":"%e7%90%86%e8%a7%a3%e5%b8%ae%e5%8a%a9%e7%94%9f%e6%88%90%ef%bc%9freca%e8%87%aa%e7%9b%91%e7%9d%a3%e8%ae%ad%e7%bb%83%e8%ae%a9%e7%bb%9f%e4%b8%80%e5%a4%9a%e6%a8%a1%e6%80%81%e6%a8%a1%e5%9e%8b%e7%9b%b4","status":"publish","type":"post","link":"http:\/\/www.applicationsofllm.com\/index.php\/2025\/09\/19\/%e7%90%86%e8%a7%a3%e5%b8%ae%e5%8a%a9%e7%94%9f%e6%88%90%ef%bc%9freca%e8%87%aa%e7%9b%91%e7%9d%a3%e8%ae%ad%e7%bb%83%e8%ae%a9%e7%bb%9f%e4%b8%80%e5%a4%9a%e6%a8%a1%e6%80%81%e6%a8%a1%e5%9e%8b%e7%9b%b4\/","title":{"rendered":"\u7406\u89e3\u5e2e\u52a9\u751f\u6210\uff1fRecA\u81ea\u76d1\u7763\u8bad\u7ec3\u8ba9\u7edf\u4e00\u591a\u6a21\u6001\u6a21\u578b\u76f4\u5347SOTA"},"content":{"rendered":"\n<p>\u7edf\u4e00\u591a\u6a21\u6001\u6a21\u578b\uff08Unified Multimodal Models, UMMs\uff09\u65e8\u5728\u5c06\u89c6\u89c9\u7406\u89e3\u548c\u751f\u6210\u7edf\u4e00\u4e8e\u5355\u4e00\u6a21\u578b\u67b6\u6784\u3002UMM \u7ee7\u627f\u4e86\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b (Multimodal Large Language Models, MLLMs) \u53ef\u4ee5\u5f88\u8f7b\u677e\u5730\u8fa8\u522b\u7269\u4f53\u7684\u5de6\u53f3\u3001\u989c\u8272\u3001\u79cd\u7c7b\u3002\u4f46\u662f\u5f88\u591a\u751f\u6210\u6a21\u578b\u8fde\u300c\u4e00\u53ea\u9ed1\u8272\u7684\u732b\u548c\u767d\u8272\u7684\u72d7\u300d\uff0c\u300c\u9ec4\u8272\u897f\u5170\u82b1\u300d\u90fd\u65e0\u6cd5\u751f\u6210\u3002\u8fd9\u4f53\u73b0\u4e86\u5f53\u524d\u7edf\u4e00\u591a\u6a21\u6001\u6a21\u578b\u5728\u89c6\u89c9\u7406\u89e3\u548c\u751f\u6210\u80fd\u529b\u4e0a\u7684\u4e0d\u5e73\u8861\uff1a\u5b83\u4eec\u5f80\u5f80\u5728\u7406\u89e3\u56fe\u50cf\u5185\u5bb9\u65b9\u9762\u8868\u73b0\u51fa\u8272\uff0c\u4f46\u5728\u6839\u636e\u6587\u672c\u63cf\u8ff0\u751f\u6210\u56fe\u50cf\u65f6\u5374\u529b\u4e0d\u4ece\u5fc3\u3002\u8fd9\u662f\u4e3a\u4ec0\u4e48\u5462\uff1f<\/p>\n\n\n\n<p>\u5b9e\u9645\u4e0a\uff0c\u56fe\u7247\u662f\u4e00\u4e2a\u300c\u7a20\u5bc6\u300d\u7684\u6a21\u6001\uff0c\u6587\u5b57\u662f\u4e00\u4e2a\u300c\u7a00\u758f\u300d\u7684\u6a21\u6001\uff0c\u4ece\u4e00\u4e2a\u7a20\u5bc6\u7684\u4fe1\u606f\u91cc\u63d0\u53d6\u7a00\u758f\u7684\u4fe1\u606f\uff08VQA\uff0cImage Captioning\uff09\u662f\u76f8\u5bf9\u8f7b\u677e\u7684\uff0c\u4f46\u662f\u8981\u4ece\u7a00\u758f\u7684\u4fe1\u606f\u53bb\u6784\u5efa\u7a20\u5bc6\u7684\u4fe1\u606f\u5219\u66f4\u4e3a\u56f0\u96be\u3002\u4f20\u7edf\u7684\u6587\u751f\u56fe\u8bad\u7ec3\u4f9d\u8d56\u5927\u89c4\u6a21\u7684\u56fe\u50cf &#8211; \u6587\u672c\u5bf9\u6570\u636e\uff0c\u8fd9\u4e9b\u6587\u672c\u63cf\u8ff0 (text caption) \u65e0\u6cd5\u5b8c\u6574\u7684\u8868\u8ff0\u56fe\u7247\u91cc\u7684\u6240\u6709\u4fe1\u606f\u3002\u6bd4\u5982\u7269\u4f53\u4f4d\u7f6e\u5173\u7cfb\u3001\u51e0\u4f55\u7ed3\u6784\uff0c\u7269\u4f53\u7684\u7eb9\u7406\u548c\u98ce\u683c\u7b49\u3002\u8fd9\u53ef\u80fd\u5bfc\u81f4\u56fe\u50cf\u751f\u6210\u6a21\u578b\u5b66\u5230\u4e0d\u5b8c\u6574\u751a\u81f3\u6709\u504f\u5dee\u7684\u89c6\u89c9\u6982\u5ff5\uff08\u4f8b\u5982\uff0c\u5c06\u300c\u897f\u5170\u82b1\u300d\u4e0e\u300c\u7eff\u8272\u300d\u8054\u7cfb\u5728\u4e00\u8d77\uff0c\u5bfc\u81f4\u6a21\u578b\u65e0\u6cd5\u751f\u6210\u300c\u9ec4\u8272\u897f\u5170\u82b1\u300d\uff09\u3002\u6211\u4eec\u79f0\u8fd9\u79cd\u6587\u672c\u76d1\u7763\u4e3a\u300c\u7a00\u758f\u76d1\u7763\u300d(sparse supervision)\u3002<\/p>\n\n\n\n<p><strong>\u65b9\u6cd5\uff1a\u91cd\u5efa\u5bf9\u9f50 (Reconstruction Alignment, RecA)<\/strong><\/p>\n\n\n\n<p>\u6709\u6ca1\u6709\u300c\u7a20\u5bc6\u76d1\u7763\u300d(Dense Supervision)\uff0c\u53ef\u4ee5\u8ba9\u6a21\u578b\u5b66\u5230\u66f4\u5b8c\u6574\u7684\u89c6\u89c9\u6982\u5ff5\u5462\uff1f\u7b54\u6848\u662f\u6709\u7684\u3002\u56fe\u7247\u672c\u8eab\u6b63\u662f\u6700\u597d\u7684\u4fe1\u606f\u8f7d\u4f53\u3002UMM \u63d0\u4f9b\u4e86\u4e00\u4e2a\u5c06\u56fe\u7247\u4f5c\u4e3a\u300c\u63d0\u793a\u8bcd\u300d(prompt) \u8f93\u5165\u7684\u673a\u4f1a\u3002\u73b0\u5728\u7684 UMM \u7684\u89c6\u89c9\u7406\u89e3\u7f16\u7801\u5668 (Visual Understanding Encoder)\uff0c\u5982 CLIP, SigLIP\uff0c\u5df2\u7ecf\u53ef\u4ee5\u628a\u56fe\u7247\u6620\u5c04\u5230\u4e86 LLM \u7684\u8bed\u4e49\u7a7a\u95f4 (language-aligned semantic space)\u3002<\/p>\n\n\n\n<p>\u4ee5\u6b64\u4e3a\u52a8\u673a\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u7b80\u5355\u800c\u6709\u6548\u7684\u540e\u8bad\u7ec3\u65b9\u6cd5 \u2014\u2014 \u91cd\u5efa\u5bf9\u9f50\uff08Reconstruction Alignment, RecA\uff09\u3002RecA \u5e76\u975e\u5bf9\u6a21\u578b\u67b6\u6784\u672c\u8eab\u505a\u51fa\u6539\u52a8\uff0c\u800c\u662f\u5728\u6a21\u578b\u5e38\u89c4\u8bad\u7ec3\u5b8c\u6210\u540e\uff0c\u989d\u5916\u8fdb\u884c\u4e00\u9636\u6bb5\u81ea\u76d1\u7763\u7684\u540e\u8bad\u7ec3\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"243\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-619-1024x243.jpg\" alt=\"\" class=\"wp-image-1537\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-619-1024x243.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-619-300x71.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-619-768x182.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-619.jpg 1109w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Arxiv\uff1ahttps:\/\/alphaxiv.org\/abs\/2509.07295<\/li>\n\n\n\n<li>\u4ee3\u7801\uff1ahttps:\/\/github.com\/HorizonWind2004\/reconstruction-alignment<\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1ahttps:\/\/reconstruction-alignment.github.io\/<\/li>\n<\/ul>\n\n\n\n<p>\u5177\u4f53\u6765\u8bf4\uff0c\u5728 RecA \u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u6a21\u578b\u9996\u5148\u5229\u7528\u5176\u89c6\u89c9\u7406\u89e3\u7f16\u7801\u5668\u4ece\u8f93\u5165\u56fe\u50cf\u63d0\u53d6\u51fa\u8bed\u4e49\u5d4c\u5165\u7279\u5f81\uff08\u4f8b\u5982\u91c7\u7528\u9884\u8bad\u7ec3\u7684 CLIP\u3001DINO \u7b49\u6a21\u578b\u83b7\u53d6\u56fe\u50cf\u7684\u9ad8\u7ef4\u8bed\u4e49\u8868\u793a\uff09\uff0c\u4e0e\u4e00\u4e2a\u6a21\u677f\u6587\u672c\u5d4c\u5165\u76f8\u878d\u5408\uff0c\u518d\u9001\u5165\u7edf\u4e00\u591a\u6a21\u6001\u6a21\u578b\uff0c\u4f7f\u5176\u4ee5\u6b64\u4e3a\u6761\u4ef6\u8bd5\u56fe\u91cd\u5efa\u51fa\u539f\u59cb\u8f93\u5165\u56fe\u50cf\u3002\u6839\u636e\u751f\u6210\u7684\u56fe\u50cf\u4e0e\u539f\u56fe\u50cf\u4e4b\u95f4\u7684\u5dee\u5f02\u8ba1\u7b97\u81ea\u76d1\u7763\u91cd\u5efa\u635f\u5931\uff0cRecA&nbsp;<strong>\u5c06\u89c6\u89c9\u7406\u89e3\u5206\u652f\u4e2d\u8574\u542b\u7684\u7ec6\u8282\u77e5\u8bc6\u6709\u6548\u5bf9\u9f50\u5230\u751f\u6210\u5206\u652f<\/strong>\u3002<\/p>\n\n\n\n<p>\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0cRecA \u7684\u8bad\u7ec3\u4e0d\u9700\u8981\u4efb\u4f55\u7684\u56fe\u50cf &#8211; \u6587\u672c\u5bf9\uff0c\u53ea\u9700\u672a\u6807\u6ce8\u7684\u56fe\u50cf\u5373\u53ef\u5b8c\u6210\u8bad\u7ec3\u3002\u8bad\u7ec3\u5b8c\u6210\u540e\uff0c\u6a21\u578b\u5728\u63a8\u7406\u65f6\u5e76\u4e0d\u9700\u8981\u989d\u5916\u8f93\u5165\u8fd9\u4e9b\u89c6\u89c9\u5d4c\u5165\uff0c\u4ecd\u7136\u50cf\u666e\u901a\u751f\u6210\u6a21\u578b\u4e00\u6837\uff0c\u4ec5\u901a\u8fc7\u6587\u672c\u63d0\u793a\u5373\u53ef\u5de5\u4f5c\uff1b\u6362\u8a00\u4e4b\uff0cRecA \u662f\u4e00\u79cd\u7eaf\u8bad\u7ec3\u9636\u6bb5\u7684\u5bf9\u9f50\u7b56\u7565\uff0c\u4e0d\u4f1a\u589e\u52a0\u63a8\u7406\u9636\u6bb5\u7684\u5f00\u9500\u6216\u6539\u53d8\u4f7f\u7528\u65b9\u5f0f\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"369\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-620-1024x369.jpg\" alt=\"\" class=\"wp-image-1538\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-620-1024x369.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-620-300x108.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-620-768x277.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-620.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p><strong>\u5b9e\u9a8c\u7ed3\u679c<\/strong><\/p>\n\n\n\n<p><strong>\u901a\u7528\u6027\uff08Generality\uff09<\/strong><\/p>\n\n\n\n<p>\u4e3a\u4e86\u9a8c\u8bc1 RecA \u7684\u6709\u6548\u6027\uff0c\u6211\u4eec\u5728\u56db\u79cd\u4ee3\u8868\u6027\u7684\u7edf\u4e00\u591a\u6a21\u6001\u6a21\u578b\u4e0a\u8fdb\u884c\u4e86\u5b9e\u9a8c\u30021. Show-o (AR), Harmon (AR+MAR), OpenUni (AR+Diffusion, Metaqueries \u5f00\u6e90\u7248), BAGEL (AR+Diffusion) \u7b49\u6a21\u578b\uff0c\u6db5\u76d6\u4e86\u5f53\u524d\u7684\u4e3b\u6d41\u67b6\u6784\u3002\u53ef\u4ee5\u53d1\u73b0\uff0cRecA \u5728\u6240\u6709\u6a21\u578b\u4e0a\u5747\u5e26\u6765\u4e86\u663e\u8457\u7684\u6027\u80fd\u63d0\u5347\uff0c\u663e\u793a\u51fa\u5176\u65b9\u6cd5\u7684\u901a\u7528\u6027\u548c\u7a33\u5065\u6027\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"284\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-621-1024x284.jpg\" alt=\"\" class=\"wp-image-1539\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-621-1024x284.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-621-300x83.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-621-768x213.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-621.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p><strong>SOTA \u7ed3\u679c\uff08State-of-the-art Results\uff09<\/strong><\/p>\n\n\n\n<p>\u6211\u4eec\u4f7f\u7528 RecA \u540e\u8bad\u7ec3\u5f97\u5230\u7684 Harmon-1.5B \u6a21\u578b\u5c55\u73b0\u51fa\u4e86\u6781\u5f3a\u7684\u63d0\u9ad8\uff0c\u5728\u4e0d\u4f7f\u7528 GPT-4o-Image \u84b8\u998f\u6570\u636e\u548c RLHF \u7684\u60c5\u51b5\u4e0b\uff0c\u5728 GenEval \u548c DPGBench \u4e0a\u8fbe\u5230\u4e86 0.86 \u548c 87.21 \u7684\u6210\u7ee9\u3002\u5982\u679c\u4f7f\u7528 GPT-4o-Image \u84b8\u998f\u6570\u636e BLIP3o-60k\uff0c\u901a\u8fc7\u4e24\u9636\u6bb5\u7b56\u7565\uff08\u5148\u8fdb\u884c\u6709\u76d1\u7763\u5fae\u8c03 SFT\uff0c\u518d\u8fdb\u884c RecA \u65e0\u76d1\u7763\u8bad\u7ec3\uff09\uff0cHarmon \u6a21\u578b\u7684\u6027\u80fd\u8fdb\u4e00\u6b65\u63d0\u5347\u5230 GenEval 0.90\uff0cDPGBench 88.15\uff0c\u5168\u9762\u5237\u65b0\u4e86\u73b0\u6709\u8bb0\u5f55\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"738\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-623-1024x738.jpg\" alt=\"\" class=\"wp-image-1541\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-623-1024x738.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-623-300x216.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-623-768x553.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-623.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p>\u5bf9\u4e8e BAGEL\uff0c\u6211\u4eec\u53d1\u73b0\u5176\u5728\u56fe\u50cf\u7f16\u8f91\u4efb\u52a1\u4e0a\u4e5f\u53d6\u5f97\u4e86\u663e\u8457\u63d0\u5347\u3002\u5728 ImgEdit \u57fa\u51c6\u4e0a\u7684\u8bc4\u5206\u4ece 3.38 \u63d0\u5347\u81f3 3.75\uff0cGEdit \u8bc4\u5206\u4ece 6.94 \u4e0a\u5347\u5230 7.25\u3002\u7ecf\u8fc7 RecA \u7684 BAGEL \u6a21\u578b\u5728\u67d0\u4e9b\u7f16\u8f91\u80fd\u529b\u4e0a\u8d85\u8d8a\u6700\u65b0\u7684 SOTA \u6a21\u578b\uff0c\u5982 Black Forest Labs \u63a8\u51fa\u7684 12 \u4ebf\u53c2\u6570\u56fe\u50cf\u7f16\u8f91\u6a21\u578b FLUX.1 Kontext\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"254\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-622-1024x254.jpg\" alt=\"\" class=\"wp-image-1540\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-622-1024x254.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-622-300x74.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-622-768x191.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-622.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p><strong>\u53ef\u89c6\u5316\u6548\u679c<\/strong><\/p>\n\n\n\n<p>\u751f\u6210\u80fd\u529b\u5c55\u793a\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"768\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-624-1024x768.jpg\" alt=\"\" class=\"wp-image-1542\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-624-1024x768.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-624-300x225.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-624-768x576.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-624.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p>\u7f16\u8f91\u80fd\u529b\u5c55\u793a\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"345\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-625-1024x345.jpg\" alt=\"\" class=\"wp-image-1543\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-625-1024x345.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-625-300x101.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-625-768x259.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-625.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p>\u8bad\u7ec3\u524d\u540e\u7684\u751f\u6210\u80fd\u529b\u5bf9\u6bd4\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"523\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-626-1024x523.jpg\" alt=\"\" class=\"wp-image-1544\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-626-1024x523.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-626-300x153.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-626-768x392.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-626.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p>\u8bad\u7ec3\u524d\u540e\u7684\u7f16\u8f91\u7ed3\u679c\u5bf9\u6bd4\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"426\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-627-1024x426.jpg\" alt=\"\" class=\"wp-image-1545\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-627-1024x426.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-627-300x125.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-627-768x320.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-627.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n","protected":false},"excerpt":{"rendered":"<p>\u7edf\u4e00\u591a\u6a21\u6001\u6a21\u578b\uff08Unified Multimodal Models, UMMs\uff09\u65e8\u5728\u5c06\u89c6\u89c9\u7406\u89e3\u548c\u751f\u6210\u7edf\u4e00\u4e8e\u5355\u4e00 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":1536,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[10],"tags":[],"class_list":["post-1535","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-10"],"_links":{"self":[{"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/posts\/1535","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/comments?post=1535"}],"version-history":[{"count":1,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/posts\/1535\/revisions"}],"predecessor-version":[{"id":1546,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/posts\/1535\/revisions\/1546"}],"wp:featuredmedia":[{"embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/media\/1536"}],"wp:attachment":[{"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/media?parent=1535"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/categories?post=1535"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/tags?post=1535"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}