{"id":4888,"date":"2024-10-17T10:35:37","date_gmt":"2024-10-17T02:35:37","guid":{"rendered":"https:\/\/nullthought.net\/?p=4888"},"modified":"2025-10-07T15:46:19","modified_gmt":"2025-10-07T07:46:19","slug":"sana%ef%bc%9a%e4%b8%80%e7%a7%8d%e7%94%a8%e4%ba%8e%e7%94%9f%e6%88%90%e9%ab%98%e5%88%86%e8%be%a8%e7%8e%87%ef%bc%88%e6%9c%80%e9%ab%98%e5%8f%af%e8%be%be4096x4096%ef%bc%89%e7%9a%84%e6%96%87%e6%9c%ac","status":"publish","type":"post","link":"https:\/\/nullthought.net\/?p=4888","title":{"rendered":"SANA\uff1a\u4e00\u79cd\u7528\u4e8e\u751f\u6210\u9ad8\u5206\u8fa8\u7387\uff08\u6700\u9ad8\u53ef\u8fbe4096\u00d74096\uff09\u7684\u6587\u672c\u5230\u56fe\u50cf\u751f\u6210\u6846\u67b6"},"content":{"rendered":"\n<p>\u8bba\u6587<strong><a href=\"https:\/\/arxiv.org\/abs\/2410.10629\" target=\"_blank\" rel=\"noreferrer noopener\">SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformers<\/a><\/strong>\u8be6\u7ec6\u4ecb\u7ecd\u4e86Sana\u6846\u67b6\uff0c\u5b83\u662f\u4e00\u79cd\u7528\u4e8e\u751f\u6210\u9ad8\u5206\u8fa8\u7387\uff08\u6700\u9ad8\u53ef\u8fbe4096\u00d74096\uff09\u7684\u6587\u672c\u5230\u56fe\u50cf\u751f\u6210\u7cfb\u7edf\u3002\u8be5\u7cfb\u7edf\u65e8\u5728\u63d0\u9ad8\u751f\u6210\u901f\u5ea6\u548c\u6548\u7387\uff0c\u5c24\u5176\u662f\u5728\u9ad8\u5206\u8fa8\u7387\u751f\u6210\u4efb\u52a1\u4e0a\uff0cSana\u7684\u6027\u80fd\u8d85\u8d8a\u4e86\u73b0\u6709\u5927\u591a\u6570\u65b9\u6cd5\u3002<\/p>\n\n\n\n<p><strong><a href=\"https:\/\/nvlabs.github.io\/Sana\/\" target=\"_blank\" rel=\"noreferrer noopener\">Sana\u6846\u67b6<\/a><\/strong>\u901a\u8fc7<strong>\u6df1\u5ea6\u538b\u7f29\u81ea\u7f16\u7801\u5668\u3001\u7ebf\u6027\u6ce8\u610f\u529b\u6269\u6563\u53d8\u6362\u5668\u3001\u89e3\u7801\u5668\u9a71\u52a8\u7684\u5c0f\u578bLLM\u6587\u672c\u7f16\u7801\u5668<\/strong>\u7b49\u591a\u9879\u521b\u65b0\u8bbe\u8ba1\uff0c\u6210\u529f\u5b9e\u73b0\u4e86\u9ad8\u6548\u7684\u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\u751f\u6210\u3002\u76f8\u6bd4\u4e8e\u73b0\u6709\u7684\u5927\u578b\u6a21\u578b\uff0cSana\u4e0d\u4ec5\u5728\u901f\u5ea6\u548c\u6548\u7387\u4e0a\u5b9e\u73b0\u4e86\u663e\u8457\u63d0\u5347\uff0c\u800c\u4e14\u80fd\u591f\u5728\u666e\u901a\u786c\u4ef6\u4e0a\u8fd0\u884c\uff0c\u6781\u5927\u5730\u964d\u4f4e\u4e86\u6587\u672c\u5230\u56fe\u50cf\u751f\u6210\u6a21\u578b\u7684\u4f7f\u7528\u95e8\u69db\u3002\u901a\u8fc7\u8fd9\u4e00\u7cfb\u5217\u7684\u4f18\u5316\u548c\u521b\u65b0\uff0cSana\u4e3a\u672a\u6765\u7684\u9ad8\u5206\u8fa8\u7387\u5185\u5bb9\u751f\u6210\u63d0\u4f9b\u4e86\u4e00\u4e2a\u9ad8\u6548\u3001\u4f4e\u6210\u672c\u7684\u89e3\u51b3\u65b9\u6848\u3002<\/p>\n\n\n\n<p>\u8bba\u6587\u4f5c\u8005\u4e3aEnze Xie, Junsong Chen, Junyu Chen, Han Cai, Haotian Tang, Yujun Lin, Zhekai Zhang, Muyang Li, Ligeng Zhu, Yao Lu, Song Han\uff0c\u6765\u81eaNVIDIA\uff08\u82f1\u4f1f\u8fbe\uff09\uff0cMIT\uff08\u9ebb\u7701\u7406\u5de5\uff09\u548cTsinghua University\uff08\u6e05\u534e\u5927\u5b66\uff09\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"665\" height=\"461\" src=\"https:\/\/nullthought.net\/wp-content\/uploads\/2024\/10\/image-11.png\" alt=\"\" class=\"wp-image-4889\" srcset=\"https:\/\/nullthought.net\/wp-content\/uploads\/2024\/10\/image-11.png 665w, https:\/\/nullthought.net\/wp-content\/uploads\/2024\/10\/image-11-300x208.png 300w\" sizes=\"auto, (max-width: 665px) 100vw, 665px\" \/><figcaption class=\"wp-element-caption\"><strong><a href=\"https:\/\/arxiv.org\/abs\/2410.10629\" target=\"_blank\" rel=\"noreferrer noopener\">SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformers<\/a><\/strong><\/figcaption><\/figure>\n\n\n\n<p>\u4ee5\u4e0b\u662f\u8bba\u6587\u7684\u8be6\u7ec6\u5206\u6790\u4ecb\u7ecd\uff1a<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">1. \u7814\u7a76\u80cc\u666f<\/h4>\n\n\n\n<p>\u968f\u7740\u6269\u6563\u6a21\u578b\u5728\u6587\u672c\u5230\u56fe\u50cf\u8f6c\u6362\u9886\u57df\u7684\u8fdb\u6b65\uff0c\u9ad8\u5206\u8fa8\u7387\u7684\u56fe\u50cf\u751f\u6210\u6210\u4e3a\u4e86\u4e00\u4e2a\u65b0\u7684\u6280\u672f\u6311\u6218\u3002\u5f53\u524d\uff0c\u8bb8\u591a\u73b0\u6709\u7684\u6a21\u578b\uff0c\u6bd4\u5982PixArt\u3001Stable Diffusion 3\uff08SD3\uff09\u3001Flux\u7b49\uff0c\u91c7\u7528\u7684\u53c2\u6570\u91cf\u5de8\u5927\uff0c\u901a\u5e38\u57288B\u523024B\u4e0d\u7b49\uff0c\u8fd9\u4f7f\u5f97\u5b83\u4eec\u7684\u8bad\u7ec3\u548c\u63a8\u7406\u6210\u672c\u975e\u5e38\u9ad8\uff0c\u96be\u4ee5\u666e\u53ca\u4f7f\u7528\u3002\u56e0\u6b64\uff0cSana\u7684\u7814\u7a76\u76ee\u6807\u662f\u5f00\u53d1\u4e00\u79cd\u8f7b\u91cf\u5316\u3001\u9ad8\u6548\u7684\u6587\u672c\u5230\u56fe\u50cf\u751f\u6210\u6a21\u578b\uff0c\u80fd\u591f\u5728\u4fdd\u6301\u9ad8\u8d28\u91cf\u548c\u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\u751f\u6210\u7684\u540c\u65f6\u663e\u8457\u964d\u4f4e\u8ba1\u7b97\u5f00\u9500\uff0c\u5e76\u4e14\u80fd\u591f\u5728\u666e\u901a\u6d88\u8d39\u7ea7\u786c\u4ef6\u4e0a\u8fd0\u884c\u3002<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">2. Sana\u67b6\u6784\u8bbe\u8ba1<\/h4>\n\n\n\n<p>Sana\u7684\u67b6\u6784\u5305\u542b\u51e0\u4e2a\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">2.1 \u6df1\u5ea6\u538b\u7f29\u81ea\u7f16\u7801\u5668\uff08Deep Compression Autoencoder\uff09<\/h5>\n\n\n\n<p>\u4f20\u7edf\u7684\u81ea\u7f16\u7801\u5668\uff08AE\uff09\u901a\u5e38\u53ea\u5c06\u56fe\u50cf\u538b\u7f298\u500d\uff08\u4f8b\u5982\uff0cAE-F8\uff09\uff0c\u4f46Sana\u5f15\u5165\u4e86\u4e00\u4e2a<strong>\u538b\u7f29\u7387\u9ad8\u8fbe32\u500d\u7684\u81ea\u7f16\u7801\u5668\uff08AE-F32\uff09<\/strong>\u3002\u901a\u8fc7\u66f4\u5927\u7684\u538b\u7f29\u7387\uff0cSana\u51cf\u5c11\u4e86\u6f5c\u5728\u6807\u8bb0\u7684\u6570\u91cf\uff0c\u4ece\u800c\u5927\u5e45\u964d\u4f4e\u4e86\u8bad\u7ec3\u548c\u751f\u6210\u7684\u590d\u6742\u5ea6\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u8bbe\u8ba1\u52a8\u673a<\/strong>\uff1a\u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\u5305\u542b\u5927\u91cf\u7684\u5197\u4f59\u4fe1\u606f\uff0c\u56e0\u6b64\u5728\u8bad\u7ec3\u548c\u63a8\u7406\u8fc7\u7a0b\u4e2d\uff0c\u8fc7\u591a\u7684\u50cf\u7d20\u4fe1\u606f\u5904\u7406\u4f1a\u9020\u6210\u6027\u80fd\u74f6\u9888\u3002\u901a\u8fc7\u66f4\u9ad8\u7684\u538b\u7f29\u6bd4\u4f8b\uff0cSana\u7684\u81ea\u7f16\u7801\u5668\u5728\u4fdd\u8bc1\u56fe\u50cf\u8d28\u91cf\u7684\u540c\u65f6\u51cf\u5c11\u4e86\u5904\u7406\u7684\u6807\u8bb0\u6570\u91cf\uff0c\u8fdb\u800c\u63d0\u9ad8\u4e86\u8bad\u7ec3\u548c\u63a8\u7406\u7684\u6548\u7387\u3002<\/li>\n\n\n\n<li><strong>\u5177\u4f53\u5b9e\u73b0<\/strong>\uff1aSana\u7684AE\u4f7f\u752832\u500d\u538b\u7f29\uff0c\u5e76\u5c06\u56fe\u50cf\u7684\u6bcf\u4e2a\u50cf\u7d20\u5206\u89e3\u4e3a32\u4e2a\u6f5c\u5728\u901a\u9053\uff08C=32\uff09\uff0c\u540c\u65f6\u964d\u4f4e\u4e86\u8865\u4e01\u5927\u5c0f\uff08Patch Size\uff09P=1\u3002\u8fd9\u79cd\u8bbe\u8ba1\u4fdd\u8bc1\u4e86\u5728\u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\u751f\u6210\u4e2d\u66f4\u6709\u6548\u7684\u538b\u7f29\u548c\u5904\u7406\u80fd\u529b\u3002<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">2.2 \u7ebf\u6027\u6269\u6563\u53d8\u6362\u5668\uff08Linear <a href=\"https:\/\/nullthought.net\/?p=6239\" target=\"_blank\" rel=\"noreferrer noopener\">Diffusion Transformers, DiT<\/a>\uff09<\/h5>\n\n\n\n<p>Sana\u7684\u53e6\u4e00\u4e2a\u5173\u952e\u7ec4\u4ef6\u662f\u7ebf\u6027\u6269\u6563\u53d8\u6362\u5668\uff0c\u5b83\u901a\u8fc7\u66ff\u6362\u4f20\u7edf\u7684\u4e8c\u6b21\u81ea\u6ce8\u610f\u529b\u673a\u5236\uff08Vanilla Self-Attention\uff09\u4e3a\u7ebf\u6027\u6ce8\u610f\u529b\uff0c\u5927\u5e45\u5ea6\u51cf\u5c11\u4e86\u8ba1\u7b97\u590d\u6742\u5ea6\u3002\u4f20\u7edf\u7684\u6ce8\u610f\u529b\u673a\u5236\u5177\u6709O(N\u00b2)\u7684\u590d\u6742\u5ea6\uff0c\u800c\u7ebf\u6027\u6ce8\u610f\u529b\u5c06\u5176\u964d\u4f4e\u5230O(N)\uff0c\u8fd9\u5bf9\u4e8e\u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\u751f\u6210\u6765\u8bf4\u662f\u4e00\u4e2a\u5de8\u5927\u7684\u6027\u80fd\u63d0\u5347\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Mix-FFN\u6a21\u5757<\/strong>\uff1aSana\u8fd8\u5f15\u5165\u4e86\u4e00\u4e2aMix-FFN\uff08\u6df7\u5408\u524d\u9988\u7f51\u7edc\uff09\u6a21\u5757\uff0c\u7ed3\u5408\u4e86\u6df1\u5ea6\u5377\u79ef\u5c42\uff083\u00d73\u5377\u79ef\uff09\uff0c\u4f7f\u5f97\u6a21\u578b\u53ef\u4ee5\u66f4\u597d\u5730\u6355\u83b7\u5c40\u90e8\u4fe1\u606f\u3002\u8be5\u8bbe\u8ba1\u8fdb\u4e00\u6b65\u63d0\u5347\u4e86\u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\u7684\u751f\u6210\u80fd\u529b\uff0c\u7279\u522b\u662f\u5728\u6ca1\u6709\u4f4d\u7f6e\u7f16\u7801\uff08NoPE\uff09\u7684\u60c5\u51b5\u4e0b\uff0c\u4ecd\u7136\u4fdd\u6301\u4e86\u51fa\u8272\u7684\u6027\u80fd\u8868\u73b0\u3002<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">2.3 \u89e3\u7801\u5668\u9a71\u52a8\u7684\u5c0f\u578bLLM\u4f5c\u4e3a\u6587\u672c\u7f16\u7801\u5668<\/h5>\n\n\n\n<p>\u4f20\u7edf\u7684\u6587\u672c\u5230\u56fe\u50cf\u6a21\u578b\u901a\u5e38\u4f7f\u7528\u7684\u662fCLIP\u6216T5\u6a21\u578b\u4f5c\u4e3a\u6587\u672c\u7f16\u7801\u5668\uff0c\u5b83\u4eec\u5728\u7406\u89e3\u6587\u672c\u548c\u56fe\u50cf\u5bf9\u9f50\u65b9\u9762\u5b58\u5728\u5c40\u9650\u6027\u3002Sana\u5219\u91c7\u7528\u4e86\u4e00\u4e2a\u57fa\u4e8e\u89e3\u7801\u5668\u7684\u5c0f\u578b\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\uff0c\u5982Gemma-2\uff0c\u6765\u63d0\u5347\u6587\u672c\u7406\u89e3\u548c\u63a8\u7406\u80fd\u529b\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u8bbe\u8ba1\u7406\u7531<\/strong>\uff1a\u4e0eT5\u76f8\u6bd4\uff0cGemma\u7b49\u89e3\u7801\u5668\u67b6\u6784\u7684LLM\u5177\u5907\u66f4\u5f3a\u7684\u6307\u4ee4\u8ddf\u968f\u80fd\u529b\u548c\u63a8\u7406\u80fd\u529b\uff0c\u7279\u522b\u662f\u5728\u590d\u6742\u7684\u7528\u6237\u6307\u4ee4\u573a\u666f\u4e2d\u3002\u901a\u8fc7\u8fd9\u79cd\u89e3\u7801\u5668\u9a71\u52a8\u7684\u8bbe\u8ba1\uff0cSana\u80fd\u591f\u66f4\u597d\u5730\u7406\u89e3\u7528\u6237\u63d0\u4f9b\u7684\u6587\u672c\u63cf\u8ff0\uff0c\u5e76\u751f\u6210\u9ad8\u8d28\u91cf\u7684\u56fe\u50cf\u3002<\/li>\n\n\n\n<li><strong>\u590d\u6742\u4eba\u5de5\u6307\u4ee4\uff08Complex Human Instruction, CHI\uff09<\/strong>\uff1aSana\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u5f15\u5165\u4e86\u590d\u6742\u7684\u4eba\u5de5\u6307\u4ee4\uff0c\u501f\u52a9\u4e8eLLM\u7684<a href=\"https:\/\/nullthought.net\/?p=5522\" target=\"_blank\" rel=\"noreferrer noopener\">\u4e0a\u4e0b\u6587\u5b66\u4e60\u80fd\u529b\uff08In-context Learning\uff09<\/a>\uff0c\u5f3a\u5316\u4e86\u6587\u672c\u548c\u56fe\u50cf\u7684\u5bf9\u9f50\u6548\u679c\u3002\u5b9e\u9a8c\u8868\u660e\uff0c\u52a0\u5165CHI\u540e\uff0c\u751f\u6210\u7684\u56fe\u50cf\u8d28\u91cf\u5f97\u5230\u4e86\u663e\u8457\u63d0\u5347\u3002<\/li>\n<\/ul>\n\n\n\n<h4 class=\"wp-block-heading\">3. \u9ad8\u6548\u7684\u8bad\u7ec3\u548c\u63a8\u7406\u7b56\u7565<\/h4>\n\n\n\n<p>Sana\u4e0d\u4ec5\u5728\u6a21\u578b\u67b6\u6784\u4e0a\u8fdb\u884c\u4e86\u521b\u65b0\uff0c\u8fd8\u5728\u8bad\u7ec3\u548c\u63a8\u7406\u8fc7\u7a0b\u4e2d\u91c7\u7528\u4e86\u591a\u9879\u4f18\u5316\u7b56\u7565\uff0c\u4ee5\u63d0\u5347\u751f\u6210\u901f\u5ea6\u548c\u56fe\u50cf\u8d28\u91cf\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">3.1 \u81ea\u52a8\u5316\u7684\u6807\u7b7e\u751f\u6210\u4e0e\u9009\u62e9<\/h5>\n\n\n\n<p>\u5728\u6570\u636e\u96c6\u7684\u6807\u6ce8\u65b9\u9762\uff0cSana\u91c7\u7528\u4e86\u81ea\u52a8\u5316\u7684\u591a\u91cd\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\u6807\u7b7e\u751f\u6210\u7b56\u7565\u3002\u6bcf\u5f20\u56fe\u7247\u90fd\u4f1a\u901a\u8fc7\u591a\u4e2aVLM\u6a21\u578b\u751f\u6210\u4e0d\u540c\u7684\u63cf\u8ff0\uff0c\u8fdb\u4e00\u6b65\u63d0\u5347\u4e86\u6807\u7b7e\u7684\u591a\u6837\u6027\u548c\u51c6\u786e\u6027\u3002\u7136\u540e\uff0cSana\u901a\u8fc7\u57fa\u4e8eCLIP\u5f97\u5206\u7684\u91c7\u6837\u7b56\u7565\uff0c\u52a8\u6001\u9009\u62e9\u9ad8\u8d28\u91cf\u7684\u6807\u7b7e\u8fdb\u884c\u8bad\u7ec3\uff0c\u4ece\u800c\u63d0\u5347\u4e86\u8bad\u7ec3\u6536\u655b\u901f\u5ea6\u548c\u56fe\u50cf-\u6587\u672c\u7684\u5bf9\u9f50\u5ea6\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">3.2 Flow-DPM-Solver\u63a8\u7406\u52a0\u901f<\/h5>\n\n\n\n<p>\u4e3a\u4e86\u8fdb\u4e00\u6b65\u52a0\u5feb\u751f\u6210\u901f\u5ea6\uff0cSana\u91c7\u7528\u4e86Flow-DPM-Solver\uff0c\u8be5\u7b97\u6cd5\u51cf\u5c11\u4e86\u751f\u6210\u6b65\u9aa4\uff0c\u5c06\u5e38\u89c4\u768428-50\u6b65\u63a8\u7406\u51cf\u5c11\u523014-20\u6b65\u3002\u540c\u65f6\uff0cFlow-DPM-Solver\u5f15\u5165\u4e86\u65b0\u7684\u63a8\u7406\u65b9\u5f0f\uff0c\u4ece\u901f\u5ea6\u548c\u8d28\u91cf\u4e0a\u90fd\u8d85\u8fc7\u4e86Flow-Euler-Solver\uff0c\u4f7f\u5f97Sana\u5728\u63a8\u7406\u65f6\u80fd\u591f\u66f4\u5feb\u5730\u751f\u6210\u9ad8\u8d28\u91cf\u56fe\u50cf\u3002<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">4. \u6027\u80fd\u8bc4\u4f30\u4e0e\u5b9e\u9a8c\u7ed3\u679c<\/h4>\n\n\n\n<p>Sana\u5728\u591a\u4e2a\u8bc4\u6d4b\u6307\u6807\u4e0a\u8868\u73b0\u4f18\u5f02\uff0c\u7279\u522b\u662f\u5728\u751f\u6210\u901f\u5ea6\u3001\u8ba1\u7b97\u6548\u7387\u4ee5\u53ca\u56fe\u50cf\u8d28\u91cf\u4e0a\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u4e0e\u73b0\u6709\u6a21\u578b\u7684\u5bf9\u6bd4<\/strong>\uff1a\u4e0eFlux\u3001SD3\u3001PixArt-\u03a3\u7b49\u6700\u5148\u8fdb\u7684\u6269\u6563\u6a21\u578b\u76f8\u6bd4\uff0cSana\u5728\u751f\u6210\u901f\u5ea6\u4e0a\u5b9e\u73b0\u4e86100\u500d\u7684\u52a0\u901f\uff0c\u4e14\u5728\u56fe\u50cf\u8d28\u91cf\u4e0a\u4fdd\u6301\u4e86\u7ade\u4e89\u529b\u3002\u7279\u522b\u662f\u57281K\u548c4K\u5206\u8fa8\u7387\u4e0b\uff0cSana\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u751f\u6210\u901f\u5ea6\u663e\u8457\u4f18\u4e8e\u540c\u7c7b\u6a21\u578b\u3002<\/li>\n\n\n\n<li><strong>\u591a\u5c3a\u5ea6\u7684\u56fe\u50cf\u751f\u6210<\/strong>\uff1aSana\u80fd\u591f\u751f\u6210\u4ece1024\u00d71024\u52304096\u00d74096\u5206\u8fa8\u7387\u7684\u56fe\u50cf\uff0c\u5e76\u4e14\u5728\u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\u751f\u6210\u4efb\u52a1\u4e2d\uff0cSana\u7684\u5ef6\u8fdf\u6bd4\u5f53\u524d\u6700\u5148\u8fdb\u7684\u6a21\u578b\u51cf\u5c11\u4e86106\u500d\u3002<\/li>\n\n\n\n<li><strong>\u5c0f\u578b\u6a21\u578b\u90e8\u7f72<\/strong>\uff1aSana-0.6B\u6a21\u578b\u53ef\u4ee5\u572816GB\u7684\u6d88\u8d39\u7ea7GPU\u4e0a\u90e8\u7f72\uff0c\u4e14\u751f\u62101024\u00d71024\u5206\u8fa8\u7387\u56fe\u50cf\u7684\u65f6\u95f4\u4e0d\u52301\u79d2\uff0c\u5c55\u793a\u4e86\u5176\u5728\u4f4e\u6210\u672c\u8bbe\u5907\u4e0a\u7684\u5f3a\u5927\u6027\u80fd\u3002<\/li>\n<\/ul>\n\n\n\n<h4 class=\"wp-block-heading\">5. \u672a\u6765\u53d1\u5c55\u65b9\u5411<\/h4>\n\n\n\n<p>\u8bba\u6587\u7684\u7ed3\u8bba\u90e8\u5206\u63d0\u51fa\u4e86Sana\u672a\u6765\u7684\u51e0\u4e2a\u6f5c\u5728\u53d1\u5c55\u65b9\u5411\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u89c6\u9891\u751f\u6210<\/strong>\uff1aSana\u672a\u6765\u53ef\u80fd\u4f1a\u8fdb\u4e00\u6b65\u6269\u5c55\u81f3\u89c6\u9891\u751f\u6210\u9886\u57df\uff0c\u4ee5\u5e94\u5bf9\u65e5\u76ca\u589e\u957f\u7684\u9ad8\u5206\u8fa8\u7387\u89c6\u9891\u751f\u6210\u9700\u6c42\u3002<\/li>\n\n\n\n<li><strong>\u751f\u6210\u5b89\u5168\u6027\u4e0e\u53ef\u63a7\u6027<\/strong>\uff1a\u5c3d\u7ba1Sana\u5728\u6027\u80fd\u4e0a\u8868\u73b0\u4f18\u5f02\uff0c\u4f46\u5728\u751f\u6210\u56fe\u50cf\u7684\u5b89\u5168\u6027\u548c\u53ef\u63a7\u6027\u65b9\u9762\u4ecd\u6709\u5f85\u6539\u8fdb\uff0c\u7279\u522b\u662f\u5728\u751f\u6210\u590d\u6742\u5185\u5bb9\uff08\u5982\u9762\u90e8\u548c\u624b\u90e8\uff09\u65f6\uff0c\u5b58\u5728\u4e00\u5b9a\u7684\u6311\u6218\u3002<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p>SANA\uff1a<a href=\"https:\/\/nvlabs.github.io\/Sana\/\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/nvlabs.github.io\/Sana\/<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u8bba\u6587SANA: Efficient High-Resolution Image Synthesis with  [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[8],"tags":[39,84,95,86,80],"class_list":["post-4888","post","type-post","status-publish","format-standard","hentry","category-tech","tag-ai","tag-nvidia","tag-transformer","tag-llm","tag-cv"],"rttpg_featured_image_url":null,"rttpg_author":{"display_name":"NullThought","author_link":"https:\/\/nullthought.net\/?author=1"},"rttpg_comment":0,"rttpg_category":"<a href=\"https:\/\/nullthought.net\/?cat=8\" rel=\"category\">Tech<\/a>","rttpg_excerpt":"\u8bba\u6587SANA: Efficient High-Resolution Image Synthesis with &hellip;","_links":{"self":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4888","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=4888"}],"version-history":[{"count":5,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4888\/revisions"}],"predecessor-version":[{"id":6242,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4888\/revisions\/6242"}],"wp:attachment":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4888"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4888"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4888"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}