{"id":8610,"date":"2026-05-30T08:37:14","date_gmt":"2026-05-30T00:37:14","guid":{"rendered":"https:\/\/infernews.com\/blog\/implementation-of-quot-collectionlora-collecting-50-effects-in-1-lora-via-multi\/"},"modified":"2026-05-30T08:37:14","modified_gmt":"2026-05-30T00:37:14","slug":"implementation-of-quot-collectionlora-collecting-50-effects-in-1-lora-via-multi","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/implementation-of-quot-collectionlora-collecting-50-effects-in-1-lora-via-multi\/","title":{"rendered":"CollectionLoRA\uff1a\u628a\u591a\u500b\u6548\u679c LoRA \u6fc3\u7e2e\u6210\u4e00\u500b"},"content":{"rendered":"
\"Repository<\/figure>\n

CollectionLoRA \u662f\u4e00\u500b\u91dd\u5c0d LoRA \u7ba1\u7406\u6210\u672c\u800c\u8a2d\u7684\u9805\u76ee\uff0c\u6838\u5fc3\u505a\u6cd5\u662f\u7528 multi-teacher on-policy distillation\uff0c\u5c07\u591a\u500b\u6548\u679c LoRAs \u7684\u6982\u5ff5\uff0c\u4ee5\u53ca few-step generation \u80fd\u529b\uff0c\u4e00\u6b21\u904e\u84b8\u993e\u9032\u55ae\u4e00 LoRA\u3002\u5c0d\u4e00\u822c\u4f7f\u7528\u8005\u4f86\u8aaa\uff0c\u610f\u601d\u5c31\u662f\u539f\u672c\u8981\u70ba\u4e0d\u540c\u6548\u679c\u5207\u63db\u6216\u4e32\u63a5\u591a\u500b adapter\uff0c\u73fe\u5728\u6709\u6a5f\u6703\u6539\u7528\u4e00\u500b\u6574\u5408\u7248\u672c\u8655\u7406\uff0c\u6d41\u7a0b\u6703\u66f4\u7c21\u6f54\u3002<\/p>\n

\u4f7f\u7528\u9019\u500b\u9805\u76ee\u6642\uff0c\u91cd\u9ede\u4e0d\u662f\u7531\u96f6\u958b\u59cb\u8a13\u7df4\uff0c\u800c\u662f\u5148\u6309\u9805\u76ee\u63d0\u4f9b\u7684 ckpt\/ \u7d50\u69cb\u653e\u597d\u6b0a\u91cd\uff0c\u4e4b\u5f8c\u4ee5 50_in_1\/ \u4f5c\u63a8\u7406\u7528\u9014\u3002\u5009\u5eab\u4ea6\u63d0\u4f9b\u516c\u958b\u7684 training and inference code\uff0c\u800c data\/manga_tone\/ \u5167\u6709\u8d77\u6b65\u7528\u7684 teacher LoRA \u548c\u793a\u7bc4\u8a13\u7df4\u8cc7\u7522\uff0c\u65b9\u4fbf\u4e86\u89e3\u6574\u500b\u8cc7\u6599\u8207\u6a21\u578b\u914d\u7f6e\u65b9\u5f0f\u3002<\/p>\n

\u5b83\u89e3\u6c7a\u7684\u554f\u984c\u76f8\u7576\u660e\u78ba\uff1a\u7576\u6548\u679c LoRA \u6578\u91cf\u6108\u4f86\u6108\u591a\uff0c\u90e8\u7f72\u3001\u5207\u63db\u8207\u758a\u52a0\u90fd\u6703\u8b8a\u5f97\u9ebb\u7169\uff0c\u914d\u5408 acceleration modules \u6642\uff0c\u9084\u53ef\u80fd\u51fa\u73fe\u4e92\u76f8\u5e72\u64fe\u3002CollectionLoRA \u5617\u8a66\u628a\u300c\u591a\u6548\u679c\u300d\u8207\u300c\u5c11\u6b65\u6578\u751f\u6210\u300d\u5408\u4f75\u5230\u540c\u4e00\u500b LoRA\uff0c\u9019\u6bd4\u55ae\u7d14\u6536\u85cf\u5927\u91cf LoRA \u66f4\u50cf\u662f\u91cd\u65b0\u6574\u7406\u6574\u500b\u5de5\u4f5c\u6d41\u3002<\/p>\n

\u9805\u76ee\u516c\u958b\u7684\u91cd\u9ede\u5305\u62ec\u5e7e\u500b\u65b9\u5411\uff1a
\n– \u53ef\u628a 50\u21921<\/strong>\uff0c\u751a\u81f3 180\u21921<\/strong> \u7684\u6548\u679c\u6559\u5e2b\u6574\u5408\u5230\u55ae\u4e00 LoRA
\n– \u5728 EffectBench<\/strong> \u4e0a\uff0c\u6587\u4e2d\u7a31\u5176\u65bc 8 NFE<\/strong> \u4e0b\u53ef\u53d6\u5f97\u8f03\u9ad8 VSA<\/strong> \u8207\u8f03\u4f4e BCR<\/strong>
\n– \u652f\u63f4 zero-shot \u7684 A \u2295 B<\/strong> \u7d44\u5408\u6548\u679c\uff0c\u63a8\u7406\u6642\u53ef\u914d\u5c0d\u5169\u500b\u5df2\u8a13\u7df4\u6559\u5e2b\uff0c\u7121\u9700\u984d\u5916\u8a13\u7df4
\n– \u5df2\u958b\u653e training and inference code\uff0c\u4f46 model weights \u4ecd\u6a19\u793a\u70ba\u672a\u5168\u9762\u958b\u653e<\/p>\n

\u9019\u500b\u9805\u76ee\u8f03\u9069\u5408\u6703\u63a5\u89f8\u751f\u6210\u6548\u679c\u63a7\u5236\u3001\u9700\u8981\u7ba1\u7406\u5927\u91cf LoRA \u7684\u7814\u7a76\u8005\u8207\u958b\u767c\u8005\uff0c\u4e5f\u9069\u5408\u60f3\u6e1b\u5c11\u90e8\u7f72\u8907\u96dc\u5ea6\u7684\u5718\u968a\u3002\u82e5\u4f60\u95dc\u5fc3\u7684\u662f\u628a\u591a\u7a2e\u98a8\u683c\u6216\u6548\u679c\u6574\u5408\u6210\u8f03\u6613\u5206\u767c\u7684\u6a21\u578b\u5f62\u5f0f\uff0cCollectionLoRA \u5c55\u793a\u7684\u65b9\u5411\u5f88\u6709\u53c3\u8003\u50f9\u503c\uff1b\u4e0d\u904e\u73fe\u968e\u6bb5\u516c\u958b\u8cc7\u8a0a\u4e3b\u8981\u96c6\u4e2d\u5728\u6846\u67b6\u3001\u6307\u6a19\u8207\u793a\u7bc4\u8cc7\u7522\uff0c\u5b8c\u6574\u6b0a\u91cd\u4f9b\u61c9\u60c5\u6cc1\u4ecd\u8981\u7559\u610f\u9805\u76ee\u5f8c\u7e8c\u66f4\u65b0\u3002<\/p>\n

GitHub\uff1a<\/strong> https:\/\/github.com\/Qwen-Applications\/CollectionLoRA<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

\u628a 50 \u751a\u81f3 180 \u500b\u6548\u679c LoRA \u96c6\u4e2d\u5230\u55ae\u4e00 LoRA\uff0c\u6e1b\u5c11\u90e8\u7f72\u6210\u672c\uff0c\u4e5f\u964d\u4f4e\u591a\u91cd adapter \u758a\u52a0\u5e36\u4f86\u7684\u5e72\u64fe\u3002<\/p>\n","protected":false},"author":8,"featured_media":8609,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","wpai_meta_description":"","footnotes":""},"categories":[133,171,157,120,149,197],"tags":[],"class_list":["post-8610","post","type-post","status-publish","format-standard","hentry","category-133","category-171","category-157","category-120","category-149","category-framework"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8610","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=8610"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8610\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/8609"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=8610"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=8610"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=8610"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}