{"id":40738,"date":"2025-08-04T11:37:25","date_gmt":"2025-08-04T03:37:25","guid":{"rendered":"https:\/\/www.1ai.net\/?p=40738"},"modified":"2025-08-04T11:37:25","modified_gmt":"2025-08-04T03:37:25","slug":"%e5%b0%8f%e7%b1%b3%e5%a3%b0%e9%9f%b3%e7%90%86%e8%a7%a3%e5%a4%a7%e6%a8%a1%e5%9e%8b-midashenglm-7b-%e5%8f%91%e5%b8%83%e5%b9%b6%e5%85%a8%e9%87%8f%e5%bc%80%e6%ba%90%ef%bc%8c22-%e4%b8%aa%e5%85%ac%e5%bc%80","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/40738.html","title":{"rendered":"Xiaomi's Sound Understanding Large Model MiDashengLM-7B Released and Open-Sourced in Full Volume, 22 Public Review Sets Refresh Best Scores"},"content":{"rendered":"<p>8 \u6708 4 \u65e5\u6d88\u606f\uff0c<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e5%b0%8f%e7%b1%b3\" title=\"\u3010\u67e5\u770b\u542b\u6709[\u5c0f\u7c73]\u6807\u7b7e\u7684\u6587\u7ae0\u3011\" target=\"_blank\" rel=\"noopener\">\u5c0f\u7c73<\/a>\u81ea\u7814\u58f0\u97f3\u7406\u89e3\u5927\u6a21\u578b <a href=\"https:\/\/www.1ai.net\/en\/tag\/midashenglm\" title=\"\u3010\u67e5\u770b\u542b\u6709[MiDashengLM]\u6807\u7b7e\u7684\u6587\u7ae0\u3011\" target=\"_blank\" rel=\"noopener\">MiDashengLM<\/a>-7B \u4eca\u65e5\u6b63\u5f0f\u53d1\u5e03\uff0c<strong>\u5e76\u5168\u91cf<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e5%bc%80%e6%ba%90\" title=\"\u3010\u67e5\u770b\u542b\u6709[\u5f00\u6e90]\u6807\u7b7e\u7684\u6587\u7ae0\u3011\" target=\"_blank\" rel=\"noopener\">\u5f00\u6e90<\/a><\/strong>\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-40739\" title=\"5e8981bbj00t0ga0y00aed000rs00fmp\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/08\/5e8981bbj00t0ga0y00aed000rs00fmp.jpg\" alt=\"5e8981bbj00t0ga0y00aed000rs00fmp\" width=\"1000\" height=\"562\" \/><\/p>\n<p>\u636e\u5c0f\u7c73\u5b98\u65b9\u4ecb\u7ecd\uff0cMiDashengLM-7B \u901f\u5ea6\u7cbe\u5ea6\u4e0a\u5b9e\u73b0\u53cc\u7a81\u7834\uff1a\u5355\u6837\u672c\u9996 Token \u5ef6\u8fdf\u4ec5\u4e3a\u540c\u7c7b\u6a21\u578b 1\/4\u3001\u540c\u663e\u5b58\u4e0b\u5e76\u53d1\u8d85 20 \u500d\uff0c<strong>\u5728 22 \u4e2a\u516c\u5f00\u8bc4\u6d4b\u96c6\u4e0a\u5237\u65b0\u591a\u6a21\u6001\u5927\u6a21\u578b\u6700\u597d\u6210\u7ee9\uff08SOTA\uff09<\/strong>\u3002<\/p>\n<p>MiDashengLM-7B \u57fa\u4e8e Xiaomi Dasheng \u4f5c\u4e3a\u97f3\u9891\u7f16\u7801\u5668\u548c Qwen2.5-Omni-7B Thinker \u4f5c\u4e3a\u81ea\u56de\u5f52\u89e3\u7801\u5668\uff0c\u901a\u8fc7\u521b\u65b0\u7684\u901a\u7528\u97f3\u9891\u63cf\u8ff0\u8bad\u7ec3\u7b56\u7565\uff0c\u5b9e\u73b0\u4e86\u5bf9\u8bed\u97f3\u3001\u73af\u5883\u58f0\u97f3\u548c\u97f3\u4e50\u7684\u7edf\u4e00\u7406\u89e3\u3002<\/p>\n<p>2024 \u5e74\uff0c\u5c0f\u7c73\u53d1\u5e03\u7684 Xiaomi Dasheng \u58f0\u97f3\u57fa\u5ea7\u6a21\u578b\u5728\u56fd\u9645\u4e0a\u9996\u6b21\u7a81\u7834 AudioSet 50+ mAP\uff0c\u5728 HEAR Benchmark \u73af\u5883\u58f0\u3001\u8bed\u97f3\u3001\u97f3\u4e50\u4e09\u5927\u9886\u57df\u5efa\u7acb\u9886\u5148\u4f18\u52bf\u5e76\u4fdd\u6301\u81f3\u4eca\u3002<\/p>\n<p><strong>Xiaomi Dasheng \u5728\u5c0f\u7c73\u7684\u667a\u80fd\u5bb6\u5c45\u548c\u6c7d\u8f66\u5ea7\u8231\u7b49\u573a\u666f\u6709\u8d85\u8fc7 30 \u9879\u843d\u5730\u5e94\u7528<\/strong>\u3002\u884c\u4e1a\u9996\u53d1\u7684\u8f66\u5916\u5524\u9192\u9632\u5fa1\u3001\u624b\u673a\u97f3\u7bb1\u5168\u5929\u5019\u76d1\u63a7\u5f02\u5e38\u58f0\u97f3\u3001\u201c\u6253\u4e2a\u54cd\u6307\u201d\u73af\u5883\u97f3\u5173\u8054 IoT \u63a7\u5236\u80fd\u529b\uff0c\u4ee5\u53ca\u5c0f\u7c73 YU7 \u4e0a\u642d\u8f7d\u7684\u589e\u5f3a\u54e8\u5175\u6a21\u5f0f\u5212\u8f66\u68c0\u6d4b\u7b49\uff0c\u80cc\u540e\u90fd\u6709 Xiaomi Dasheng \u4f5c\u4e3a\u6838\u5fc3\u7b97\u6cd5\u7684\u8d4b\u80fd\u3002<\/p>\n<p>MiDashengLM \u7684\u8bad\u7ec3\u6570\u636e\u7531 100% \u7684\u516c\u5f00\u6570\u636e\u6784\u6210\uff0c\u6a21\u578b\u4ee5\u5bbd\u677e\u7684 Apache License 2.0 \u53d1\u5e03\uff0c\u540c\u65f6\u652f\u6301\u5b66\u672f\u548c\u5546\u4e1a\u5e94\u7528\u3002<\/p>\n<p>\u5c0f\u7c73\u8868\u793a\uff0c\u4e0d\u540c\u4e8e Qwen2.5-Omni \u7b49\u672a\u516c\u5f00\u8bad\u7ec3\u6570\u636e\u7ec6\u8282\u7684\u6a21\u578b\uff0c<strong>MiDashengLM \u5b8c\u6574\u516c\u5f00\u4e86 77 \u4e2a\u6570\u636e\u6e90\u7684\u8be6\u7ec6\u914d\u6bd4<\/strong>\uff0c\u6280\u672f\u62a5\u544a\u4e2d\u8be6\u7ec6\u4ecb\u7ecd\u4e86\u4ece\u97f3\u9891\u7f16\u7801\u5668\u9884\u8bad\u7ec3\u5230\u6307\u4ee4\u5fae\u8c03\u7684\u5168\u6d41\u7a0b\u3002<\/p>\n<p>\u4f5c\u4e3a\u5c0f\u7c73\u201c\u4eba\u8f66\u5bb6\u5168\u751f\u6001\u201d\u6218\u7565\u7684\u5173\u952e\u6280\u672f\uff0cMiDashengLM \u901a\u8fc7\u7edf\u4e00\u7406\u89e3\u8bed\u97f3\u3001\u73af\u5883\u58f0\u4e0e\u97f3\u4e50\u7684\u8de8\u9886\u57df\u80fd\u529b\uff0c\u4e0d\u4ec5\u80fd\u542c\u61c2\u7528\u6237\u5468\u56f4\u53d1\u751f\u4e86\u4ec0\u4e48\u4e8b\u60c5\uff0c<strong>\u8fd8\u80fd\u5206\u6790\u53d1\u73b0\u8fd9\u4e9b\u4e8b\u60c5\u7684\u9690\u85cf\u542b\u4e49\uff0c\u63d0\u9ad8\u7528\u6237\u573a\u666f\u7406\u89e3\u7684\u6cdb\u5316\u6027<\/strong>\u3002<\/p>\n<p>\u57fa\u4e8e MiDashengLM \u7684\u6a21\u578b\u901a\u8fc7\u81ea\u7136\u8bed\u8a00\u548c\u7528\u6237\u4ea4\u4e92\uff0c\u4e3a\u7528\u6237\u63d0\u66f4\u4eba\u6027\u5316\u7684\u6c9f\u901a\u548c\u53cd\u9988\uff0c\u6bd4\u5982\u5728\u7528\u6237\u7ec3\u4e60\u5531\u6b4c\u6216\u7ec3\u4e60\u5916\u8bed\u65f6\u63d0\u4f9b\u53d1\u97f3\u53cd\u9988\u5e76\u5236\u5b9a\u9488\u5bf9\u6027\u63d0\u5347\u65b9\u6848\uff0c\u53c8\u6bd4\u5982\u5728\u7528\u6237\u9a7e\u9a76\u8f66\u8f86\u65f6\u5b9e\u65f6\u5bf9\u7528\u6237\u5173\u4e8e\u73af\u5883\u58f0\u97f3\u7684\u63d0\u95ee\u505a\u51fa\u89e3\u7b54\u3002<\/p>\n<p>MiDashengLM \u4ee5 Xiaomi Dasheng \u97f3\u9891\u7f16\u7801\u5668\u4e3a\u6838\u5fc3\u7ec4\u4ef6\uff0c\u662f Xiaomi Dasheng \u7cfb\u5217\u6a21\u578b\u7684\u91cd\u8981\u5347\u7ea7\u3002\u5728\u5f53\u524d\u7248\u672c\u7684\u57fa\u7840\u4e0a\uff0c\u5c0f\u7c73\u5df2\u7740\u624b\u5bf9\u8be5\u6a21\u578b\u505a\u8ba1\u7b97\u6548\u7387\u7684\u8fdb\u4e00\u6b65\u5347\u7ea7\uff0c<strong>\u5bfb\u6c42\u7ec8\u7aef\u8bbe\u5907\u4e0a\u53ef\u79bb\u7ebf\u90e8\u7f72\uff0c\u5e76\u5b8c\u5584\u57fa\u4e8e\u7528\u6237\u81ea\u7136\u8bed\u8a00\u63d0\u793a\u7684\u58f0\u97f3\u7f16\u8f91\u7b49\u66f4\u5168\u9762\u7684\u529f\u80fd<\/strong>\u3002<\/p>\n<p>1AI\u9644 MiDashengLM \u5f00\u6e90\u5730\u5740\uff1a<\/p>\n<ul>\n<li><strong>GitHub \u4e3b\u9875\uff1a<\/strong>https:\/\/github.com\/xiaomi-research\/dasheng-lm<\/li>\n<li><strong>\u6280\u672f\u62a5\u544a<\/strong>\uff1ahttps:\/\/github.com\/xiaomi-research\/dasheng-lm\/tree\/main\/technical_report<\/li>\n<li><strong>\u6a21\u578b\u53c2\u6570\uff08Hugging Face\uff09\uff1a<\/strong>https:\/\/huggingface.co\/mispeech\/midashenglm-7b<\/li>\n<li><strong>\u6a21\u578b\u53c2\u6570\uff08\u9b54\u642d\u793e\u533a\uff09\uff1a<\/strong>https:\/\/modelscope.cn\/models\/midasheng\/midashenglm-7b<\/li>\n<li><strong>\u7f51\u9875<\/strong><strong>\u00a0<\/strong><strong>Demo\uff1a<\/strong><strong>\u00a0<\/strong>https:\/\/xiaomi-research.github.io\/dasheng-lm<\/li>\n<li><strong>\u4ea4\u4e92<\/strong><strong>\u00a0<\/strong><strong>Demo\uff1a<\/strong>https:\/\/huggingface.co\/spaces\/mispeech\/MiDashengLM<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>August 4 news, millet self-research sound understanding large model MiDashengLM-7B today officially released, and full open source. According to Xiaomi's official introduction, MiDashengLM-7B achieves double breakthroughs in speed and accuracy: the delay of the first Token of a single sample is only 1\/4 of similar models, and the concurrency is more than 20 times under the same memory, and it refreshes the best results of multimodal large models in 22 public evaluation sets (SOTA). Based on Xiaomi Dasheng as the audio encoder and Qwen2.5-Omni-7B Thinker as the autoregressive decoder, MiDashengLM-7B achieves the best results in speech, ambient sound, and music through an innovative universal audio description training strategy.<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[146],"tags":[7335,1114,219],"collection":[],"class_list":["post-40738","post","type-post","status-publish","format-standard","hentry","category-news","tag-midashenglm","tag-1114","tag-219"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/40738","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=40738"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/40738\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=40738"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=40738"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=40738"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=40738"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}