{"id":35791,"date":"2025-05-21T19:45:42","date_gmt":"2025-05-21T11:45:42","guid":{"rendered":"https:\/\/www.1ai.net\/?p=35791"},"modified":"2025-05-21T19:45:42","modified_gmt":"2025-05-21T11:45:42","slug":"%e8%b0%b7%e6%ad%8c%e5%8f%91%e5%b8%83-gemma-3n-%e7%ab%af%e4%be%a7%e5%a4%9a%e6%a8%a1%e6%80%81%e6%a8%a1%e5%9e%8b%ef%bc%8c2gb-%e5%86%85%e5%ad%98%e4%b9%9f%e8%83%bd%e7%8e%a9%e8%bd%ac-ai","status":"publish","type":"post","link":"http:\/\/www.1ai.net\/en\/35791.html","title":{"rendered":"Google Releases Gemma 3n End-Side Multi-Modal Model for AI with 2GB of RAM"},"content":{"rendered":"<p>5 \u6708 21 \u65e5\u6d88\u606f\uff0c\u4eca\u5929\u53ec\u5f00\u7684 2025 \u5e74 I\/O \u5f00\u53d1\u8005\u5927\u4f1a\u4e0a\uff0c<a href=\"http:\/\/www.1ai.net\/en\/tag\/%e8%b0%b7%e6%ad%8c\" title=\"\u3010\u67e5\u770b\u542b\u6709[\u8c37\u6b4c]\u6807\u7b7e\u7684\u6587\u7ae0\u3011\" target=\"_blank\" rel=\"noopener\">\u8c37\u6b4c<\/a>\u63a8\u51fa\u4e86 <a href=\"http:\/\/www.1ai.net\/en\/tag\/gemma-3n\" title=\"\u3010\u67e5\u770b\u542b\u6709[Gemma 3n]\u6807\u7b7e\u7684\u6587\u7ae0\u3011\" target=\"_blank\" rel=\"noopener\">Gemma 3n<\/a> \u7aef\u4fa7\u591a\u6a21\u6001<a href=\"http:\/\/www.1ai.net\/en\/tag\/ai\" title=\"\u3010\u67e5\u770b\u542b\u6709[AI]\u6807\u7b7e\u7684\u6587\u7ae0\u3011\" target=\"_blank\" rel=\"noopener\">AI<\/a>\u6a21\u578b\uff0c<strong>\u53ef\u4ee5\u5728\u624b\u673a\u3001\u5e73\u677f\u548c\u7b14\u8bb0\u672c\u7535\u8111\u4e0a\u672c\u5730\u8fd0\u884c\uff0c\u5904\u7406\u97f3\u9891\u3001\u6587\u672c\u3001\u56fe\u7247\u548c\u89c6\u9891\u591a\u79cd\u6570\u636e\u7c7b\u578b\u3002<\/strong><\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-35792\" title=\"1e1ff986j00swm0n00082d000v900hkp\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/05\/1e1ff986j00swm0n00082d000v900hkp.jpg\" alt=\"1e1ff986j00swm0n00082d000v900hkp\" width=\"1125\" height=\"632\" \/><\/p>\n<p>Gemma \u4ea7\u54c1\u7ecf\u7406 Gus Martins \u900f\u9732\uff0cGemma 3n \u53ef\u4ee5\u5728\u5185\u5b58\u5c0f\u4e8e 2GB \u7684\u8bbe\u5907\u4e0a\u8fd0\u884c\uff0c\u91c7\u7528 Gemini Nano \u540c\u6b3e\u67b6\u6784\uff0c\u4f46\u6027\u80fd\u8868\u73b0\u66f4\u51fa\u8272\u3002\u4ed6\u5f3a\u8c03\uff1a\u201cGemma 3n \u4e3a\u8bbe\u5907\u7aef AI \u5e94\u7528\u5e26\u6765\u4e86\u60ca\u4eba\u7684\u8868\u73b0\u529b\u3002\u201d<\/p>\n<p>1AI\u63f4\u5f15\u535a\u6587\u4ecb\u7ecd\uff0cGemma 3n \u7684\u6700\u5927\u4eae\u70b9\u5728\u4e8e\u91c7\u7528\u4e86\u8c37\u6b4c DeepMind \u5f00\u53d1\u7684 Per-Layer Embeddings\uff08PLE\uff09\u6280\u672f\uff0c\u663e\u8457\u964d\u4f4e\u4e86\u6a21\u578b\u7684\u5185\u5b58\u9700\u6c42\u3002\u5c3d\u7ba1\u5176\u53c2\u6570\u91cf\u5206\u522b\u4e3a 5B \u548c 8B\uff0c\u4f46\u5185\u5b58\u5360\u7528\u4ec5\u76f8\u5f53\u4e8e 2B \u548c 4B \u6a21\u578b\u3002<\/p>\n<p>\u8c37\u6b4c\u5ba3\u79f0\uff0cGemma 3n \u7684\u5185\u5b58\u5360\u7528\u5206\u522b\u4f4e\u81f3 2GB \u548c 3GB\u3002\u6b64\u5916\uff0c\u901a\u8fc7 PLE\u3001KVC \u5171\u4eab\u548c\u9ad8\u7ea7\u6fc0\u6d3b\u91cf\u5316\u7b49\u6280\u672f\uff0cGemma 3n \u5728\u79fb\u52a8\u8bbe\u5907\u4e0a\u7684\u54cd\u5e94\u901f\u5ea6\u63d0\u5347 1.5 \u500d\uff0c\u8d28\u91cf\u4e5f\u4f18\u4e8e Gemma 3 4B \u7248\u672c\u3002<\/p>\n<p>Gemma 3n \u5728\u975e\u82f1\u8bed\u8bed\u8a00\u5904\u7406\u4e0a\u4e5f\u6709\u7a81\u7834\uff0c\u5c24\u5176\u5728\u65e5\u8bed\u3001\u5fb7\u8bed\u3001\u97e9\u8bed\u3001\u897f\u73ed\u7259\u8bed\u548c\u6cd5\u8bed\u4e2d\u8868\u73b0\u4f18\u5f02\uff0c\u5728\u591a\u8bed\u8a00\u57fa\u51c6\u6d4b\u8bd5 WMT24++ \u4e2d\uff0c\u8be5\u6a21\u578b\u5f97\u5206\u9ad8\u8fbe 50.1%\u3002<\/p>\n<p>\u7528\u6237\u65e0\u9700\u4efb\u4f55\u8bbe\u7f6e\uff0c\u76ee\u524d\u53ef\u901a\u8fc7 Google AI Studio \u76f4\u63a5\u5728\u6d4f\u89c8\u5668\u4e2d\u4f7f\u7528 Gemma 3n\uff1b\u5f00\u53d1\u8005\u5219\u53ef\u901a\u8fc7 Google AI Edge \u96c6\u6210\u672c\u5730\u529f\u80fd\uff0c\u652f\u6301\u6587\u672c\u548c\u56fe\u50cf\u7406\u89e3\u4e0e\u751f\u6210\uff0c\u672a\u6765\u8fd8\u5c06\u6269\u5c55\u66f4\u591a\u80fd\u529b\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>On May 21, at today ' s I\/O Developer Conference, Google launched the Gemma 3n end multimodular AI model, which can be operated locally on mobile phones, tablets and laptops, processing audio, text, pictures and video data types. Gemma Product Manager Gus Martins revealed that Gemma 3n could operate on equipment less than 2GB, using the Gemini Nano equivalent structure, but performed better. He stressed: \"Gemma 3n has brought amazing performance to the end-of-equipment AI application.\" 1AI quotes Boven, Gemma 3n's greatest highlight is Google DeepMind<\/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":[411,6708,6707,281],"collection":[],"class_list":["post-35791","post","type-post","status-publish","format-standard","hentry","category-news","tag-ai","tag-gemma-3n","tag-6707","tag-281"],"acf":[],"_links":{"self":[{"href":"http:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/35791","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=35791"}],"version-history":[{"count":0,"href":"http:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/35791\/revisions"}],"wp:attachment":[{"href":"http:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=35791"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=35791"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=35791"},{"taxonomy":"collection","embeddable":true,"href":"http:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=35791"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}