{"id":35070,"date":"2024-12-18T00:00:01","date_gmt":"2024-12-17T16:00:01","guid":{"rendered":"https:\/\/17aitech.com\/?p=35070"},"modified":"2024-12-18T00:00:01","modified_gmt":"2024-12-17T16:00:01","slug":"%e9%9d%9etransformer%e6%9e%b6%e6%9e%84%e7%ab%99%e8%b5%b7%e6%9d%a5%e4%ba%86%ef%bc%81%e9%a6%96%e4%b8%aa%e7%ba%af%e6%97%a0%e6%b3%a8%e6%84%8f%e5%8a%9b%e5%a4%a7%e6%a8%a1%e5%9e%8b%ef%bc%8c%e8%b6%85%e8%b6%8a","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=35070","title":{"rendered":"\u975eTransformer\u67b6\u6784\u7ad9\u8d77\u6765\u4e86\uff01\u9996\u4e2a\u7eaf\u65e0\u6ce8\u610f\u529b\u5927\u6a21\u578b\uff0c\u8d85\u8d8a\u5f00\u6e90\u5de8\u5934Llama 3.1"},"content":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2024-08-13-8\" target=\"_blank\">\u975eTransformer\u67b6\u6784\u7ad9\u8d77\u6765\u4e86\uff01\u9996\u4e2a\u7eaf\u65e0\u6ce8\u610f\u529b\u5927\u6a21\u578b\uff0c\u8d85\u8d8a\u5f00\u6e90\u5de8\u5934Llama 3.1<\/a><\/p>\n<blockquote data-author-name=\"\" data-content-utf8-length=\"32\" data-source-title=\"\" data-type=\"2\" data-url=\"\">\n<section>\n<section>\n<section>Mamba \u67b6\u6784\u7684\u5927\u6a21\u578b\u53c8\u4e00\u6b21\u5411 Transformer \u53d1\u8d77\u4e86\u6311\u6218\u3002<\/section>\n<\/section>\n<\/section>\n<\/blockquote>\n<section><\/section>\n<section>Mamba \u67b6\u6784\u6a21\u578b\u8fd9\u6b21\u7ec8\u4e8e\u8981\u300c\u7ad9\u300d\u8d77\u6765\u4e86\uff1f\u81ea 2023 \u5e74 12 \u6708\u9996\u6b21\u63a8\u51fa\u4ee5\u6765\uff0cMamba \u4fbf\u6210\u4e3a\u4e86 Transformer \u7684\u5f3a\u6709\u529b\u7ade\u4e89\u5bf9\u624b\u3002<\/section>\n<section><\/section>\n<section>\u6b64\u540e\uff0c\u91c7\u7528 Mamba \u67b6\u6784\u7684\u6a21\u578b\u4e0d\u65ad\u51fa\u73b0\uff0c\u6bd4\u5982 Mistral \u53d1\u5e03\u7684\u9996\u4e2a\u57fa\u4e8e Mamba \u67b6\u6784\u7684\u5f00\u6e90\u5927\u6a21\u578b Codestral 7B\u3002<\/section>\n<section><\/section>\n<section>\u4eca\u5929\uff0c\u963f\u5e03\u624e\u6bd4\u6280\u672f\u521b\u65b0\u7814\u7a76\u6240\uff08TII\uff09\u53d1\u5e03\u4e86\u4e00\u4e2a<strong>\u65b0\u7684\u5f00\u6e90 Mamba \u6a21\u578b \u2014\u2014Falcon Mamba 7B<\/strong>\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-83b0d93936d7f01f0a778d594c78ef8e.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-83b0d93936d7f01f0a778d594c78ef8e.png\"><\/a><\/p>\n<section>\u5148\u6765\u603b\u7ed3\u4e00\u6ce2 Falcon Mamba 7B \u7684\u4eae\u70b9\uff1a\u65e0\u9700\u589e\u52a0\u5185\u5b58\u5b58\u50a8\uff0c\u5c31\u53ef\u4ee5\u5904\u7406\u4efb\u610f\u957f\u5ea6\u7684\u5e8f\u5217\uff0c\u5e76\u4e14\u80fd\u591f\u5728\u5355\u4e2a 24GB A10 GPU \u4e0a\u8fd0\u884c\u3002<\/section>\n<section><\/section>\n<section>\u76ee\u524d\u53ef\u4ee5\u5728\u00a0Hugging Face \u4e0a\u67e5\u770b\u5e76\u4f7f\u7528 Falcon Mamba 7B\uff0c\u8fd9\u4e2a\u4ec5\u7528\u56e0\u679c\u89e3\u7801\u5668\u7684\u6a21\u578b\u91c7\u7528\u4e86\u65b0\u9896\u7684\u00a0<strong>Mamba \u72b6\u6001\u7a7a\u95f4<mark data-type=\"tech_tasks\" data-id=\"fee178e8-ee20-42fc-8f8f-1d41c1f34e8f\">\u8bed\u8a00\u6a21\u578b<\/mark>\uff08State Space Language Model, SSLM\uff09\u67b6\u6784<\/strong>\u6765\u5904\u7406\u5404\u79cd<mark data-type=\"tech_tasks\" data-id=\"f02bb483-1c90-4010-849a-f2d7e3f0ad59\">\u6587\u672c\u751f\u6210<\/mark>\u4efb\u52a1\u3002<\/section>\n<section><\/section>\n<section>\u4ece\u7ed3\u679c\u6765\u770b\uff0cFalcon Mamba 7B \u5728\u4e00\u4e9b<mark data-type=\"concepts\" data-id=\"308c3a45-0fee-4ec6-858e-85b15f440fc0\">\u57fa\u51c6<\/mark>\u4e0a\u8d85\u8d8a\u540c\u5c3a\u5bf8\u7ea7\u522b\u7684\u9886\u5148\u6a21\u578b\uff0c\u5305\u62ec Meta \u7684 Llama 3 8B\u3001Llama 3.1 8B \u548c Mistral 7B\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-b8e7829b6525df2bc8e8615252338ea9.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-b8e7829b6525df2bc8e8615252338ea9.png\"><\/a><\/p>\n<section>Falcon Mamba 7B \u5206\u4e3a\u56db\u4e2a\u53d8\u4f53\u6a21\u578b\uff0c\u5206\u522b\u662f\u57fa\u7840\u7248\u672c\u3001\u6307\u4ee4\u5fae\u8c03\u7248\u672c\u30014bit \u7248\u672c\u548c\u6307\u4ee4\u5fae\u8c03 4bit \u7248\u672c\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-4fe65a96ae30bb5031db2837a7c46fb5.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-4fe65a96ae30bb5031db2837a7c46fb5.png\"><\/a><\/p>\n<section>\u4f5c\u4e3a\u4e00\u4e2a\u5f00\u6e90\u6a21\u578b\uff0c Falcon Mamba 7B \u91c7\u7528\u4e86\u57fa\u4e8e Apache 2.0 \u7684\u8bb8\u53ef\u8bc1\u300cFalcon License 2.0\u300d\uff0c\u652f\u6301\u7814\u7a76\u548c\u5e94\u7528\u76ee\u7684\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-0d8d95c7d2a71861981aac6b19a795b4.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-0d8d95c7d2a71861981aac6b19a795b4.png\"><\/a><\/p>\n<section>Hugging Face \u5730\u5740\uff1ahttps:\/\/huggingface.co\/tiiuae\/falcon-mamba-7b<\/section>\n<section><\/section>\n<section>Falcon Mamba 7B \u4e5f\u6210\u4e3a\u4e86\u7ee7 Falcon 180B\u3001Falcon 40B \u548c Falcon 2 \u4e4b\u540e\uff0cTII \u5f00\u6e90\u7684\u7b2c\u56db\u4e2a\u6a21\u578b\uff0c\u5e76\u4e14\u662f<strong>\u9996\u4e2a Mamba SSLM \u67b6\u6784\u6a21\u578b<\/strong>\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-c5ea5fb16d8510cb4f7505dbc8c08d31.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-c5ea5fb16d8510cb4f7505dbc8c08d31.png\"><\/a><\/p>\n<section><strong>\u9996\u4e2a\u901a\u7528\u7684\u5927\u578b\u7eaf Mamba \u6a21\u578b<\/strong><\/section>\n<section><\/section>\n<section>\u4e00\u76f4\u4ee5\u6765\uff0c\u57fa\u4e8e Transformer \u7684\u6a21\u578b\u4e00\u76f4\u5360\u636e\u7740\u751f\u6210\u5f0f AI \u7684\u7edf\u6cbb\u5730\u4f4d\uff0c\u7136\u800c\uff0c\u7814\u7a76\u4eba\u5458\u6ce8\u610f\u5230\uff0cTransformer \u67b6\u6784\u5728\u5904\u7406\u8f83\u957f\u7684\u6587\u672c\u4fe1\u606f\u65f6\u53ef\u80fd\u4f1a\u9047\u5230\u56f0\u96be\u3002<\/section>\n<section><\/section>\n<section>\u672c\u8d28\u4e0a\uff0cTransformer \u4e2d\u7684<mark data-type=\"tech_methods\" data-id=\"60bee267-89dc-4606-ba24-6b3f7d1f6189\">\u6ce8\u610f\u529b\u673a\u5236<\/mark>\u901a\u8fc7\u5c06\u6bcf\u4e2a\u5355\u8bcd\uff08\u6216 token\uff09\u4e0e\u6587\u672c\u4e2d\u7684\u6bcf\u4e2a\u5355\u8bcd\u8fdb\u884c\u6bd4\u8f83\u6765\u7406\u89e3\u4e0a\u4e0b\u6587\uff0c\u5b83\u9700\u8981\u66f4\u591a\u7684\u8ba1\u7b97\u80fd\u529b\u548c\u5185\u5b58\u9700\u6c42\u6765\u5904\u7406\u4e0d\u65ad\u589e\u957f\u7684\u4e0a\u4e0b\u6587\u7a97\u53e3\u3002<\/section>\n<section><\/section>\n<section>\u4f46\u662f\u5982\u679c\u4e0d\u76f8\u5e94\u5730\u6269\u5c55\u8ba1\u7b97\u8d44\u6e90\uff0c\u6a21\u578b\u63a8\u7406\u901f\u5ea6\u5c31\u4f1a\u53d8\u6162\uff0c\u8d85\u8fc7\u4e00\u5b9a\u957f\u5ea6\u7684\u6587\u672c\u5c31\u6ca1\u6cd5\u5904\u7406\u4e86\u3002\u4e3a\u4e86\u514b\u670d\u8fd9\u4e9b\u969c\u788d\uff0c\u72b6\u6001\u7a7a\u95f4<mark data-type=\"tech_tasks\" data-id=\"fee178e8-ee20-42fc-8f8f-1d41c1f34e8f\">\u8bed\u8a00\u6a21\u578b<\/mark> (SSLM) \u67b6\u6784\u5e94\u8fd0\u800c\u751f\uff0c\u8be5\u67b6\u6784\u901a\u8fc7\u5728\u5904\u7406\u5355\u8bcd\u65f6\u4e0d\u65ad\u66f4\u65b0\u72b6\u6001\u6765\u5de5\u4f5c\uff0c\u5df2\u6210\u4e3a\u4e00\u79cd\u6709\u524d\u9014\u7684\u66ff\u4ee3\u65b9\u6848\uff0c\u5305\u62ec TII \u5728\u5185\u7684\u5f88\u591a\u673a\u6784\u90fd\u5728\u90e8\u7f72\u8fd9\u79cd\u67b6\u6784\u3002<\/section>\n<section><\/section>\n<section>Falcon Mamba 7B \u91c7\u7528\u4e86\u5361\u5185\u57fa\u6885\u9686\u5927\u5b66\u548c\u666e\u6797\u65af\u987f\u5927\u5b66\u7814\u7a76\u4eba\u5458\u6700\u521d\u5728 2023 \u5e74 12 \u6708\u7684\u4e00\u7bc7\u8bba\u6587\u4e2d\u63d0\u51fa\u7684\u00a0<a data-itemshowtype=\"0\" data-linktype=\"2\" href=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzA3MzI4MjgzMw==&amp;mid=2650899414&amp;idx=1&amp;sn=b6c34617d6d0f45b3ad3da9ea6385206&amp;chksm=84e447a8b393cebe030f1b2dc3e5372e3258f772bdf8510134e1d551d82558225a8454399619&amp;scene=21#wechat_redirect\" target=\"_blank\">Mamba SSM \u67b6\u6784<\/a>\u3002<\/section>\n<section><\/section>\n<section>\u8be5\u67b6\u6784\u4f7f\u7528\u4e00\u79cd\u9009\u62e9\u673a\u5236\uff0c\u5141\u8bb8\u6a21\u578b\u6839\u636e\u8f93\u5165\u52a8\u6001\u8c03\u6574\u5176<mark data-type=\"concepts\" data-id=\"2e982b73-88e2-41e8-a430-f7ae5a9af4bf\">\u53c2\u6570<\/mark>\u3002\u8fd9\u6837\uff0c\u6a21\u578b\u53ef\u4ee5\u5173\u6ce8\u6216\u5ffd\u7565\u7279\u5b9a\u8f93\u5165\uff0c\u7c7b\u4f3c\u4e8e<mark data-type=\"tech_methods\" data-id=\"60bee267-89dc-4606-ba24-6b3f7d1f6189\">\u6ce8\u610f\u529b\u673a\u5236<\/mark>\u5728 Transformer \u4e2d\u7684\u5de5\u4f5c\u65b9\u5f0f\uff0c\u540c\u65f6\u63d0\u4f9b\u5904\u7406\u957f\u6587\u672c\u5e8f\u5217\uff08\u4f8b\u5982\u6574\u672c\u4e66\uff09\u7684\u80fd\u529b\uff0c\u800c\u65e0\u9700\u989d\u5916\u7684\u5185\u5b58\u6216\u8ba1\u7b97\u8d44\u6e90\u3002<\/section>\n<section><\/section>\n<section>TII \u6307\u51fa\uff0c\u8be5\u65b9\u6cd5\u4f7f\u6a21\u578b\u9002\u7528\u4e8e\u4f01\u4e1a\u7ea7<mark data-type=\"tech_tasks\" data-id=\"7237ee05-f07e-4fd9-a9ac-96e5bc1f50e9\">\u673a\u5668\u7ffb\u8bd1<\/mark>\u3001\u6587\u672c\u6458\u8981\u3001<mark data-type=\"tech_tasks\" data-id=\"6e614199-9e49-450e-9078-61fb2b122da9\">\u8ba1\u7b97\u673a\u89c6\u89c9<\/mark>\u548c\u97f3\u9891\u5904\u7406\u4efb\u52a1\u4ee5\u53ca\u4f30\u8ba1\u548c\u9884\u6d4b\u7b49\u4efb\u52a1\u3002<\/section>\n<section><\/section>\n<section><strong>\u8bad\u7ec3\u6570\u636e<\/strong><\/section>\n<section><\/section>\n<section>Falcon Mamba 7B\u00a0<strong>\u8bad\u7ec3\u6570\u636e\u9ad8\u8fbe 5500GT<\/strong> \uff0c\u4e3b\u8981\u7531\u00a0RefinedWeb \u6570\u636e\u96c6\u7ec4\u6210\uff0c\u5e76\u6dfb\u52a0\u4e86\u6765\u81ea\u516c\u5171\u6e90\u7684\u9ad8\u8d28\u91cf\u6280\u672f\u6570\u636e\u3001\u4ee3\u7801\u6570\u636e\u548c\u6570\u5b66\u6570\u636e\u3002\u6240\u6709\u6570\u636e\u901a\u8fc7 Falcon-7B\/11B \u6807\u8bb0\u5668\u8fdb\u884c tokenized \u64cd\u4f5c\u3002<\/section>\n<section><\/section>\n<section>\u4e0e\u5176\u4ed6 Falcon \u7cfb\u5217\u6a21\u578b\u7c7b\u4f3c\uff0cFalcon Mamba 7B \u91c7\u7528\u591a\u9636\u6bb5\u8bad\u7ec3\u7b56\u7565\u8fdb\u884c\u8bad\u7ec3\uff0c<strong>\u4e0a\u4e0b\u6587\u957f\u5ea6\u4ece 2048 \u589e\u52a0\u5230\u4e86 8192<\/strong>\u3002\u6b64\u5916\uff0c\u53d7\u5230\u8bfe\u7a0b\u5b66\u4e60\u6982\u5ff5\u7684\u542f\u53d1\uff0cTII \u5728\u6574\u4e2a\u8bad\u7ec3\u9636\u6bb5\u7cbe\u5fc3\u9009\u62e9\u4e86\u6df7\u5408\u6570\u636e\uff0c\u5145\u5206\u8003\u8651\u4e86\u6570\u636e\u7684\u591a\u6837\u6027\u548c\u590d\u6742\u6027\u3002<\/section>\n<section><\/section>\n<section>\u5728\u6700\u540e\u7684\u8bad\u7ec3\u9636\u6bb5\uff0cTII \u4f7f\u7528\u4e86\u4e00\u5c0f\u90e8\u5206\u9ad8\u8d28\u91cf\u7cbe\u9009\u6570\u636e\uff08\u5373\u6765\u81ea Fineweb-edu \u7684\u6837\u672c\uff09\uff0c\u4ee5\u8fdb\u4e00\u6b65\u63d0\u5347\u6027\u80fd\u3002<\/section>\n<section><\/section>\n<section><strong>\u8bad\u7ec3\u8fc7\u7a0b\u3001\u8d85<mark data-type=\"concepts\" data-id=\"2e982b73-88e2-41e8-a430-f7ae5a9af4bf\">\u53c2\u6570<\/mark><\/strong><\/section>\n<section><\/section>\n<section>Falcon Mamba 7B \u7684\u5927\u90e8\u5206\u8bad\u7ec3\u662f<strong>\u5728 256 \u4e2a H100 80GB GPU \u4e0a\u5b8c\u6210<\/strong>\u7684\uff0c\u91c7\u7528\u4e86 3D \u5e76\u884c\uff08TP=1\u3001PP=1\u3001DP=256\uff09\u4e0e ZeRO \u76f8\u7ed3\u5408\u7684\u7b56\u7565\u3002\u4e0b\u56fe\u4e3a\u6a21\u578b\u8d85<mark data-type=\"concepts\" data-id=\"2e982b73-88e2-41e8-a430-f7ae5a9af4bf\">\u53c2\u6570<\/mark>\u7ec6\u8282\uff0c\u5305\u62ec\u7cbe\u5ea6\u3001<mark data-type=\"concepts\" data-id=\"fa50298e-1a85-4af0-ae96-a82708f4b610\">\u4f18\u5316\u5668<\/mark>\u3001\u6700\u5927<mark data-type=\"concepts\" data-id=\"444ad8d1-7262-46c9-94b7-fe0aa9227006\">\u5b66\u4e60\u7387<\/mark>\u3001<mark data-type=\"concepts\" data-id=\"149a12cf-10c2-4555-9899-cc6dee319ef5\">\u6743\u91cd<\/mark>\u8870\u51cf\u548c batch \u5927\u5c0f\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-5c37c464f26e9c014ec81b9e959512e8.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-5c37c464f26e9c014ec81b9e959512e8.png\"><\/a><\/p>\n<section>\u5177\u4f53\u800c\u8a00\uff0cFalcon Mamba 7B \u7ecf\u8fc7\u4e86 AdamW <mark data-type=\"concepts\" data-id=\"fa50298e-1a85-4af0-ae96-a82708f4b610\">\u4f18\u5316\u5668<\/mark>\u3001WSD\uff08\u9884\u70ed &#8211; \u7a33\u5b9a &#8211; \u8870\u51cf\uff09<mark data-type=\"concepts\" data-id=\"444ad8d1-7262-46c9-94b7-fe0aa9227006\">\u5b66\u4e60\u7387<\/mark>\u8ba1\u5212\u7684\u8bad\u7ec3\uff0c \u5e76\u4e14\u5728\u524d 50 GT \u7684\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0cbatch \u5927\u5c0f\u4ece b_min=128 \u589e\u52a0\u5230\u4e86 b_max=2048\u3002<\/section>\n<section><\/section>\n<section>\u5728\u7a33\u5b9a\u9636\u6bb5\uff0cTII \u4f7f\u7528\u4e86\u6700\u5927<mark data-type=\"concepts\" data-id=\"444ad8d1-7262-46c9-94b7-fe0aa9227006\">\u5b66\u4e60\u7387<\/mark> \u03b7_max=6.4\u00d710^\u22124\uff0c\u7136\u540e\u4f7f\u7528\u8d85\u8fc7 500GT \u7684\u6307\u6570\u8ba1\u5212\u5c06\u5176\u8870\u51cf\u5230\u6700\u5c0f\u503c<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-fc60fb5b7c262382ce5132beeb351852.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-fc60fb5b7c262382ce5132beeb351852.png\"><\/a>\u3002\u540c\u65f6\uff0cTII \u5728\u52a0\u901f\u9636\u6bb5\u91c7\u7528\u4e86 BatchScaling \u4ee5\u91cd\u65b0\u8c03\u6574<mark data-type=\"concepts\" data-id=\"444ad8d1-7262-46c9-94b7-fe0aa9227006\">\u5b66\u4e60\u7387<\/mark> \u03b7\uff0c\u4f7f\u5f97 Adam \u566a\u58f0\u6e29\u5ea6<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-336262667dad950ef31f420299fc362c.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-336262667dad950ef31f420299fc362c.png\"><\/a>\u4fdd\u6301\u6052\u5b9a\u3002<\/section>\n<section><\/section>\n<section><strong>\u6574\u4e2a\u6a21\u578b\u8bad\u7ec3\u82b1\u8d39\u4e86\u5927\u7ea6\u4e24\u4e2a\u6708\u65f6\u95f4<\/strong>\u3002<\/section>\n<section><\/section>\n<section><strong>\u6a21\u578b\u8bc4\u4f30<\/strong><\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u4e86\u89e3 Falcon Mamba 7B \u4e0e\u540c\u5c3a\u5bf8\u7ea7\u522b\u9886\u5148\u7684 Transformer \u6a21\u578b\u76f8\u6bd4\u5982\u4f55\uff0c\u8be5\u7814\u7a76\u8fdb\u884c\u4e86\u4e00\u9879\u6d4b\u8bd5\uff0c\u4ee5\u786e\u5b9a\u4f7f\u7528\u5355\u4e2a 24GB A10GPU \u65f6\u6a21\u578b\u53ef\u4ee5\u5904\u7406\u7684\u6700\u5927\u4e0a\u4e0b\u6587\u957f\u5ea6\u3002<\/section>\n<section><\/section>\n<section>\u7ed3\u679c\u663e\u793a\uff0cFalcon Mamba \u80fd\u591f\u6bd4\u5f53\u524d\u7684 Transformer \u6a21\u578b\u9002\u5e94\u66f4\u5927\u7684\u5e8f\u5217\uff0c\u540c\u65f6<strong>\u7406\u8bba\u4e0a\u80fd\u591f\u9002\u5e94\u65e0\u9650\u7684\u4e0a\u4e0b\u6587\u957f\u5ea6<\/strong>\u3002\u00a0<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-d033fefe2be687851718b88821a90c12.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-d033fefe2be687851718b88821a90c12.png\"><\/a><\/p>\n<section>\u63a5\u4e0b\u6765\uff0c\u7814\u7a76\u8005\u4f7f\u7528\u6279\u5904\u7406\u5927\u5c0f\u4e3a 1 \uff0c\u786c\u4ef6\u91c7\u7528 H100 GPU \u7684\u8bbe\u7f6e\u4e2d\u6d4b\u91cf\u6a21\u578b\u751f\u6210\u541e\u5410\u91cf\u3002\u7ed3\u679c\u5982\u4e0b\u56fe\u6240\u793a\uff0cFalcon Mamba \u4ee5\u6052\u5b9a\u7684\u541e\u5410\u91cf\u751f\u6210\u6240\u6709 token\uff0c\u5e76\u4e14 CUDA \u5cf0\u503c\u5185\u5b58\u6ca1\u6709\u4efb\u4f55\u589e\u52a0\u3002\u5bf9\u4e8e Transformer \u6a21\u578b\uff0c\u5cf0\u503c\u5185\u5b58\u4f1a\u589e\u52a0\uff0c\u751f\u6210\u901f\u5ea6\u4f1a\u968f\u7740\u751f\u6210\u7684 token \u6570\u91cf\u7684\u589e\u52a0\u800c\u51cf\u6162\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-52da79c9108c01209b46b2f254f1c566.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-52da79c9108c01209b46b2f254f1c566.png\"><\/a><\/p>\n<section>\u5373\u4f7f\u5728\u6807\u51c6\u7684\u884c\u4e1a<mark data-type=\"concepts\" data-id=\"308c3a45-0fee-4ec6-858e-85b15f440fc0\">\u57fa\u51c6<\/mark>\u6d4b\u8bd5\u4e2d\uff0c\u65b0\u6a21\u578b\u7684\u6027\u80fd\u4e5f\u4f18\u4e8e\u6216\u63a5\u8fd1\u4e8e\u6d41\u884c\u7684 transformer \u6a21\u578b\u4ee5\u53ca\u7eaf\u72b6\u6001\u7a7a\u95f4\u6a21\u578b\u548c\u6df7\u5408\u72b6\u6001\u7a7a\u95f4\u6a21\u578b\u3002<\/section>\n<section><\/section>\n<section>\u4f8b\u5982\uff0c\u5728 Arc\u3001TruthfulQA \u548c GSM8K <mark data-type=\"concepts\" data-id=\"308c3a45-0fee-4ec6-858e-85b15f440fc0\">\u57fa\u51c6<\/mark>\u6d4b\u8bd5\u4e2d\uff0cFalcon Mamba 7B \u7684\u5f97\u5206\u5206\u522b\u4e3a 62.03%\uff0c53.42% \u548c 52.54%\uff0c\u8d85\u8fc7\u4e86 Llama 3 8 B, Llama 3.1 8B, Gemma 7B \u548c Mistral 7B\u3002\u7136\u800c\uff0c\u5728 MMLU \u548c Hellaswag <mark data-type=\"concepts\" data-id=\"308c3a45-0fee-4ec6-858e-85b15f440fc0\">\u57fa\u51c6<\/mark>\u6d4b\u8bd5\u4e2d\uff0cFalcon Mamba 7B \u8fdc\u8fdc\u843d\u540e\u4e8e\u8fd9\u4e9b\u6a21\u578b\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-3389a6c3a40d20700ba83a88b4294f38.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-3389a6c3a40d20700ba83a88b4294f38.png\"><\/a><\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-edd55dabe0e88106f50e9edd99c8c179.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-edd55dabe0e88106f50e9edd99c8c179.png\"><\/a><\/p>\n<section>TII \u9996\u5e2d\u7814\u7a76\u5458 Hakim Hacid \u5728\u4e00\u4efd\u58f0\u660e\u4e2d\u8868\u793a\uff1aFalcon Mamba 7B \u7684\u53d1\u5e03\u4ee3\u8868\u7740\u8be5\u673a\u6784\u5411\u524d\u8fc8\u51fa\u7684\u91cd\u5927\u4e00\u6b65\uff0c\u5b83\u6fc0\u53d1\u4e86\u65b0\u7684\u89c2\u70b9\uff0c\u5e76\u8fdb\u4e00\u6b65\u63a8\u52a8\u4e86\u5bf9\u667a\u80fd\u7cfb\u7edf\u7684\u63a2\u7d22\u3002\u5728 TII\uff0c\u4ed6\u4eec\u6b63\u5728\u7a81\u7834 SSLM \u548c transformer \u6a21\u578b\u7684\u754c\u9650\uff0c\u4ee5\u6fc0\u53d1\u751f\u6210\u5f0f AI \u7684\u8fdb\u4e00\u6b65\u521b\u65b0\u3002<\/section>\n<section><\/section>\n<section>\u76ee\u524d\uff0cTII \u7684 Falcon \u7cfb\u5217<mark data-type=\"tech_tasks\" data-id=\"fee178e8-ee20-42fc-8f8f-1d41c1f34e8f\">\u8bed\u8a00\u6a21\u578b<\/mark>\u4e0b\u8f7d\u91cf\u5df2\u8d85\u8fc7 4500 \u4e07\u6b21 \u2014\u2014 \u6210\u4e3a\u963f\u8054\u914b\u6700\u6210\u529f\u7684 LLM \u7248\u672c\u4e4b\u4e00\u3002<\/section>\n<section><\/section>\n<section>Falcon Mamba 7B \u8bba\u6587\u5373\u5c06\u653e\u51fa\uff0c\u5927\u5bb6\u53ef\u4ee5\u7b49\u4e00\u7b49\u3002<\/section>\n<section><\/section>\n<section><em><sup>\u53c2\u8003\u94fe\u63a5\uff1a<\/sup><\/em><\/section>\n<section><sup><em>https:\/\/huggingface.co\/blog\/falconmamba<\/em><\/sup><\/section>\n<section><em><sup>https:\/\/venturebeat.com\/ai\/falcon-mamba-7bs-powerful-new-ai-architecture-offers-alternative-to-transformer-models\/<\/sup><\/em><\/section>\n<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2024-08-13-8\" target=\"_blank\">\u975eTransformer\u67b6\u6784\u7ad9\u8d77\u6765\u4e86\uff01\u9996\u4e2a\u7eaf\u65e0\u6ce8\u610f\u529b\u5927\u6a21\u578b\uff0c\u8d85\u8d8a\u5f00\u6e90\u5de8\u5934Llama 3.1<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u975eTransforme [&hellip;]<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"","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":"","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":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","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":[27],"tags":[68,66],"class_list":["post-35070","post","type-post","status-publish","format-standard","hentry","category-news","tag-68","tag-66"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.4 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u975eTransformer\u67b6\u6784\u7ad9\u8d77\u6765\u4e86\uff01\u9996\u4e2a\u7eaf\u65e0\u6ce8\u610f\u529b\u5927\u6a21\u578b\uff0c\u8d85\u8d8a\u5f00\u6e90\u5de8\u5934Llama 3.1 - \u4e00\u8d77AI\u6280\u672f<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/17aitech.com\/?p=35070\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/17aitech.com\/?p=35070\",\"url\":\"https:\/\/17aitech.com\/?p=35070\",\"name\":\"\u975eTransformer\u67b6\u6784\u7ad9\u8d77\u6765\u4e86\uff01\u9996\u4e2a\u7eaf\u65e0\u6ce8\u610f\u529b\u5927\u6a21\u578b\uff0c\u8d85\u8d8a\u5f00\u6e90\u5de8\u5934Llama 3.1 - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\/\/17aitech.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/17aitech.com\/?p=35070#primaryimage\"},\"image\":{\"@id\":\"https:\/\/17aitech.com\/?p=35070#primaryimage\"},\"thumbnailUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-83b0d93936d7f01f0a778d594c78ef8e.png\",\"datePublished\":\"2024-12-17T16:00:01+00:00\",\"author\":{\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\/\/17aitech.com\/?p=35070#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/17aitech.com\/?p=35070\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/?p=35070#primaryimage\",\"url\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-83b0d93936d7f01f0a778d594c78ef8e.png\",\"contentUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-83b0d93936d7f01f0a778d594c78ef8e.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/17aitech.com\/?p=35070#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/17aitech.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u975eTransformer\u67b6\u6784\u7ad9\u8d77\u6765\u4e86\uff01\u9996\u4e2a\u7eaf\u65e0\u6ce8\u610f\u529b\u5927\u6a21\u578b\uff0c\u8d85\u8d8a\u5f00\u6e90\u5de8\u5934Llama 3.1\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/17aitech.com\/#website\",\"url\":\"https:\/\/17aitech.com\/\",\"name\":\"\u4e00\u8d77AI\u6280\u672f\",\"description\":\"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca\",\"alternateName\":\"\u4e00\u8d77AI\u6280\u672f\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/17aitech.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\",\"name\":\"AI\u5c0f\u52a9\u624b\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/image\/\",\"url\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"contentUrl\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"caption\":\"AI\u5c0f\u52a9\u624b\"},\"description\":\"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e\",\"url\":\"https:\/\/17aitech.com\/?page_id=33738&user=3\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u975eTransformer\u67b6\u6784\u7ad9\u8d77\u6765\u4e86\uff01\u9996\u4e2a\u7eaf\u65e0\u6ce8\u610f\u529b\u5927\u6a21\u578b\uff0c\u8d85\u8d8a\u5f00\u6e90\u5de8\u5934Llama 3.1 - \u4e00\u8d77AI\u6280\u672f","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/17aitech.com\/?p=35070","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=35070","url":"https:\/\/17aitech.com\/?p=35070","name":"\u975eTransformer\u67b6\u6784\u7ad9\u8d77\u6765\u4e86\uff01\u9996\u4e2a\u7eaf\u65e0\u6ce8\u610f\u529b\u5927\u6a21\u578b\uff0c\u8d85\u8d8a\u5f00\u6e90\u5de8\u5934Llama 3.1 - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=35070#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=35070#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-83b0d93936d7f01f0a778d594c78ef8e.png","datePublished":"2024-12-17T16:00:01+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=35070#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=35070"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=35070#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-83b0d93936d7f01f0a778d594c78ef8e.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/08\/frc-83b0d93936d7f01f0a778d594c78ef8e.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=35070#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u975eTransformer\u67b6\u6784\u7ad9\u8d77\u6765\u4e86\uff01\u9996\u4e2a\u7eaf\u65e0\u6ce8\u610f\u529b\u5927\u6a21\u578b\uff0c\u8d85\u8d8a\u5f00\u6e90\u5de8\u5934Llama 3.1"}]},{"@type":"WebSite","@id":"https:\/\/17aitech.com\/#website","url":"https:\/\/17aitech.com\/","name":"\u4e00\u8d77AI\u6280\u672f","description":"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca","alternateName":"\u4e00\u8d77AI\u6280\u672f","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/17aitech.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d","name":"AI\u5c0f\u52a9\u624b","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/#\/schema\/person\/image\/","url":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","contentUrl":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","caption":"AI\u5c0f\u52a9\u624b"},"description":"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e","url":"https:\/\/17aitech.com\/?page_id=33738&user=3"}]}},"_links":{"self":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/35070","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=35070"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/35070\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=35070"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=35070"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=35070"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}