{"id":37658,"date":"2025-02-02T00:00:55","date_gmt":"2025-02-01T16:00:55","guid":{"rendered":"https:\/\/17aitech.com\/?p=37658"},"modified":"2025-02-02T00:00:55","modified_gmt":"2025-02-01T16:00:55","slug":"aaai-2025-%e5%a4%a7%e6%a8%a1%e5%9e%8b%e6%8e%a8%e7%90%86%e5%8a%a0%e9%80%9f%e6%96%b0%e8%8c%83%e5%bc%8f%ef%bc%9a%e5%8a%a0%e9%80%9f%e6%af%94%e9%ab%98%e8%be%be3-51%e5%80%8d%e3%80%81%e6%88%90%e6%9c%ac","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=37658","title":{"rendered":"AAAI 2025 | \u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u65b0\u8303\u5f0f\uff1a\u52a0\u901f\u6bd4\u9ad8\u8fbe3.51\u500d\u3001\u6210\u672c\u964d\u81f31\/3"},"content":{"rendered":"

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:AAAI 2025 | \u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u65b0\u8303\u5f0f\uff1a\u52a0\u901f\u6bd4\u9ad8\u8fbe3.51\u500d\u3001\u6210\u672c\u964d\u81f31\/3<\/a><\/p>\n

<\/a><\/p>\n

\n
\n
AIxiv\u4e13\u680f\u662f\u673a\u5668\u4e4b\u5fc3\u53d1\u5e03\u5b66\u672f\u3001\u6280\u672f\u5185\u5bb9\u7684\u680f\u76ee\u3002\u8fc7\u53bb\u6570\u5e74\uff0c\u673a\u5668\u4e4b\u5fc3AIxiv\u4e13\u680f\u63a5\u6536\u62a5\u9053\u4e862000\u591a\u7bc7\u5185\u5bb9\uff0c\u8986\u76d6\u5168\u7403\u5404\u5927\u9ad8\u6821\u4e0e\u4f01\u4e1a\u7684\u9876\u7ea7\u5b9e\u9a8c\u5ba4\uff0c\u6709\u6548\u4fc3\u8fdb\u4e86\u5b66\u672f\u4ea4\u6d41\u4e0e\u4f20\u64ad\u3002\u5982\u679c\u60a8\u6709\u4f18\u79c0\u7684\u5de5\u4f5c\u60f3\u8981\u5206\u4eab\uff0c\u6b22\u8fce\u6295\u7a3f\u6216\u8005\u8054\u7cfb\u62a5\u9053\u3002\u6295\u7a3f\u90ae\u7bb1\uff1aliyazhou@jiqizhixin.com\uff1bzhaoyunfeng@jiqizhixin.com<\/section>\n<\/section>\n<\/blockquote>\n
<\/section>\n
\u8fd1\u65e5\uff0c\u4e2d\u56fd\u7535\u4fe1\u7ffc\u652f\u4ed8\u9488\u5bf9\u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u7684\u6700\u65b0\u7814\u7a76\u6210\u679c\u300aFalcon: Faster and Parallel Inference of Large Language Models through Enhanced Semi-Autoregressive Drafting and Custom-Designed Decoding Tree\u300b\u5df2\u88ab AAAI 2025 \u63a5\u6536\u3002<\/section>\n
<\/section>\n
\u8bba\u6587\u4e2d\u63d0\u51fa\u7684 Falcon \u65b9\u6cd5\u662f\u4e00\u79cd\u589e\u5f3a\u534a\u81ea\u56de\u5f52\u6295\u673a\u89e3\u7801\u6846\u67b6<\/strong>\uff0c\u65e8\u5728\u589e\u5f3a draft model \u7684\u5e76\u884c\u6027\u548c\u8f93\u51fa\u8d28\u91cf\uff0c\u4ee5\u6709\u6548\u63d0\u5347\u5927\u6a21\u578b\u7684\u63a8\u7406\u901f\u5ea6\u3002Falcon \u53ef\u4ee5\u5b9e\u73b0\u7ea6 2.91-3.51 \u500d\u7684\u52a0\u901f\u6bd4\uff0c\u5728\u591a\u79cd\u6570\u636e\u96c6\u4e0a\u83b7\u5f97\u4e86\u5f88\u597d\u7684\u7ed3\u679c\uff0c\u5e76\u5df2\u5e94\u7528\u5230\u7ffc\u652f\u4ed8\u591a\u4e2a\u5b9e\u9645\u4e1a\u52a1\u4e2d\u3002<\/section>\n
<\/a><\/section>\n
\u8bba\u6587\u5730\u5740\uff1ahttps:\/\/arxiv.org\/pdf\/2412.12639<\/section>\n
<\/section>\n
1. \u7814\u7a76\u80cc\u666f<\/strong><\/section>\n
<\/section>\n
\u5927\u578b\u8bed\u8a00\u6a21\u578b (LLMs) \u5728\u5404\u79cd\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u5c55\u73b0\u4e86\u5353\u8d8a\u7684\u8868\u73b0\uff0c\u7136\u800c\u7531\u4e8e\u81ea\u56de\u5f52 (AR) \u89e3\u7801\u65b9\u5f0f\uff0cLLMs \u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\u4e5f\u9762\u4e34\u7740\u663e\u8457\u7684\u8ba1\u7b97\u5f00\u9500\u548c\u5ef6\u8fdf\u74f6\u9888\u3002<\/section>\n
<\/section>\n
\u4e3a\u6b64\uff0c\u7814\u7a76\u5b66\u8005\u63d0\u51fa Speculative Decoding (\u6295\u673a\u91c7\u6837) \u65b9\u6cd5\u3002Speculative Decoding \u4f1a\u9009\u62e9\u4e00\u4e2a\u6bd4\u539f\u59cb\u6a21\u578b (Target Model) \u8f7b\u91cf\u7684 LLM \u4f5c\u4e3a Draft Model\uff0c\u5728 Draft \u9636\u6bb5\u4f7f\u7528 Draft Model \u8fde\u7eed\u751f\u6210\u82e5\u5e72\u4e2a\u5019\u9009 Token\u3002\u5728 Verify \u9636\u6bb5\uff0c\u5c06\u5f97\u5230\u7684\u5019\u9009 Token \u5e8f\u5217\u653e\u5165\u5230\u539f\u59cb LLM \u505a\u9a8c\u8bc1 & Next Token \u751f\u6210\uff0c\u5b9e\u73b0\u5e76\u884c\u89e3\u7801\u3002\u901a\u8fc7\u5c06\u8ba1\u7b97\u8d44\u6e90\u5bfc\u5411\u4e8e\u9a8c\u8bc1\u9884\u5148\u751f\u6210\u7684 token\uff0cSpeculative Decoding \u5927\u5927\u51cf\u5c11\u4e86\u8bbf\u95ee LLM \u53c2\u6570\u6240\u9700\u7684\u5185\u5b58\u64cd\u4f5c\uff0c\u4ece\u800c\u63d0\u5347\u4e86\u6574\u4f53\u63a8\u7406\u6548\u7387\u3002<\/section>\n
<\/section>\n
\u73b0\u6709\u7684\u6295\u673a\u91c7\u6837\u4e3b\u8981\u91c7\u7528\u4e24\u79cd Draft \u7b56\u7565\uff1a\u81ea\u56de\u5f52 (AR) \u548c\u534a\u81ea\u56de\u5f52 (SAR) draft<\/strong>\u3002AR draft \u987a\u5e8f\u751f\u6210 token\uff0c\u6bcf\u4e2a token \u4f9d\u8d56\u4e8e\u524d\u9762\u7684 token\u3002\u8fd9\u79cd\u987a\u5e8f\u4f9d\u8d56\u6027\u9650\u5236\u4e86 draft \u6a21\u578b\u7684\u5e76\u884c\u6027\uff0c\u5bfc\u81f4\u663e\u8457\u7684\u65f6\u95f4\u5f00\u9500\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0cSAR draft \u540c\u65f6\u751f\u6210\u591a\u4e2a token\uff0c\u589e\u5f3a\u4e86 draft \u8fc7\u7a0b\u7684\u5e76\u884c\u5316\u3002\u7136\u800c\uff0cSAR draft \u7684\u4e00\u4e2a\u91cd\u8981\u5c40\u9650\u662f\u5b83\u65e0\u6cd5\u5b8c\u5168\u6355\u6349\u76f8\u540c block \u5185 draft tokens \u4e4b\u95f4\u7684\u76f8\u4e92\u4f9d\u8d56\u5173\u7cfb\uff0c\u53ef\u80fd\u5bfc\u81f4\u751f\u6210\u7684 token \u63a5\u53d7\u7387\u8f83\u4f4e\u3002<\/section>\n
<\/section>\n
\u56e0\u6b64\uff0c\u5728\u6295\u673a\u91c7\u6837\u4e2d\uff0c\u5e73\u8861\u4f4e draft \u5ef6\u8fdf\u4e0e\u9ad8\u63a8\u6d4b\u51c6\u786e\u6027\u4ee5\u52a0\u901f LLMs \u7684\u63a8\u7406\u901f\u5ea6\uff0c\u662f\u4e00\u4e2a\u91cd\u5927\u6311\u6218\u3002<\/section>\n
<\/section>\n
\u4e3a\u6b64\uff0c\u7ffc\u652f\u4ed8\u63d0\u51fa\u4e86 Falcon\uff0c\u4e00\u4e2a\u589e\u5f3a\u7684\u534a\u81ea\u56de\u5f52\uff08SAR\uff09\u6295\u673a\u89e3\u7801\u6846\u67b6\uff0c\u65e8\u5728\u589e\u5f3a draft model \u7684\u5e76\u884c\u6027\u548c\u8f93\u51fa\u8d28\u91cf\uff0c\u4ece\u800c\u63d0\u5347 LLMs \u7684\u63a8\u7406\u6548\u7387\u3002Falcon \u96c6\u6210\u4e86 Coupled Sequential Glancing Distillation\uff08CSGD\uff09\u65b9\u6cd5\uff0c\u63d0\u9ad8\u4e86 SAR draft model \u7684 token \u63a5\u53d7\u7387\u3002<\/section>\n
<\/section>\n
\u6b64\u5916\uff0cFalcon\u00a0\u8fd8\u8bbe\u8ba1\u4e86\u4e00\u79cd\u4e13\u95e8\u7684 decoding tree \u6765\u652f\u6301 SAR \u91c7\u6837\uff0c\u4f7f\u5f97 draft model \u53ef\u4ee5\u5728\u4e00\u6b21\u524d\u5411\u4f20\u64ad\u4e2d\u751f\u6210\u591a\u4e2a token\uff0c\u5e76\u4e14\u4e5f\u80fd\u591f\u652f\u6301\u591a\u6b21\u524d\u5411\u4f20\u64ad\u3002\u8fd9\u79cd\u8bbe\u8ba1\u6709\u6548\u63d0\u5347 LLMs \u5bf9 token \u7684\u63a5\u53d7\u7387\uff0c\u8fdb\u4e00\u6b65\u52a0\u5feb\u4e86\u63a8\u7406\u901f\u5ea6\u3002<\/section>\n
<\/section>\n
2.\u00a0<\/strong>\u7814\u7a76\u65b9\u6cd5<\/strong><\/section>\n
<\/section>\n
Falcon\u00a0\u7684\u67b6\u6784\u5982\u56fe 1 \u6240\u793a\uff0c\u53ef\u4ee5\u770b\u5230\uff0c\u8be5\u534a\u81ea\u56de\u5f52\u89e3\u7801\u6846\u67b6\u4e3b\u8981\u7531\u4e09\u4e2a\u7ec4\u4ef6\u6784\u6210\uff1aEmbedding Layer\u3001LM-Head\u00a0\u548c\u534a\u81ea\u56de\u5f52\u89e3\u7801 Head<\/strong>\u3002<\/section>\n
<\/a><\/section>\n
\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u56fe 1 Falcon \u6846\u67b6\u56fe<\/sup><\/em><\/section>\n
<\/section>\n
\u5177\u4f53\u6765\u8bb2\uff0cFalcon \u5c06\u4e00\u4e2a\u65f6\u95f4\u6b65\u957f\u4e4b\u524d\u7684\u8fde\u7eed\u7279\u5f81\u5e8f\u5217\u548c\u5f53\u524d token \u5e8f\u5217\u8fde\u63a5\u8d77\u6765\uff0c\u4ee5\u540c\u65f6\u9884\u6d4b\u63a5\u4e0b\u6765\u7684 k \u4e2a\u6807\u8bb0\u3002\u4f8b\u5982\uff0c\u5f53 k = 2 \u65f6\uff0cFalcon \u4f7f\u7528\u521d\u59cb\u7279\u5f81\u5e8f\u5217 (f1, f2) \u548c\u63d0\u524d\u4e00\u4e2a\u65f6\u95f4\u6b65\u957f\u7684\u6807\u8bb0\u5e8f\u5217 (t2, t3) \u6765\u9884\u6d4b\u7279\u5f81\u5e8f\u5217 (f3, f4)\u3002\u968f\u540e\uff0c\u5c06\u9884\u6d4b\u5f97\u5230\u7684\u7279\u5f81 (f3, f4) \u4e0e\u4e0b\u4e00\u4e2a\u6807\u8bb0\u5e8f\u5217 (t4, t5) \u8fde\u63a5\uff0c\u5f62\u6210\u65b0\u7684\u8f93\u5165\u5e8f\u5217\u3002\u8fd9\u4e2a\u65b0\u8f93\u5165\u5e8f\u5217\u7528\u4e8e\u9884\u6d4b\u540e\u7eed\u7684\u7279\u5f81\u5e8f\u5217 (f5, f6) \u548c\u6807\u8bb0\u5e8f\u5217 (t6, t7)\uff0c\u4ece\u800c\u4fc3\u8fdb draft \u8fc7\u7a0b\u7684\u7ee7\u7eed\u3002Draft model \u591a\u6b21 forward \u4e4b\u540e\u751f\u6210\u7684 token \u88ab\u7ec4\u7ec7\u6210\u6811\u7ed3\u6784\uff0c\u8f93\u5165\u5230\u5927\u6a21\u578b\u4e2d\u8fdb\u884c verify\uff0c\u901a\u8fc7 verify \u7684 token \u88ab\u5927\u6a21\u578b\u63a5\u6536\uff0c\u5e76\u57fa\u4e8e\u6b64\u57fa\u7840\u5f00\u59cb\u4e0b\u4e00\u4e2a\u5faa\u73af\u3002<\/section>\n
<\/section>\n
2.1 Coupled Sequential Glancing Distillation<\/strong><\/section>\n
<\/section>\n
\u5f53\u524d\u63a8\u6d4b\u89e3\u7801\u65b9\u6cd5\u7684\u51c6\u786e\u6027\u76f8\u5bf9\u8f83\u4f4e\uff0c\u4e3b\u8981\u539f\u56e0\u662f token \u4e4b\u95f4\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\u4e0d\u8db3\u3002CSGD \u901a\u8fc7\u7528\u771f\u5b9e token \u548c hidden states \u66ff\u6362\u4e00\u4e9b\u521d\u59cb\u9884\u6d4b\u6765\u6539\u5584\u8fd9\u4e00\u70b9\uff0c\u5c06\u6b63\u786e\u4fe1\u606f\u91cd\u65b0\u6ce8\u5165\u89e3\u7801\u8fc7\u7a0b\u4e2d\uff0c\u4ece\u800c\u63d0\u9ad8\u540e\u7eed\u9884\u6d4b\u7684\u51c6\u786e\u6027\u548c\u8fde\u8d2f\u6027\u3002\u6a21\u578b\u7ed3\u6784\u53ca\u8bad\u7ec3\u6d41\u7a0b\u5982\u4e0b\u56fe\uff1a<\/section>\n
<\/a><\/section>\n
\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0\u56fe 2 CGSD \u65b9\u6cd5\u793a\u610f\u56fe<\/sup><\/em><\/section>\n
<\/section>\n
\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u4e00\u4e2a\u65f6\u95f4\u6b65\u957f\u4e4b\u524d\u7684\u8fde\u7eed\u7279\u5f81\u5e8f\u5217\u548c\u5f53\u524d token \u5e8f\u5217\u8fde\u63a5\u8d77\u6765\uff0c\u5e76\u8f93\u5165\u5230 draft model \u4e2d\uff0c\u5f62\u6210\u4e00\u4e2a\u878d\u5408\u5e8f\u5217\uff0c\u5176\u7ef4\u5ea6\u4e3a (bs, seq_len, 2 * hidden_dim)\u3002<\/section>\n
<\/section>\n
draft model \u7531\u4e00\u4e2a\u6df7\u5408 Transformer \u7f51\u7edc\u7ec4\u6210\uff0c\u8be5\u7f51\u7edc\u5305\u62ec\u4e24\u5c42 LSTM\u3001Relaxed Causal-Masked \u591a\u5934\u6ce8\u610f\u529b\u673a\u5236\uff0c\u4ee5\u53ca MLP \u7f51\u7edc\u3002\u5176\u4e2d LSTM \u7f51\u7edc\u5c06\u878d\u5408\u5e8f\u5217\u7684\u7ef4\u5ea6\u51cf\u5c11\u5230 (bs, seq_len, hidden_dim)\uff0c\u5e76\u4fdd\u7559\u5173\u4e8e\u8fc7\u53bb token \u7684\u4fe1\u606f\uff0c\u4ece\u800c\u63d0\u9ad8\u6a21\u578b\u7684\u51c6\u786e\u6027\u3002Relaxed Causal-Masked \u591a\u5934\u6ce8\u610f\u529b\u673a\u5236\u80fd\u591f\u5728\u4fdd\u6301\u56e0\u679c\u5173\u7cfb\u7684\u540c\u65f6\uff0c\u4e13\u6ce8\u4e8e\u8f93\u5165\u5e8f\u5217\u7684\u76f8\u5173\u90e8\u5206\u3002MLP \u5c42\u8fdb\u4e00\u6b65\u5904\u7406\u8fd9\u4e9b\u4fe1\u606f\uff0c\u4ee5\u505a\u51fa\u6700\u7ec8\u9884\u6d4b\u3002<\/section>\n
<\/section>\n
\u5f53\u5e8f\u5217\u9996\u6b21\u901a\u8fc7 draft model \u540e\uff0c\u4f1a\u751f\u6210\u521d\u59cb\u7684 token \u9884\u6d4b<\/a>\u3002\u7136\u540e\uff0c\u6211\u4eec\u8ba1\u7b97 draft model \u7684\u9884\u6d4b\u4e0e\u771f\u5b9e token Y \u4e4b\u95f4\u7684\u6c49\u660e\u8ddd\u79bb\uff0c\u4ee5\u6b64\u6765\u8861\u91cf\u9884\u6d4b\u7684\u51c6\u786e\u6027\u3002\u63a5\u4e0b\u6765\uff0c\u6211\u4eec\u5c06\u4e00\u5b9a\u6570\u91cf\u8fde\u7eed\u9884\u6d4b\u7684 token \u5e8f\u5217<\/a>\u548c\u7279\u5f81\u5e8f\u5217<\/a>\u66ff\u6362\u4e3a\u6765\u81ea LLMs \u7684\u6b63\u786e token \u5e8f\u5217<\/a>\u548c\u7279\u5f81\u5e8f\u5217<\/a>\u3002<\/section>\n
<\/section>\n
CSGD \u4e0e\u4f20\u7edf\u7684 glancing \u65b9\u6cd5\u4e0d\u540c\uff0c\u540e\u8005\u4ec5\u968f\u673a\u66ff\u6362 token\u3002\u76f8\u53cd\uff0cCSGD \u9009\u62e9\u6027\u5730\u540c\u65f6\u66ff\u6362\u9884\u6d4b\u4e4b\u524d\u7684\u8fde\u7eed token \u548c\u7279\u5f81\u5e8f\u5217\uff0c\u5982\u56fe 2 \u4e2d\u865a\u7ebf\u6846\u6807\u6ce8\u7684 choice 1\u3001choice 2\u3001choice3 \u6240\u793a\u3002\u8fd9\u79cd\u65b9\u6cd5\u589e\u5f3a\u4e86\u5bf9 token \u95f4\u7684\u5173\u7cfb\u7684\u7406\u89e3\uff0c\u5e76\u786e\u4fdd draft model \u80fd\u591f\u6709\u6548\u5229\u7528\u63d0\u524d\u65f6\u95f4\u6b65\u957f\u7684 token \u5e8f\u5217\uff0c\u8fd9\u5728 SAR \u89e3\u7801\u4e2d\u5c24\u4e3a\u91cd\u8981\u3002\u968f\u540e\uff0c\u4fee\u6b63\u540e\u7684 token \u548c\u7279\u5f81\u5e8f\u5217\u88ab\u91cd\u65b0\u8f93\u5165\u5230 draft model \u4e2d\u4ee5\u8ba1\u7b97\u8bad\u7ec3\u635f\u5931\u3002<\/section>\n
<\/section>\n
\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u6211\u4eec\u91c7\u7528\u4e86\u77e5\u8bc6\u84b8\u998f\uff0c\u635f\u5931\u51fd\u6570\u5305\u62ec draft model \u7684\u8f93\u51fa\u7279\u5f81\u4e0e\u771f\u5b9e\u7279\u5f81\u4e4b\u95f4\u7684\u56de\u5f52\u635f\u5931\u4ee5\u53ca\u84b8\u998f\u635f\u5931\uff0c\u5177\u4f53\u7684\u635f\u5931\u51fd\u6570\u5982\u4e0b\uff1a<\/section>\n
<\/a><\/section>\n
2.2 Custom-Designed Decoding Tree<\/strong><\/section>\n
<\/section>\n
\u5f53\u524d\u57fa\u4e8e\u6811\u7684\u63a8\u6d4b\u89e3\u7801\u65b9\u6cd5\u901a\u8fc7\u5728\u6bcf\u4e2a\u8d77\u8349\u6b65\u9aa4\u751f\u6210\u591a\u4e2a draft token \u6765\u63d0\u5347\u63a8\u6d4b\u6548\u7387\u3002\u7136\u800c\uff0c\u8fd9\u4e9b\u65b9\u6cd5\u4ecd\u7136\u9700\u8981 draft model \u6309\u987a\u5e8f\u751f\u6210 token\uff0c\u8fd9\u9650\u5236\u4e86\u63a8\u6d4b\u6548\u7387\u7684\u8fdb\u4e00\u6b65\u63d0\u9ad8\u3002\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e00\u5c40\u9650\u6027\uff0cCDT (Custom-Designed Decoding Tree) \u652f\u6301 draft model \u5728\u4e00\u6b21\u524d\u5411\u4f20\u9012\u4e2d\u751f\u6210\u591a\u4e2a token (k \u4e2a)\uff0c\u5e76\u4e14\u5728\u6bcf\u4e2a draft \u6b65\u9aa4\u4e2d\u652f\u6301\u591a\u6b21\u524d\u5411\u4f20\u9012\u3002\u56e0\u6b64\uff0c\u4e0e\u73b0\u6709\u65b9\u6cd5\u76f8\u6bd4\uff0cCDT \u751f\u6210\u7684\u8349\u7a3f\u6807\u8bb0\u6570\u91cf\u662f\u5176 k \u500d\u3002<\/section>\n
<\/section>\n
Draft model \u591a\u6b21 forward \u4e4b\u540e\uff0c\u751f\u6210\u7684 token \u88ab\u7ec4\u7ec7\u6210\u6811\u7ed3\u6784\uff0c\u8f93\u5165\u5230\u5927\u6a21\u578b\u4e2d\u8fdb\u884c verify\u3002LLM \u4f7f\u7528\u57fa\u4e8e\u6811\u7684\u5e76\u884c\u89e3\u7801\u673a\u5236\u6765\u9a8c\u8bc1\u5019\u9009 token \u5e8f\u5217\u7684\u6b63\u786e\u6027\uff0c\u88ab\u63a5\u53d7\u7684 token \u53ca\u5176\u76f8\u5e94\u7684\u7279\u5f81\u5e8f\u5217\u4f1a\u5728\u540e\u7eed\u7ee7\u7eed\u8fdb\u884c\u524d\u5411\u4f20\u9012\u3002\u5728\u4f20\u7edf\u7684\u81ea\u56de\u5f52\uff08AR\uff09\u89e3\u7801\u4e2d\uff0c\u4f7f\u7528\u56e0\u679c\u63a9\u7801\uff0c\u5176\u7ed3\u6784\u4e3a\u4e0b\u4e09\u89d2\u77e9\u9635\u3002\u5b83\u786e\u4fdd\u4e86\u524d\u9762\u7684 token \u4e0d\u80fd\u8bbf\u95ee\u540e\u9762\u7684\u4fe1\u606f\u3002<\/section>\n
<\/section>\n
\u76f8\u6bd4\u4e4b\u4e0b\uff0cFalcon \u91c7\u7528\u4e86\u4e00\u79cd causal \u56e0\u679c\u63a9\u7801 (\u5982\u56fe 3 \u6240\u793a)\uff0c\u5141\u8bb8\u6a21\u578b\u8bbf\u95ee\u540c\u4e00 k*k \u7684 block \u5185\u7684 token \u4ee5\u53ca\u76f8\u5e94\u7684\u4e4b\u524d\u7684\u8fde\u7eed token\u3002\u8fd9\u4e00\u589e\u5f3a\u663e\u8457\u63d0\u9ad8\u4e86 drafter \u751f\u6210 token \u7684\u6548\u7387\uff0c\u4f7f LLM \u80fd\u591f\u540c\u65f6\u9a8c\u8bc1\u66f4\u591a\u7684 token\uff0c\u4ece\u800c\u52a0\u5feb\u4e86 LLM \u7684\u6574\u4f53\u63a8\u7406\u901f\u5ea6\u3002<\/section>\n
<\/a><\/section>\n
\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0\u56fe 3 Custom-Designed Decoding Tree \u65b9\u6cd5\u793a\u610f\u56fe<\/sup><\/em><\/section>\n
<\/section>\n
3.\u00a0<\/strong>\u5b9e\u9a8c\u7ed3\u679c<\/strong><\/section>\n
<\/section>\n
\u6211\u4eec\u5728\u591a\u4e2a\u6570\u636e\u96c6\u548c\u591a\u4e2a\u6a21\u578b\u4e0a\u8fdb\u884c\u4e86\u5e7f\u6cdb\u7684\u5b9e\u9a8c\uff0c\u9a8c\u8bc1\u4e86\u672c\u6587\u65b9\u6cd5\u7684\u6709\u6548\u6027\u3002\u548c\u73b0\u6709\u7684\u65b9\u6cd5\u76f8\u6bd4\uff0cFalcon \u5c55\u73b0\u4e86\u4f18\u8d8a\u7684\u6027\u80fd\uff0c\u5177\u4f53\u5982\u4e0b\u56fe\uff1a<\/section>\n
<\/a><\/section>\n
\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u56fe 4 Falcon \u5b9e\u9a8c\u7ed3\u679c\u56fe<\/sup><\/em><\/section>\n
<\/section>\n
4.\u00a0<\/strong>\u4e1a\u52a1\u6f5c\u529b<\/strong><\/section>\n
<\/section>\n
Falcon \u5927\u6a21\u578b\u53ef\u4ee5\u5b9e\u73b0\u7ea6 2.91-3.51 \u500d\u7684\u52a0\u901f\u6bd4\uff0c\u76f8\u5f53\u4e8e\u540c\u7b49\u6761\u4ef6\u4e0b\u63a8\u7406\u6210\u672c\u4e0b\u964d\u81f3\u7ea6\u539f\u5148\u7684 1\/3\uff0c\u4ece\u800c\u5927\u5e45\u964d\u4f4e\u4e86\u5927\u6a21\u578b\u63a8\u7406\u8ba1\u7b97\u76f8\u5173\u6210\u672c\u3002<\/section>\n
<\/section>\n
\u5f53\u524d\uff0cFalcon \u6280\u672f\u5df2\u8f6c\u5316\u81f3\u7ffc\u652f\u4ed8\u5927\u6a21\u578b\u4ea7\u54c1 InsightAI \u5e73\u53f0\uff0c\u5e76\u5df2\u670d\u52a1\u8bf8\u5982\u7ffc\u652f\u4ed8\u6570\u5b57\u4eba\u5ba2\u670d\u3001\u501f\u94b1-\u7ffc\u5c0f\u6a59\u3001\u4eba\u529b-\u7ffc\u70b9\u901a\u3001\u8d22\u52a1-\u7ffc\u5c0f\u8d22\u7b49\u591a\u4e2a\u4e1a\u52a1\u5e94\u7528\u3002<\/section>\n
<\/section>\n
5.\u00a0<\/strong>\u603b\u7ed3<\/strong><\/section>\n
<\/section>\n
\u6295\u673a\u91c7\u6837\u662f\u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u7684\u4e00\u4e2a\u6838\u5fc3\u65b9\u6cd5\u3002\u5f53\u524d\uff0c\u4e3b\u8981\u7684\u6311\u6218\u662f\u5982\u4f55\u63d0\u5347 draft model \u7684\u51c6\u786e\u7387\u3001\u91c7\u6837\u6548\u7387\uff0c\u5e76\u63d0\u5347\u5927\u6a21\u578b\u7684\u9a8c\u8bc1\u6548\u7387\u3002\u6587\u7ae0\u63d0\u51fa\u4e86 Falcon \u65b9\u6cd5\uff0c\u4e00\u79cd\u57fa\u4e8e\u589e\u5f3a\u534a\u81ea\u56de\u5f52\u6295\u673a\u89e3\u7801\u6846\u67b6\u3002Falcon \u901a\u8fc7 CSGD \u8fd9\u79cd\u8bad\u7ec3\u65b9\u6cd5\u4ee5\u53ca\u534a\u81ea\u56de\u5f52\u7684\u6a21\u578b\u8bbe\u8ba1\uff0c\u663e\u8457\u63d0\u5347\u4e86 draft model \u7684\u9884\u6d4b\u51c6\u786e\u7387\u4ee5\u53ca\u91c7\u6837\u6548\u7387\u3002\u6b64\u5916\uff0c\u4e3a\u4e86\u8ba9\u5927\u6a21\u578b\u80fd\u9a8c\u8bc1\u66f4\u591a\u7684 token\uff0c\u672c\u6587\u7cbe\u5fc3\u8bbe\u8ba1\u4e86\u4e00\u4e2a decoding tree\uff0c\u6709\u6548\u63d0\u5347\u4e86 draft model \u7684\u6548\u7387\uff0c\u4ece\u800c\u63d0\u5347\u4e86\u9a8c\u8bc1\u6548\u7387\u3002Falcon \u5728\u591a\u79cd\u6570\u636e\u96c6\u4e0a\u53ef\u4ee5\u5b9e\u73b0\u7ea6 2.91-3.51x \u7684\u52a0\u901f\u6bd4\u5e76\u5e94\u7528\u5230\u7ffc\u652f\u4ed8\u7684\u4f17\u591a\u4e1a\u52a1\u4e2d\uff0c\u83b7\u5f97\u4e86\u5f88\u597d\u7684\u6548\u679c\u3002<\/section>\n
<\/section>\n
6. \u516c\u53f8\u7b80\u4ecb<\/strong><\/section>\n
<\/section>\n
\u5929\u7ffc\u7535\u5b50\u5546\u52a1\u6709\u9650\u516c\u53f8\uff08\u7ffc\u652f\u4ed8\uff09\u662f\u4e2d\u56fd\u7535\u4fe1\u96c6\u56e2\u6709\u9650\u516c\u53f8\u6210\u5458\u4f01\u4e1a\u3002\u516c\u53f8\u575a\u6301\u4ee5\u79d1\u6280\u521b\u65b0\u5f15\u9886\u6218\u7565\u65b0\u5174\u4e1a\u52a1\u53d1\u5c55\uff0c\u8fd0\u7528\u5927\u6570\u636e\u3001\u4eba\u5de5\u667a\u80fd\u3001\u533a\u5757\u94fe\u7b49\u5173\u952e\u6280\u672f\uff0c\u79ef\u6781\u63a2\u7d22\u652f\u4ed8 + AI \u670d\u52a1\u4f53\u9a8c\uff0c\u5c06 AI \u5e94\u7528\u4e8e\u652f\u4ed8\u670d\u52a1\u3001\u667a\u80fd\u5ba2\u670d\u3001\u98ce\u9669\u7ba1\u7406\u7b49\u591a\u4e2a\u65b9\u9762\uff0c\u4ee5\u66f4\u597d\u63a8\u52a8\u4ea7\u4e1a\u6570\u5b57\u5316\u8f6c\u578b\u5347\u7ea7\uff0c\u52a9\u529b\u6570\u5b57\u4e2d\u56fd\u5efa\u8bbe\u53d1\u5c55\u3002<\/section>\n

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:AAAI 2025 | \u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u65b0\u8303\u5f0f\uff1a\u52a0\u901f\u6bd4\u9ad8\u8fbe3.51\u500d\u3001\u6210\u672c\u964d\u81f31\/3<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:AAAI 2025 | […]<\/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],"class_list":["post-37658","post","type-post","status-publish","format-standard","hentry","category-news","tag-68"],"yoast_head":"\nAAAI 2025 | \u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u65b0\u8303\u5f0f\uff1a\u52a0\u901f\u6bd4\u9ad8\u8fbe3.51\u500d\u3001\u6210\u672c\u964d\u81f31\/3 - \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=37658\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=37658#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=37658\"},\"author\":{\"name\":\"AI\u5c0f\u52a9\u624b\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"headline\":\"AAAI 2025 | \u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u65b0\u8303\u5f0f\uff1a\u52a0\u901f\u6bd4\u9ad8\u8fbe3.51\u500d\u3001\u6210\u672c\u964d\u81f31\\\/3\",\"datePublished\":\"2025-02-01T16:00:55+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=37658\"},\"wordCount\":276,\"commentCount\":0,\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=37658#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2025\\\/01\\\/frc-962f41f7888e8f21de6eba24e156e28d.png\",\"keywords\":[\"\u5927\u6a21\u578b\"],\"articleSection\":[\"\u884c\u4e1a\u8d44\u8baf\"],\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=37658#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=37658\",\"url\":\"https:\\\/\\\/17aitech.com\\\/?p=37658\",\"name\":\"AAAI 2025 | \u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u65b0\u8303\u5f0f\uff1a\u52a0\u901f\u6bd4\u9ad8\u8fbe3.51\u500d\u3001\u6210\u672c\u964d\u81f31\\\/3 - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=37658#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=37658#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2025\\\/01\\\/frc-962f41f7888e8f21de6eba24e156e28d.png\",\"datePublished\":\"2025-02-01T16:00:55+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=37658#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=37658\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=37658#primaryimage\",\"url\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2025\\\/01\\\/frc-962f41f7888e8f21de6eba24e156e28d.png\",\"contentUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2025\\\/01\\\/frc-962f41f7888e8f21de6eba24e156e28d.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=37658#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\\\/\\\/17aitech.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"AAAI 2025 | \u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u65b0\u8303\u5f0f\uff1a\u52a0\u901f\u6bd4\u9ad8\u8fbe3.51\u500d\u3001\u6210\u672c\u964d\u81f31\\\/3\"}]},{\"@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\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"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":"AAAI 2025 | \u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u65b0\u8303\u5f0f\uff1a\u52a0\u901f\u6bd4\u9ad8\u8fbe3.51\u500d\u3001\u6210\u672c\u964d\u81f31\/3 - \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=37658","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/17aitech.com\/?p=37658#article","isPartOf":{"@id":"https:\/\/17aitech.com\/?p=37658"},"author":{"name":"AI\u5c0f\u52a9\u624b","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"headline":"AAAI 2025 | \u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u65b0\u8303\u5f0f\uff1a\u52a0\u901f\u6bd4\u9ad8\u8fbe3.51\u500d\u3001\u6210\u672c\u964d\u81f31\/3","datePublished":"2025-02-01T16:00:55+00:00","mainEntityOfPage":{"@id":"https:\/\/17aitech.com\/?p=37658"},"wordCount":276,"commentCount":0,"image":{"@id":"https:\/\/17aitech.com\/?p=37658#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-962f41f7888e8f21de6eba24e156e28d.png","keywords":["\u5927\u6a21\u578b"],"articleSection":["\u884c\u4e1a\u8d44\u8baf"],"inLanguage":"zh-Hans","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/17aitech.com\/?p=37658#respond"]}]},{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=37658","url":"https:\/\/17aitech.com\/?p=37658","name":"AAAI 2025 | \u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u65b0\u8303\u5f0f\uff1a\u52a0\u901f\u6bd4\u9ad8\u8fbe3.51\u500d\u3001\u6210\u672c\u964d\u81f31\/3 - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=37658#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=37658#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-962f41f7888e8f21de6eba24e156e28d.png","datePublished":"2025-02-01T16:00:55+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=37658#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=37658"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=37658#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-962f41f7888e8f21de6eba24e156e28d.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-962f41f7888e8f21de6eba24e156e28d.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=37658#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"AAAI 2025 | \u5927\u6a21\u578b\u63a8\u7406\u52a0\u901f\u65b0\u8303\u5f0f\uff1a\u52a0\u901f\u6bd4\u9ad8\u8fbe3.51\u500d\u3001\u6210\u672c\u964d\u81f31\/3"}]},{"@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":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","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\/37658","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=37658"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/37658\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=37658"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=37658"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=37658"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}