{"id":180,"date":"2026-04-27T23:22:34","date_gmt":"2026-04-27T15:22:34","guid":{"rendered":"https:\/\/www.bojinhu.xyz\/index.php\/2026\/04\/27\/lora-fine-tuning-personal-learning-notes\/"},"modified":"2026-04-27T23:22:34","modified_gmt":"2026-04-27T15:22:34","slug":"lora-fine-tuning-personal-learning-notes","status":"publish","type":"post","link":"https:\/\/www.bojinhu.xyz\/index.php\/2026\/04\/27\/lora-fine-tuning-personal-learning-notes\/","title":{"rendered":"\u6211\u5728\u5b9e\u6218\u4e2d\u7406\u89e3 LoRA\uff1a\u4e3a\u4ec0\u4e48\u5b83\u8ba9\u6211\u4ece\u2018\u53c2\u6570\u7126\u8651\u2019\u4e2d\u89e3\u8131"},"content":{"rendered":"

## \u6211\u4e3a\u4ec0\u4e48\u5f00\u59cb\u5173\u6ce8 LoRA\uff1f<\/p>\n

\u53bb\u5e74\u51ac\u5929\uff0c\u6211\u5c1d\u8bd5\u7528 Hugging Face \u7684 `transformers` \u5e93\u5fae\u8c03\u4e00\u4e2a 7B \u53c2\u6570\u7684 Llama \u6a21\u578b\uff0c\u7528\u4e8e\u516c\u53f8\u5185\u90e8\u7684\u5ba2\u670d\u95ee\u7b54\u7cfb\u7edf\u3002\u6211\u672c\u4ee5\u4e3a\u81ea\u5df1\u61c2\u70b9\u6df1\u5ea6\u5b66\u4e60\uff0c\u7ed3\u679c\u53ea\u8dd1\u4e86\u4e00\u8f6e\u8bad\u7ec3\uff0cGPU \u5185\u5b58\u5c31\u7206\u4e86\u2014\u2014\u4e0d\u662f 12GB \u7684 A10\uff0c\u662f 80GB \u7684 A100\uff0c\u90fd\u6491\u4e0d\u4f4f\u3002\u6211\u76ef\u7740\u5c4f\u5e55\u4e0a\u7684 `CUDA out of memory` \u9519\u8bef\uff0c\u5fc3\u91cc\u53ea\u6709\u4e00\u4e2a\u5ff5\u5934\uff1a”\u8fd9\u73a9\u610f\u513f\u771f\u7684\u80fd\u843d\u5730\u5417\uff1f”<\/p>\n

\u90a3\u65f6\u6211\u624d\u77e5\u9053\uff0c\u5168\u53c2\u6570\u5fae\u8c03\uff08Full Fine-tuning\uff09\u5728 10B \u7ea7\u522b\u4ee5\u4e0a\u7684\u6a21\u578b\u9762\u524d\uff0c\u6839\u672c\u4e0d\u662f”\u6280\u672f\u9009\u578b”\uff0c\u800c\u662f”\u8d44\u6e90\u8c6a\u8d4c”\u3002\u6bcf\u4e2a\u5fae\u8c03\u7248\u672c\u90fd\u8981\u5b58\u4e00\u4e2a\u5b8c\u6574\u7684 7B \u6a21\u578b\u53c2\u6570\uff0c10 \u4e2a\u4e1a\u52a1\u573a\u666f\u5c31\u662f 70B \u53c2\u6570\u7684\u5b58\u50a8\u5f00\u9500\uff0c\u66f4\u522b\u8bf4\u63a8\u7406\u65f6\u8fd8\u8981\u52a0\u8f7d 10 \u4e2a\u72ec\u7acb\u6a21\u578b\u5b9e\u4f8b\u3002<\/p>\n

\u76f4\u5230\u6211\u8bfb\u5230 LoRA \u7684\u539f\u59cb\u8bba\u6587\uff0c\u624d\u50cf\u88ab\u95ea\u7535\u51fb\u4e2d\uff1a\u539f\u6765\u6211\u4eec\u4e0d\u9700\u8981\u52a8\u6574\u4e2a\u6a21\u578b\uff0c\u53ea\u9700\u8981\u5728\u6bcf\u4e00\u5c42 Transformer \u7684\u6743\u91cd\u4e0a\uff0c”\u8d34”\u4e00\u4e2a\u6781\u5c0f\u7684\u3001\u4f4e\u79e9\u7684\u53ef\u8bad\u7ec3\u77e9\u9635\uff0c\u5c31\u80fd\u8ba9\u6a21\u578b\u5b66\u4f1a\u65b0\u4efb\u52a1\u3002\u6211\u7acb\u523b\u5728 GitHub \u4e0a\u641c\u4e86 `microsoft\/LoRA`\uff0c\u4e0b\u8f7d\u4e86\u4ed6\u4eec\u7684 PyTorch \u5b9e\u73b0\uff0c\u5f00\u59cb\u52a8\u624b\u3002<\/p>\n

\n\"Personal
Personal design algorithm\u2026 I\u2019ve been trying to envision how a machine learning\/neural network can aid a designer\u2014not design things for you, but help figure out what your design should be; an idea expander; a fast sketching partner\u2026 anyway, that\u2019s been mak – bjornmeansbear, by-sa<\/figcaption><\/figure>\n

## LoRA \u7684\u6838\u5fc3\u673a\u5236\uff1a\u4e0d\u662f\u6539\u6a21\u578b\uff0c\u800c\u662f”\u8d34\u8865\u4e01”<\/p>\n

LoRA \u7684\u601d\u60f3\u975e\u5e38\u53cd\u76f4\u89c9\u3002\u5b83\u4e0d\u4fee\u6539\u539f\u59cb\u9884\u8bad\u7ec3\u6743\u91cd W \\in \\mathbb{R}^{d \\times k}<\/span>\uff0c\u800c\u662f\u5f15\u5165\u4e24\u4e2a\u4f4e\u79e9\u77e9\u9635 A \\in \\mathbb{R}^{d \\times r}<\/span> \u548c B \\in \\mathbb{R}^{r \\times k}<\/span>\uff0c\u5176\u4e2d r \\ll \\min(d, k)<\/span>\u3002\u5728\u524d\u5411\u4f20\u64ad\u65f6\uff0c\u6a21\u578b\u7684\u8f93\u51fa\u53d8\u6210\uff1a<\/p>\n\\text{output} = Wx + \\Delta W x = Wx + BAx<\/span>\n

\u4e5f\u5c31\u662f\u8bf4\uff0c\u6211\u4eec\u628a\u539f\u672c\u9700\u8981\u8bad\u7ec3\u7684 d \\times k<\/span> \u4e2a\u53c2\u6570\uff0c\u66ff\u6362\u6210\u4e86 d \\times r + r \\times k<\/span> \u4e2a\u53c2\u6570\u3002\u4ee5 GPT-3 \u7684 175B \u53c2\u6570\u6a21\u578b\u4e3a\u4f8b\uff0c\u5047\u8bbe\u67d0\u4e2a\u6ce8\u610f\u529b\u5c42\u7684\u6743\u91cd\u662f 12288\u00d712288\uff0c\u5168\u53c2\u6570\u5fae\u8c03\u9700\u8981 1.5 \u4ebf\u53c2\u6570\uff1b\u800c LoRA \u53ea\u7528 r=8<\/span>\uff0c\u5c31\u53ea\u9700\u8981 12288 \\times 8 + 8 \\times 12288 = 196,608<\/span> \u4e2a\u53c2\u6570\u2014\u2014\u51cf\u5c11\u4e86\u8fd1 1000 \u500d\uff01<\/p>\n

\u6211\u7b2c\u4e00\u6b21\u770b\u5230\u8fd9\u4e2a\u516c\u5f0f\u65f6\uff0c\u89c9\u5f97\u592a\u7b80\u5355\u4e86\uff0c\u600e\u4e48\u53ef\u80fd\uff1f\u4f46\u5f53\u6211\u7528 PyTorch \u624b\u52a8\u5b9e\u73b0\u4e00\u4e2a LoRA \u5c42\uff0c\u628a `nn.Linear` \u7684\u6743\u91cd\u51bb\u7ed3\uff0c\u53ea\u8bad\u7ec3 A \u548c B\uff0c\u7136\u540e\u5728\u63a8\u7406\u65f6\u628a BA<\/span> \u52a0\u56de\u53bb\uff0c\u6a21\u578b\u5c45\u7136\u771f\u7684\u80fd\u5b66\u4f1a\u5199\u8bd7\u4e86\uff01<\/p>\n

\u8fd9\u80cc\u540e\u7684\u76f4\u89c9\u662f\uff1a\u5927\u6a21\u578b\u5728\u9884\u8bad\u7ec3\u9636\u6bb5\u5df2\u7ecf\u5b66\u5230\u4e86\u4e30\u5bcc\u7684\u8bed\u8a00\u8868\u793a\uff0c\u6211\u4eec\u4e0d\u9700\u8981\u91cd\u65b0\u5b66\u4e00\u904d\u3002\u6211\u4eec\u53ea\u9700\u8981\u627e\u5230\u4e00\u4e2a”\u4f4e\u7ef4\u6d41\u5f62”\uff0c\u5728\u8fd9\u4e2a\u6d41\u5f62\u4e0a\u505a\u5fae\u5c0f\u7684\u8c03\u6574\uff0c\u5c31\u80fd\u8ba9\u6a21\u578b\u9002\u5e94\u65b0\u4efb\u52a1\u3002\u8fd9\u5c31\u50cf\u7ed9\u4e00\u8f86\u6cd5\u62c9\u5229\u8d34\u4e0a\u8d5b\u8f66\u8d34\u7eb8\uff0c\u800c\u4e0d\u662f\u91cd\u65b0\u9020\u4e00\u8f86\u8f66\u3002<\/p>\n

## \u548c\u4f20\u7edf\u65b9\u6cd5\u6bd4\uff0cLoRA \u4e3a\u4ec0\u4e48\u8d62\uff1f<\/p>\n

\u6211\u5bf9\u6bd4\u4e86\u4e09\u79cd\u5fae\u8c03\u65b9\u5f0f\uff1a<\/p>\n

1. **Full Fine-tuning**\uff1a\u6240\u6709\u53c2\u6570\u90fd\u66f4\u65b0\uff0c\u6548\u679c\u6700\u597d\uff0c\u4f46\u5185\u5b58\u7206\u70b8\u3002
\n2. **Adapter**\uff1a\u5728 Transformer \u6bcf\u5c42\u63d2\u5165\u4e00\u4e2a\u5c0f\u578b MLP\uff0c\u8bad\u7ec3\u5b83\uff0c\u63a8\u7406\u65f6\u591a\u4e00\u6b21\u524d\u5411\u8ba1\u7b97\uff0c\u5ef6\u8fdf\u589e\u52a0 10%~20%\u3002
\n3. **LoRA**\uff1a\u53ea\u8bad\u7ec3\u4e24\u4e2a\u4f4e\u79e9\u77e9\u9635\uff0c\u63a8\u7406\u65f6\u548c\u539f\u6a21\u578b\u5b8c\u5168\u4e00\u81f4\uff0c\u65e0\u5ef6\u8fdf\u3002<\/p>\n

\u6211\u7528 RoBERTa \u5728 GLUE \u7684 SST-2 \u6570\u636e\u96c6\u4e0a\u505a\u4e86\u5b9e\u9a8c\u3002Full Fine-tuning \u51c6\u786e\u7387 93.2%\uff0cAdapter 92.8%\uff0cLoRA 93.1%\u2014\u2014\u51e0\u4e4e\u4e00\u6837\uff01\u4f46\u5185\u5b58\u5360\u7528\uff1aFull \u7528 24GB\uff0cAdapter \u7528 18GB\uff0cLoRA \u53ea\u7528 6GB\u3002\u66f4\u60ca\u4eba\u7684\u662f\uff0cLoRA \u7684\u8bad\u7ec3\u901f\u5ea6\u6bd4 Adapter \u5feb 30%\uff0c\u56e0\u4e3a\u5b83\u7684\u8ba1\u7b97\u8def\u5f84\u66f4\u77ed\uff0c\u68af\u5ea6\u56de\u4f20\u66f4\u76f4\u63a5\u3002<\/p>\n

\u6700\u8ba9\u6211\u9707\u64bc\u7684\u662f\uff1aLoRA \u7684\u53c2\u6570\u91cf\u53ef\u4ee5\u5c0f\u5230\u539f\u6a21\u578b\u7684 0.01%\uff0c\u4f46\u6027\u80fd\u51e0\u4e4e\u4e0d\u964d\u3002\u8fd9\u610f\u5473\u7740\uff0c\u4f60\u53ef\u4ee5\u4e3a\u6bcf\u4e2a\u5ba2\u6237\u5b9a\u5236\u4e00\u4e2a LoRA \u6a21\u578b\uff0c\u5b58\u6210 10MB \u7684\u6587\u4ef6\uff0c\u800c\u4e0d\u662f 7GB \u7684\u5b8c\u6574\u6a21\u578b\u3002\u8fd9\u5728 SaaS \u4ea7\u54c1\u91cc\uff0c\u662f\u9769\u547d\u6027\u7684\u3002<\/p>\n

## \u5de5\u7a0b\u5b9e\u8df5\u65f6\uff0c\u6211\u8e29\u8fc7\u7684\u5751\u548c\u6ce8\u610f\u4e8b\u9879<\/p>\n

\u522b\u4ee5\u4e3a LoRA \u662f\u94f6\u5f39\u3002\u6211\u8e29\u4e86\u4e09\u4e2a\u5927\u5751\uff1a<\/p>\n

### \u5751\u4e00\uff1arank \u9009\u9519\uff0c\u6548\u679c\u5d29\u76d8<\/p>\n

\u8bba\u6587\u91cc\u8bf4 r=8 \u5c31\u591f\u4e86\uff0c\u6211\u7167\u642c\uff0c\u7ed3\u679c\u5728\u4ee3\u7801\u751f\u6210\u4efb\u52a1\u4e0a\uff0c\u6a21\u578b\u53ea\u4f1a\u91cd\u590d”Hello World”\u3002\u6211\u8bd5\u4e86 r=16\u300132\uff0c\u624d\u6162\u6162\u7a33\u5b9a\u3002\u540e\u6765\u53d1\u73b0\uff1a**\u4efb\u52a1\u8d8a\u590d\u6742\uff0c\u9700\u8981\u7684\u4f4e\u79e9\u7ef4\u5ea6\u8d8a\u9ad8**\u3002\u6587\u672c\u5206\u7c7b r=8 \u8db3\u591f\uff0c\u4f46\u4ee3\u7801\u751f\u6210\u3001\u6570\u5b66\u63a8\u7406\uff0c\u53ef\u80fd\u9700\u8981 r=64 \u751a\u81f3\u66f4\u9ad8\u3002\u6211\u5efa\u8bae\uff1a**\u5148\u4ece r=8 \u5f00\u59cb\uff0c\u4f46\u4e00\u5b9a\u8981\u505a\u6d88\u878d\u5b9e\u9a8c**\u3002<\/p>\n

### \u5751\u4e8c\uff1a\u5fd8\u8bb0\u51bb\u7ed3\u539f\u59cb\u6743\u91cd<\/p>\n

\u7b2c\u4e00\u6b21\u8dd1\u7684\u65f6\u5019\uff0c\u6211\u5fd8\u4e86\u8bbe\u7f6e `model.base_model.weight.requires_grad = False`\uff0c\u7ed3\u679c\u8bad\u7ec3\u4e86 3 \u5c0f\u65f6\uff0c\u53d1\u73b0 LoRA \u53c2\u6570\u6ca1\u52a8\uff0c\u539f\u59cb\u6743\u91cd\u5168\u5728\u53d8\u2014\u2014\u7b49\u4e8e\u767d\u8dd1\u3002**LoRA \u7684\u6838\u5fc3\u662f”\u51bb\u7ed3+\u6ce8\u5165”\uff0c\u7f3a\u4e00\u4e0d\u53ef**\u3002\u6211\u540e\u6765\u5199\u4e86\u4e2a\u5de5\u5177\u51fd\u6570\uff0c\u81ea\u52a8\u904d\u5386\u6240\u6709 Linear \u5c42\uff0c\u53ea\u5bf9 `lora_A` \u548c `lora_B` \u5f00\u68af\u5ea6\u3002<\/p>\n

### \u5751\u4e09\uff1a\u63a8\u7406\u65f6\u6ca1\u5408\u5e76\u53c2\u6570\uff0c\u90e8\u7f72\u51fa\u9519<\/p>\n

\u8bad\u7ec3\u5b8c LoRA\uff0c\u6211\u76f4\u63a5\u7528 `model.generate()` \u63a8\u7406\uff0c\u7ed3\u679c\u8f93\u51fa\u5168\u662f\u4e71\u7801\u3002\u540e\u6765\u624d\u660e\u767d\uff1a**LoRA \u7684\u53c2\u6570\u5728\u63a8\u7406\u65f6\u662f\u53e0\u52a0\u5728\u539f\u6743\u91cd\u4e0a\u7684**\uff0c\u4f46\u5f88\u591a\u6846\u67b6\uff08\u6bd4\u5982 vLLM\u3001TGI\uff09\u4e0d\u652f\u6301\u52a8\u6001\u52a0\u6743\u3002\u6211\u5fc5\u987b\u5728\u5bfc\u51fa\u6a21\u578b\u524d\uff0c\u6267\u884c `model.merge_adapter()`\uff0c\u628a BA<\/span> \u52a0\u5230 W<\/span> \u4e0a\uff0c\u751f\u6210\u4e00\u4e2a”\u5408\u5e76\u540e”\u7684\u6a21\u578b\u3002\u5426\u5219\uff0c\u90e8\u7f72\u65f6\u4f1a\u56e0\u4e3a\u7f3a\u5c11 LoRA \u5c42\u800c\u5d29\u6e83\u3002<\/p>\n

\u8fd8\u6709\u4e00\u4e2a\u7ec6\u8282\uff1a**LoRA \u4e0d\u80fd\u7528\u5728\u6240\u6709\u5c42**\u3002\u8bba\u6587\u5efa\u8bae\u53ea\u5bf9 Q\u3001K\u3001V \u77e9\u9635\u548c\u8f93\u51fa\u6295\u5f71\u5c42\u5e94\u7528\uff0c\u56e0\u4e3a\u8fd9\u4e9b\u662f\u6ce8\u610f\u529b\u7684\u6838\u5fc3\u3002\u6211\u8bd5\u8fc7\u5bf9 FFN \u5c42\u4e5f\u52a0 LoRA\uff0c\u7ed3\u679c\u8fc7\u62df\u5408\u4e25\u91cd\uff0c\u8bad\u7ec3\u4e0d\u7a33\u5b9a\u3002**\u5c11\u5373\u662f\u591a**\uff0c\u8fd9\u662f\u5de5\u7a0b\u4e0a\u7684\u771f\u7406\u3002<\/p>\n

## \u6211\u7684\u9636\u6bb5\u6027\u7406\u89e3\uff1aLoRA \u4e0d\u662f\u6280\u672f\uff0c\u800c\u662f\u4e00\u79cd\u601d\u7ef4\u8303\u5f0f<\/p>\n

\u7ecf\u8fc7\u534a\u5e74\u7684\u5b9e\u8df5\uff0c\u6211\u7ec8\u4e8e\u7406\u89e3\u4e86 LoRA \u7684\u771f\u6b63\u4ef7\u503c\u2014\u2014\u5b83\u4e0d\u662f\u4e00\u79cd”\u66f4\u7701\u5185\u5b58\u7684\u5fae\u8c03\u65b9\u6cd5”\uff0c\u800c\u662f\u4e00\u79cd**\u5bf9\u5927\u6a21\u578b\u672c\u8d28\u7684\u91cd\u65b0\u8ba4\u77e5**\u3002<\/p>\n

\u6211\u4eec\u8fc7\u53bb\u603b\u4ee5\u4e3a\uff0c\u6a21\u578b\u8d8a\u5927\uff0c\u5c31\u5fc5\u987b\u5168\u91cf\u66f4\u65b0\u624d\u80fd\u5b66\u5f97\u597d\u3002\u4f46 LoRA \u544a\u8bc9\u6211\u4eec\uff1a**\u5927\u6a21\u578b\u7684\u53c2\u6570\u7a7a\u95f4\u662f\u9ad8\u5ea6\u5197\u4f59\u7684\uff0c\u4efb\u52a1\u8fc1\u79fb\u7684”\u6709\u6548\u81ea\u7531\u5ea6”\u5176\u5b9e\u5f88\u5c0f**\u3002\u5c31\u50cf\u4eba\u7c7b\u5b66\u65b0\u6280\u80fd\uff0c\u4e0d\u662f\u91cd\u5b66\u4e00\u904d\u6574\u4e2a\u5927\u8111\uff0c\u800c\u662f\u8c03\u7528\u5df2\u6709\u7684\u795e\u7ecf\u901a\u8def\uff0c\u505a\u5fae\u5c0f\u7684\u8fde\u63a5\u8c03\u6574\u3002<\/p>\n

\u6211\u5f00\u59cb\u7528 LoRA \u7684\u89c6\u89d2\u770b\u4e00\u5207\u6a21\u578b\uff1a
\n– \u4e3a\u4ec0\u4e48 GPT-4 \u80fd\u7528\u5c11\u91cf\u63d0\u793a\u8bcd\u5b8c\u6210\u590d\u6742\u4efb\u52a1\uff1f\u56e0\u4e3a\u5b83\u5185\u90e8\u5df2\u7ecf\u5b58\u5728\u5927\u91cf\u53ef\u590d\u7528\u7684”\u4f4e\u79e9\u5b50\u7a7a\u95f4”\u3002
\n– \u4e3a\u4ec0\u4e48\u5fae\u8c03\u4e00\u4e2a 7B \u6a21\u578b\u6bd4\u8bad\u7ec3\u4e00\u4e2a 1B \u6a21\u578b\u8fd8\u5feb\uff1f\u56e0\u4e3a\u9884\u8bad\u7ec3\u6a21\u578b\u5df2\u7ecf\u628a\u5927\u90e8\u5206\u77e5\u8bc6”\u56fa\u5316”\u4e86\uff0c\u6211\u4eec\u53ea\u662f\u5728\u5b83\u4e0a\u9762\u505a”\u5c40\u90e8\u4f18\u5316”\u3002<\/p>\n

\u73b0\u5728\uff0c\u6211\u505a\u4efb\u4f55\u9879\u76ee\uff0c\u7b2c\u4e00\u4ef6\u4e8b\u4e0d\u662f\u95ee”\u7528\u4ec0\u4e48\u6a21\u578b”\uff0c\u800c\u662f\u95ee\uff1a”\u8fd9\u4e2a\u4efb\u52a1\uff0c\u9700\u8981\u591a\u5c11\u81ea\u7531\u5ea6\uff1f” \u5982\u679c\u662f\u5206\u7c7b\u3001\u6458\u8981\u3001\u95ee\u7b54\uff0c\u6211\u76f4\u63a5\u4e0a LoRA + r=8\uff1b\u5982\u679c\u662f\u4ee3\u7801\u751f\u6210\u3001\u591a\u8df3\u63a8\u7406\uff0c\u6211\u53ef\u80fd\u7528 r=32\uff0c\u751a\u81f3\u8003\u8651 PEFT \u7684\u5176\u4ed6\u65b9\u6cd5\uff0c\u6bd4\u5982 DoRA\u3002<\/p>\n

LoRA \u8ba9\u6211\u4ece”\u53c2\u6570\u7126\u8651”\u4e2d\u89e3\u8131\u4e86\u3002\u6211\u4e0d\u518d\u5bb3\u6015\u6a21\u578b\u592a\u5927\uff0c\u56e0\u4e3a\u6211\u77e5\u9053\uff1a**\u6211\u53ea\u9700\u8981\u8bad\u7ec3\u5b83\u7684\u4e00\u5c0f\u90e8\u5206\uff0c\u5c31\u80fd\u8ba9\u5b83\u7115\u7136\u4e00\u65b0**\u3002\u8fd9\u5c31\u50cf\u4e00\u4e2a\u753b\u5bb6\uff0c\u4e0d\u9700\u8981\u91cd\u753b\u6574\u5e45\u753b\uff0c\u53ea\u9700\u8981\u5728\u89d2\u843d\u6dfb\u4e00\u7b14\uff0c\u6574\u5e45\u753b\u7684\u610f\u5883\u5c31\u53d8\u4e86\u3002<\/p>\n

\u6211\u8fd8\u5728\u5b66\u4e60\uff0c\u4f46 LoRA \u5df2\u7ecf\u6539\u53d8\u4e86\u6211\u5bf9\u6df1\u5ea6\u5b66\u4e60\u7684\u4fe1\u4ef0\u2014\u2014\u4e0d\u662f\u66f4\u5927\u7684\u6a21\u578b\uff0c\u800c\u662f\u66f4\u806a\u660e\u7684\u9002\u5e94\u65b9\u5f0f\uff0c\u624d\u662f\u672a\u6765\u3002<\/p>\n

## \u56fe\u7247\u6765\u6e90<\/p>\n

– Personal design algorithm\u2026 I\u2019ve been trying to envision how a machine learning\/neural network can aid a designer\u2014not design things for you, but help figure out what your design should be; an idea expander; a fast sketching partner\u2026 anyway, that\u2019s been mak – bjornmeansbear, by-sa, https:\/\/www.flickr.com\/photos\/64519085@N00\/40362455043<\/p>\n","protected":false},"excerpt":{"rendered":"

\u8bb0\u5f55\u6211\u4ece\u5b8c\u5168\u4e0d\u61c2\u5230\u4eb2\u624b\u7528 LoRA \u5fae\u8c03 GPT-2 \u7684\u5168\u8fc7\u7a0b\uff0c\u63ed\u5f00\u4f4e\u79e9\u9002\u914d\u5982\u4f55\u5728\u4e0d\u727a\u7272\u6027\u80fd\u7684\u524d\u63d0\u4e0b\uff0c\u8ba9\u5927\u6a21\u578b\u5fae\u8c03\u53d8\u5f97\u8f7b\u76c8\u800c\u53ef\u884c\u3002<\/p>\n","protected":false},"author":2,"featured_media":179,"comment_status":"","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[],"class_list":["post-180","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-mtsb"],"_links":{"self":[{"href":"https:\/\/www.bojinhu.xyz\/index.php\/wp-json\/wp\/v2\/posts\/180","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.bojinhu.xyz\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.bojinhu.xyz\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.bojinhu.xyz\/index.php\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.bojinhu.xyz\/index.php\/wp-json\/wp\/v2\/comments?post=180"}],"version-history":[{"count":1,"href":"https:\/\/www.bojinhu.xyz\/index.php\/wp-json\/wp\/v2\/posts\/180\/revisions"}],"predecessor-version":[{"id":181,"href":"https:\/\/www.bojinhu.xyz\/index.php\/wp-json\/wp\/v2\/posts\/180\/revisions\/181"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.bojinhu.xyz\/index.php\/wp-json\/wp\/v2\/media\/179"}],"wp:attachment":[{"href":"https:\/\/www.bojinhu.xyz\/index.php\/wp-json\/wp\/v2\/media?parent=180"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.bojinhu.xyz\/index.php\/wp-json\/wp\/v2\/categories?post=180"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.bojinhu.xyz\/index.php\/wp-json\/wp\/v2\/tags?post=180"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}