{"id":37362,"date":"2025-05-15T22:27:37","date_gmt":"2025-05-15T14:27:37","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/37362.html"},"modified":"2025-05-15T22:27:37","modified_gmt":"2025-05-15T14:27:37","slug":"pytorch%e9%9a%8f%e6%9c%ba%e6%95%b0%e6%8e%a7%e5%88%b6%e5%85%a8%e6%8c%87%e5%8d%97%ef%bc%9a%e4%bb%8e%e7%a7%8d%e5%ad%90%e8%ae%be%e7%bd%ae%e5%88%b0%e7%8a%b6%e6%80%81%e7%ae%a1%e7%90%86","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/37362.html","title":{"rendered":"PyTorch\u968f\u673a\u6570\u63a7\u5236\u5168\u6307\u5357\uff1a\u4ece\u79cd\u5b50\u8bbe\u7f6e\u5230\u72b6\u6001\u7ba1\u7406"},"content":{"rendered":"

\u4e3b\u8981\u51fd\u6570\u603b\u89c8<\/h2>\n\n\u51fd\u6570\u4f5c\u7528\u9002\u7528\u8303\u56f4<\/tr>\n\n\n\n\n\n\n\n\n
torch.seed()<\/td>\n\u81ea\u52a8\u751f\u6210\u5e76\u8bbe\u7f6e\u968f\u673a\u79cd\u5b50<\/td>\n\u4e3b\u8981\u5f71\u54cd CPU，GPU \u884c\u4e3a\u4f9d\u8d56\u5177\u4f53\u7248\u672c<\/td>\n<\/tr>\n
torch.initial_seed()<\/td>\n\u67e5\u8be2\u521d\u59cb\u79cd\u5b50\u503c<\/td>\nCPU<\/td>\n<\/tr>\n
torch.manual_seed()<\/td>\n\u8bbe\u7f6e\u5168\u5c40\u968f\u673a\u79cd\u5b50<\/td>\nCPU\/CUDA<\/td>\n<\/tr>\n
torch.get_rng_state()<\/td>\n\u83b7\u53d6 CPU \u968f\u673a\u72b6\u6001<\/td>\nCPU<\/td>\n<\/tr>\n
torch.set_rng_state()<\/td>\n\u6062\u590d CPU \u968f\u673a\u72b6\u6001<\/td>\nCPU<\/td>\n<\/tr>\n
torch.cuda.get_rng_state()<\/td>\n\u83b7\u53d6 GPU \u968f\u673a\u72b6\u6001<\/td>\nCUDA<\/td>\n<\/tr>\n
torch.cuda.set_rng_state()<\/td>\n\u6062\u590d GPU \u968f\u673a\u72b6\u6001<\/td>\nCUDA<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

torch.seed()<\/h2>\n
    \n
  • \u5728 PyTorch \u4e2d，torch.seed() \u662f\u4e00\u4e2a\u7528\u4e8e\u8bbe\u7f6e\u968f\u673a\u6570\u751f\u6210\u5668（RNG）\u79cd\u5b50\u7684\u51fd\u6570，\u4ee5\u786e\u4fdd\u5b9e\u9a8c\u7684\u53ef\u91cd\u590d\u6027\u3002<\/li>\n
  • torch.seed() \u4f1a\u5c06\u6240\u6709\u968f\u673a\u6570\u751f\u6210\u5668（\u5305\u62ec CPU \u548c CUDA \u8bbe\u5907）\u7684\u79cd\u5b50\u8bbe\u7f6e\u4e3a\u540c\u4e00\u4e2a\u968f\u673a\u521d\u59cb\u503c（\u57fa\u4e8e\u7cfb\u7edf\u65f6\u95f4\u6216\u71b5\u6e90）\u3002\u5b83\u8fd4\u56de\u4e00\u4e2a 64 \u4f4d\u6574\u6570\u4f5c\u4e3a\u751f\u6210\u7684\u79cd\u5b50\u503c\u3002<\/li>\n<\/ul>\n
    \n

    seed =<\/span> torch.<\/span>seed(<\/span>)<\/span><\/p>\n

    \n

    \u4f5c\u7528\u8303\u56f4<\/h3>\n
  • PyTorch \u7684\u5168\u5c40\u968f\u673a\u72b6\u6001：\u5f71\u54cd\u6240\u6709\u4f7f\u7528 PyTorch \u968f\u673a\u6570\u751f\u6210\u5668\u7684\u64cd\u4f5c，\u4f8b\u5982：\n
      \n
    • \u6a21\u578b\u6743\u91cd\u521d\u59cb\u5316（\u5982 torch.nn.init）<\/li>\n
    • \u6570\u636e\u96c6\u7684\u968f\u673a\u6253\u4e71（\u5982 DataLoader(shuffle=True)）<\/li>\n
    • \u968f\u673a\u589e\u5f3a（\u5982\u56fe\u50cf\u53d8\u6362\u4e2d\u7684 RandomCrop）<\/li>\n
    • \u4efb\u4f55\u4f7f\u7528 torch.rand()\u3001torch.randn() \u7b49\u51fd\u6570\u7684\u64cd\u4f5c\u3002<\/li>\n<\/ul>\n<\/li>\n
    • CUDA \u8bbe\u5907：\u5982\u679c\u4f7f\u7528\u4e86 GPU，CUDA \u7684\u968f\u673a\u79cd\u5b50\u4e5f\u4f1a\u88ab\u540c\u6b65\u8bbe\u7f6e\u3002<\/li>\n
      \n

      \u4e0e manual_seed() \u7684\u533a\u522b<\/h3>\n
        \n
      • torch.seed() ：\u81ea\u52a8\u751f\u6210\u4e00\u4e2a\u968f\u673a\u79cd\u5b50\u5e76\u8bbe\u7f6e\u5b83（\u65e0\u9700\u53c2\u6570），\u9002\u5408\u5feb\u901f\u5b9e\u9a8c\u4f46\u4e0d\u7cbe\u786e\u63a7\u5236\u79cd\u5b50\u503c\u3002<\/li>\n
      • torch.manual_seed(seed) ：\u9700\u8981\u624b\u52a8\u6307\u5b9a\u4e00\u4e2a\u6574\u6570\u4f5c\u4e3a\u79cd\u5b50（\u5982 torch.manual_seed(42)），\u9002\u5408\u9700\u8981\u5b8c\u5168\u53ef\u590d\u73b0\u7684\u5b9e\u9a8c\u3002<\/li>\n<\/ul>\n
        \n

        \u6817\u5b50<\/h3>\n

        import<\/span> torch<\/p>\n

        # \u81ea\u52a8\u751f\u6210\u5e76\u8bbe\u7f6e\u968f\u673a\u79cd\u5b50<\/span>
        \nseed =<\/span> torch.<\/span>seed(<\/span>)<\/span>
        \nprint<\/span>(<\/span>"Generated seed:"<\/span>,<\/span> seed)<\/span> # \u4f8b\u5982\u8f93\u51fa: 123456789<\/span><\/p>\n

        # \u9a8c\u8bc1\u968f\u673a\u6027<\/span>
        \na =<\/span> torch.<\/span>rand(<\/span>1<\/span>)<\/span> # \u968f\u673a\u5f20\u91cf<\/span>
        \nb =<\/span> torch.<\/span>rand(<\/span>1<\/span>)<\/span> # \u53e6\u4e00\u4e2a\u968f\u673a\u5f20\u91cf<\/span>
        \nprint<\/span>(<\/span>a,<\/span> b)<\/span> # \u6bcf\u6b21\u8fd0\u884c\u7ed3\u679c\u4e0d\u540c（\u9664\u975e\u79cd\u5b50\u56fa\u5b9a）<\/span><\/p>\n

        \n

        \u6ce8\u610f\u4e8b\u9879<\/h3>\n
      • \u591a\u8bbe\u5907\/\u8fdb\u7a0b： \u5728\u5206\u5e03\u5f0f\u8bad\u7ec3\u4e2d，\u9700\u4e3a\u6bcf\u4e2a\u8fdb\u7a0b\u5355\u72ec\u8bbe\u7f6e\u79cd\u5b50（\u53ef\u80fd\u8fd8\u9700\u8bbe\u7f6e torch.cuda.manual_seed_all()）\u3002<\/li>\n
      • \u5176\u4ed6\u5e93\u7684\u5f71\u54cd： PyTorch \u7684\u79cd\u5b50\u4e0d\u63a7\u5236 NumPy\u3001Python \u5185\u7f6e random \u7b49\u5e93，\u9700\u5355\u72ec\u8bbe\u7f6e（\u5982 np.random.seed(42)）\u3002<\/li>\n
      • \u6027\u80fd： \u9891\u7e41\u8c03\u7528 torch.seed() \u53ef\u80fd\u5f71\u54cd\u6027\u80fd，\u5efa\u8bae\u5728\u5b9e\u9a8c\u5f00\u59cb\u65f6\u53ea\u8bbe\u7f6e\u4e00\u6b21\u3002<\/li>\n
        \n

        \u5e38\u89c1\u7528\u9014<\/h3>\n
          \n
        • \u8c03\u8bd5：\u901a\u8fc7\u56fa\u5b9a\u79cd\u5b50\u590d\u73b0\u95ee\u9898\u3002<\/li>\n
        • \u5b9e\u9a8c\u5bf9\u6bd4：\u786e\u4fdd\u4e0d\u540c\u6a21\u578b\u5728\u76f8\u540c\u6570\u636e\u5212\u5206\u548c\u521d\u59cb\u5316\u6761\u4ef6\u4e0b\u8bad\u7ec3\u3002<\/li>\n<\/ul>\n

          torch.manual_seed()<\/h2>\n
            \n
          • \u5728 PyTorch \u4e2d，torch.manual_seed() \u662f\u4e00\u4e2a\u7528\u4e8e\u624b\u52a8\u8bbe\u7f6e\u968f\u673a\u6570\u751f\u6210\u5668（RNG）\u79cd\u5b50\u7684\u51fd\u6570，\u76ee\u7684\u662f\u786e\u4fdd\u5b9e\u9a8c\u7684\u5b8c\u5168\u53ef\u590d\u73b0\u6027\u3002\n
              \n
            • \u4e3a CPU \u548c GPU（CUDA） \u7684\u968f\u673a\u6570\u751f\u6210\u5668\u8bbe\u7f6e\u4e00\u4e2a\u786e\u5b9a\u7684\u79cd\u5b50\u503c\u3002<\/li>\n
            • \u5f71\u54cd\u6240\u6709\u57fa\u4e8e PyTorch \u7684\u968f\u673a\u64cd\u4f5c（\u5982\u5f20\u91cf\u521d\u59cb\u5316\u3001\u6570\u636e\u6253\u4e71\u3001Dropout \u7b49）\u3002<\/li>\n
            • \u4e0e torch.seed()（\u81ea\u52a8\u751f\u6210\u968f\u673a\u79cd\u5b50）\u4e0d\u540c，manual_seed() \u9700\u8981\u663e\u5f0f\u6307\u5b9a\u79cd\u5b50\u503c\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n
              \n

              import<\/span> torch
              \nseed=<\/span>14<\/span>
              \ntorch.<\/span>manual_seed(<\/span>seed)<\/span> # seed \u5fc5\u987b\u662f\u4e00\u4e2a\u6574\u6570（\u901a\u5e38\u9009 42\u3001123 \u7b49）<\/span><\/p>\n\n\u53c2\u6570\u7c7b\u578b\u8bf4\u660e<\/tr>\n\n\n
              seed<\/td>\nint<\/td>\n\u4efb\u610f\u6574\u6570（\u8303\u56f4：0 \u5230 2\u2076\u2074-1）<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
              \n

              \u4f5c\u7528\u8303\u56f4<\/h3>\n
            • PyTorch \u5168\u5c40\u968f\u673a\u72b6\u6001：\n
                \n
              • \u5f71\u54cd torch.rand(), torch.randn(), torch.randint() \u7b49\u968f\u673a\u51fd\u6570\u3002<\/li>\n
              • \u63a7\u5236\u6a21\u578b\u521d\u59cb\u5316（\u5982 nn.Linear \u7684\u6743\u91cd）\u3002<\/li>\n
              • \u6570\u636e\u52a0\u8f7d\u5668\u7684\u6253\u4e71（\u5982 DataLoader(shuffle=True)）\u3002<\/li>\n<\/ul>\n<\/li>\n
              • CUDA \u8bbe\u5907：\u5982\u679c\u4f7f\u7528 GPU，CUDA \u7684\u968f\u673a\u79cd\u5b50\u4e5f\u4f1a\u88ab\u540c\u6b65\u8bbe\u7f6e（\u7b49\u540c\u4e8e\u8c03\u7528 torch.cuda.manual_seed_all()）\u3002<\/li>\n
                \n

                \u4e0e torch.seed() \u5bf9\u6bd4<\/h3>\n\n\u51fd\u6570\u79cd\u5b50\u6765\u6e90\u9002\u7528\u573a\u666f<\/tr>\n\n\n\n
                torch.seed()<\/td>\n\u81ea\u52a8\u751f\u6210<\/td>\n\u5feb\u901f\u5b9e\u9a8c，\u65e0\u9700\u7cbe\u786e\u590d\u73b0<\/td>\n<\/tr>\n
                torch.manual_seed()<\/td>\n\u624b\u52a8\u6307\u5b9a<\/td>\n\u9700\u8981\u5b8c\u5168\u53ef\u590d\u73b0\u7684\u7ed3\u679c<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

                \u6817\u5b50<\/h3>\n

                import<\/span> torch<\/p>\n

                # \u8bbe\u7f6e\u968f\u673a\u79cd\u5b50\u4e3a 42<\/span>
                \ntorch.<\/span>manual_seed(<\/span>42<\/span>)<\/span><\/p>\n

                # \u751f\u6210\u968f\u673a\u5f20\u91cf（\u7ed3\u679c\u53ef\u590d\u73b0）<\/span>
                \na =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>)<\/span>
                \nprint<\/span>(<\/span>a)<\/span>
                \n# \u8f93\u51fa（\u6bcf\u6b21\u8fd0\u884c\u76f8\u540c）：<\/span>
                \n# tensor([[0.8823, 0.9150],<\/span>
                \n# [0.3829, 0.9593]])<\/span><\/p>\n

                # \u91cd\u65b0\u8bbe\u7f6e\u76f8\u540c\u79cd\u5b50\u4f1a\u5f97\u5230\u76f8\u540c\u7ed3\u679c<\/span>
                \ntorch.<\/span>manual_seed(<\/span>42<\/span>)<\/span>
                \nb =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>)<\/span>
                \nprint<\/span>(<\/span>b ==<\/span> a)<\/span> # \u6240\u6709\u5143\u7d20\u4e3a True<\/span><\/p>\n

                \n

                \u6ce8\u610f\u4e8b\u9879<\/h3>\n
              • GPU \u786e\u5b9a\u6027：\u5373\u4f7f\u8bbe\u7f6e\u4e86\u79cd\u5b50，CUDA \u64cd\u4f5c\u53ef\u80fd\u56e0\u5e76\u884c\u8ba1\u7b97\u5bfc\u81f4\u8f7b\u5fae\u5dee\u5f02\u3002\u82e5\u9700\u5b8c\u5168\u786e\u5b9a\u6027，\u9700\u989d\u5916\u8bbe\u7f6e\u3002（\u6ce8\u610f：\u53ef\u80fd\u964d\u4f4e\u8bad\u7ec3\u901f\u5ea6）torch.<\/span>backends.<\/span>cudnn.<\/span>deterministic =<\/span> True<\/span> # \u542f\u7528\u786e\u5b9a\u6027\u7b97\u6cd5<\/span>
                \ntorch.<\/span>backends.<\/span>cudnn.<\/span>benchmark =<\/span> False<\/span> # \u5173\u95ed\u81ea\u52a8\u4f18\u5316<\/span>\n <\/li>\n
              • \u591a\u8fdb\u7a0b\/\u591aGPU： \u5728\u5206\u5e03\u5f0f\u8bad\u7ec3\u4e2d，\u9700\u4e3a\u6bcf\u4e2a\u8fdb\u7a0b\u5355\u72ec\u8bbe\u7f6e\u79cd\u5b50\u3002torch.<\/span>manual_seed(<\/span>42<\/span> +<\/span> rank)<\/span> # rank \u662f\u8fdb\u7a0b\u7f16\u53f7<\/span>
                \ntorch.<\/span>cuda.<\/span>manual_seed_all(<\/span>42<\/span> +<\/span> rank)<\/span>\n <\/li>\n
              • \u5176\u4ed6\u5e93\u7684\u79cd\u5b50：PyTorch \u7684\u79cd\u5b50\u4e0d\u63a7\u5236 NumPy \u6216 Python \u5185\u7f6e random \u6a21\u5757，\u9700\u5355\u72ec\u8bbe\u7f6e：import<\/span> numpy as<\/span> np
                \nimport<\/span> random
                \nnp.<\/span>random.<\/span>seed(<\/span>42<\/span>)<\/span>
                \nrandom.<\/span>seed(<\/span>42<\/span>)<\/span>\n <\/li>\n
                \n

                \u5e38\u89c1\u7528\u9014<\/h3>\n
                  \n
                • \u8c03\u8bd5\u6a21\u578b：\u901a\u8fc7\u56fa\u5b9a\u79cd\u5b50\u590d\u73b0\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u7684\u9519\u8bef\u3002<\/li>\n
                • \u5b9e\u9a8c\u5bf9\u6bd4：\u786e\u4fdd\u4e0d\u540c\u6a21\u578b\u5728\u76f8\u540c\u521d\u59cb\u5316\u6761\u4ef6\u4e0b\u516c\u5e73\u6bd4\u8f83\u3002<\/li>\n
                • \u6559\u5b66\u6f14\u793a：\u8ba9\u5b66\u751f\u83b7\u5f97\u4e0e\u6559\u7a0b\u4e00\u81f4\u7684\u968f\u673a\u7ed3\u679c\u3002<\/li>\n<\/ul>\n
                  \n

                  torch.initial_seed()<\/h2>\n
                    \n
                  • \u5728 PyTorch \u4e2d，torch.initial_seed() \u662f\u4e00\u4e2a\u7528\u4e8e\u83b7\u53d6\u5f53\u524d\u968f\u673a\u6570\u751f\u6210\u5668（RNG）\u7684\u521d\u59cb\u79cd\u5b50\u503c\u7684\u51fd\u6570\u3002\u5b83\u901a\u5e38\u7528\u4e8e\u8c03\u8bd5\u6216\u9a8c\u8bc1\u968f\u673a\u79cd\u5b50\u7684\u8bbe\u7f6e\u60c5\u51b5\u3002<\/li>\n
                  • \u8fd4\u56de\u5f53\u524d\u968f\u673a\u6570\u751f\u6210\u5668\u7684\u521d\u59cb\u79cd\u5b50\u503c（\u5373\u901a\u8fc7 torch.manual_seed() \u6216 torch.seed() \u8bbe\u7f6e\u7684\u79cd\u5b50）\u3002<\/li>\n
                  • \u5982\u679c\u672a\u624b\u52a8\u8bbe\u7f6e\u79cd\u5b50，\u5219\u8fd4\u56de\u4e00\u4e2a\u57fa\u4e8e\u7cfb\u7edf\u65f6\u95f4\u6216\u9ed8\u8ba4\u71b5\u6e90\u751f\u6210\u7684\u79cd\u5b50\u503c\u3002<\/li>\n
                  • \u4e3b\u8981\u7528\u4e8e\u8c03\u8bd5\u6216\u68c0\u67e5\u968f\u673a\u72b6\u6001，\u786e\u4fdd\u5b9e\u9a8c\u7684\u53ef\u590d\u73b0\u6027\u3002<\/li>\n<\/ul>\n
                    \n

                    seed =<\/span> torch.<\/span>initial_seed(<\/span>)<\/span><\/p>\n

                      \n
                    • \u8fd4\u56de\u503c：\u4e00\u4e2a int \u7c7b\u578b\u7684\u6574\u6570，\u8868\u793a\u5f53\u524d RNG \u7684\u521d\u59cb\u79cd\u5b50\u3002<\/li>\n<\/ul>\n
                      \n

                      \u6817\u5b50<\/h3>\n

                      \u793a\u4f8b 1：\u83b7\u53d6\u9ed8\u8ba4\u79cd\u5b50<\/h4>\n

                      import<\/span> torch<\/p>\n

                      # \u672a\u8bbe\u7f6e\u79cd\u5b50\u65f6，\u8fd4\u56de\u81ea\u52a8\u751f\u6210\u7684\u79cd\u5b50<\/span>
                      \ndefault_seed =<\/span> torch.<\/span>initial_seed(<\/span>)<\/span>
                      \nprint<\/span>(<\/span>"Default seed:"<\/span>,<\/span> default_seed)<\/span> # \u4f8b\u5982：42（\u6bcf\u6b21\u8fd0\u884c\u53ef\u80fd\u4e0d\u540c）<\/span><\/p>\n

                      \u793a\u4f8b 2：\u83b7\u53d6\u624b\u52a8\u8bbe\u7f6e\u7684\u79cd\u5b50<\/h4>\n

                      import<\/span> torch<\/p>\n

                      # \u624b\u52a8\u8bbe\u7f6e\u79cd\u5b50<\/span>
                      \ntorch.<\/span>manual_seed(<\/span>42<\/span>)<\/span><\/p>\n

                      # \u83b7\u53d6\u5f53\u524d\u521d\u59cb\u79cd\u5b50<\/span>
                      \ncurrent_seed =<\/span> torch.<\/span>initial_seed(<\/span>)<\/span>
                      \nprint<\/span>(<\/span>"Current seed:"<\/span>,<\/span> current_seed)<\/span> # \u8f93\u51fa：42<\/span><\/p>\n

                      \u793a\u4f8b 3：\u9a8c\u8bc1\u79cd\u5b50\u662f\u5426\u751f\u6548<\/h4>\n

                      import<\/span> torch<\/p>\n

                      # \u8bbe\u7f6e\u79cd\u5b50\u5e76\u751f\u6210\u968f\u673a\u5f20\u91cf<\/span>
                      \ntorch.<\/span>manual_seed(<\/span>42<\/span>)<\/span>
                      \na =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>)<\/span><\/p>\n

                      # \u68c0\u67e5\u521d\u59cb\u79cd\u5b50<\/span>
                      \nprint<\/span>(<\/span>"Initial seed:"<\/span>,<\/span> torch.<\/span>initial_seed(<\/span>)<\/span>)<\/span> # \u8f93\u51fa：42<\/span><\/p>\n

                      # \u91cd\u65b0\u8bbe\u7f6e\u76f8\u540c\u79cd\u5b50，\u5e94\u5f97\u5230\u76f8\u540c\u968f\u673a\u5f20\u91cf<\/span>
                      \ntorch.<\/span>manual_seed(<\/span>42<\/span>)<\/span>
                      \nb =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>)<\/span><\/p>\n

                      print<\/span>(<\/span>"a == b:"<\/span>,<\/span> torch.<\/span>all<\/span>(<\/span>a ==<\/span> b)<\/span>)<\/span> # \u8f93\u51fa：True<\/span><\/p>\n

                      \n

                      \u6ce8\u610f\u4e8b\u9879<\/h3>\n
                    • \n

                      initial_seed() vs manual_seed()<\/p>\n

                        \n
                      • torch.manual_seed(seed)：\u8bbe\u7f6e\u968f\u673a\u79cd\u5b50\u3002<\/li>\n
                      • torch.initial_seed()：\u83b7\u53d6\u5f53\u524d\u521d\u59cb\u79cd\u5b50\u503c（\u4e0d\u4fee\u6539 RNG \u72b6\u6001）\u3002<\/li>\n<\/ul>\n<\/li>\n
                      • \n

                        \u4e0e torch.seed() \u7684\u533a\u522b<\/p>\n

                          \n
                        • torch.seed()：\u91cd\u65b0\u751f\u6210\u5e76\u8bbe\u7f6e\u968f\u673a\u79cd\u5b50（\u8fd4\u56de\u65b0\u79cd\u5b50）\u3002<\/li>\n
                        • torch.initial_seed()：\u4ec5\u67e5\u8be2\u5f53\u524d\u521d\u59cb\u79cd\u5b50（\u4e0d\u4fee\u6539 RNG）\u3002<\/li>\n<\/ul>\n<\/li>\n
                        • \n

                          GPU（CUDA）\u79cd\u5b50<\/p>\n

                            \n
                          • torch.initial_seed() \u4ec5\u8fd4\u56de CPU \u968f\u673a\u6570\u751f\u6210\u5668\u7684\u79cd\u5b50，\u4e0d\u6d89\u53ca CUDA\u3002<\/li>\n
                          • \u8981\u83b7\u53d6 CUDA \u79cd\u5b50，\u9700\u4f7f\u7528 torch.cuda.initial_seed()\u3002<\/li>\n<\/ul>\n<\/li>\n
                          • \n

                            \u591a\u8fdb\u7a0b\/\u5206\u5e03\u5f0f\u8bad\u7ec3<\/p>\n

                              \n
                            • \u5728\u5206\u5e03\u5f0f\u8bad\u7ec3\u4e2d，\u6bcf\u4e2a\u8fdb\u7a0b\u53ef\u80fd\u6709\u4e0d\u540c\u7684\u521d\u59cb\u79cd\u5b50，\u9700\u5355\u72ec\u68c0\u67e5\u3002<\/li>\n<\/ul>\n<\/li>\n
                              \n
                            • \u5178\u578b\u5e94\u7528\u573a\u666f<\/li>\n
                                \n
                              • \u8c03\u8bd5\u968f\u673a\u521d\u59cb\u5316\u95ee\u9898：\u68c0\u67e5\u6a21\u578b\u6743\u91cd\u521d\u59cb\u5316\u662f\u5426\u4e00\u81f4\u3002<\/li>\n
                              • \u5b9e\u9a8c\u590d\u73b0：\u8bb0\u5f55\u521d\u59cb\u79cd\u5b50，\u786e\u4fdd\u540e\u7eed\u5b9e\u9a8c\u53ef\u590d\u73b0\u3002<\/li>\n
                              • \u968f\u673a\u6027\u9a8c\u8bc1：\u786e\u8ba4 DataLoader \u6216 Dropout \u662f\u5426\u6309\u9884\u671f\u5de5\u4f5c\u3002<\/li>\n<\/ul>\n
                                \n

                                \u603b\u7ed3\u5bf9\u6bd4<\/h3>\n\n\u51fd\u6570\u4f5c\u7528\u8fd4\u56de\u503c<\/tr>\n\n\n\n\n
                                torch.manual_seed()<\/td>\n\u8bbe\u7f6e\u968f\u673a\u79cd\u5b50<\/td>\nNone<\/td>\n<\/tr>\n
                                torch.seed()<\/td>\n\u81ea\u52a8\u751f\u6210\u5e76\u8bbe\u7f6e\u968f\u673a\u79cd\u5b50<\/td>\n\u65b0\u79cd\u5b50 (int)<\/td>\n<\/tr>\n
                                torch.initial_seed()<\/td>\n\u83b7\u53d6\u5f53\u524d\u521d\u59cb\u79cd\u5b50<\/td>\n\u5f53\u524d\u79cd\u5b50 (int)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

                                \u5982\u679c\u9700\u8981\u5728 PyTorch \u4e2d\u786e\u4fdd\u5b9e\u9a8c\u7684\u53ef\u590d\u73b0\u6027，\u5efa\u8bae\u7ed3\u5408：<\/p>\n

                                torch.<\/span>manual_seed(<\/span>42<\/span>)<\/span>
                                \ntorch.<\/span>backends.<\/span>cudnn.<\/span>deterministic =<\/span> True<\/span> # \u542f\u7528\u786e\u5b9a\u6027 CUDA \u8ba1\u7b97<\/span>
                                \ntorch.<\/span>backends.<\/span>cudnn.<\/span>benchmark =<\/span> False<\/span> # \u5173\u95ed\u81ea\u52a8\u4f18\u5316<\/span><\/p>\n

                                torch.get_rng_state()<\/h2>\n
                                  \n
                                • \u5728 PyTorch \u4e2d，torch.get_rng_state() \u662f\u4e00\u4e2a\u7528\u4e8e\u83b7\u53d6\u5f53\u524d\u968f\u673a\u6570\u751f\u6210\u5668（RNG）\u5b8c\u6574\u72b6\u6001\u7684\u51fd\u6570，\u5b83\u8fd4\u56de\u4e00\u4e2a\u8868\u793a\u5185\u90e8\u968f\u673a\u72b6\u6001\u7684\u5f20\u91cf\u3002\u8fd9\u4e2a\u529f\u80fd\u5728\u9700\u8981\u4fdd\u5b58\u548c\u6062\u590d\u968f\u673a\u72b6\u6001\u65f6\u975e\u5e38\u6709\u7528，\u4f8b\u5982\u5728\u5b9e\u9a8c\u7684\u67d0\u4e2a\u9636\u6bb5\u51bb\u7ed3\u968f\u673a\u6027，\u7a0d\u540e\u6062\u590d\u3002<\/li>\n<\/ul>\n
                                  \n
                                    \n
                                  • \u8fd4\u56de\u4e00\u4e2a torch.ByteTensor，\u8868\u793a PyTorch \u5f53\u524d CPU \u968f\u673a\u6570\u751f\u6210\u5668\u7684\u5185\u90e8\u72b6\u6001\u3002\u8be5\u72b6\u6001\u53ef\u4ee5\u4fdd\u5b58\u5230\u78c1\u76d8\u6216\u4f20\u9012\u7ed9 torch.set_rng_state() \u4ee5\u6062\u590d\u4e4b\u524d\u7684\u968f\u673a\u72b6\u6001\u3002<\/li>\n
                                  • \u4e0d\u5f71\u54cd CUDA \u7684\u968f\u673a\u72b6\u6001（GPU \u7684\u968f\u673a\u72b6\u6001\u9700\u4f7f\u7528 torch.cuda.get_rng_state()）\u3002<\/li>\n<\/ul>\n
                                    \n

                                    rng_state =<\/span> torch.<\/span>get_rng_state(<\/span>)<\/span><\/p>\n

                                      \n
                                    • \u8fd4\u56de\u503c：\u4e00\u4e2a torch.ByteTensor，\u5305\u542b\u5f53\u524d RNG \u7684\u5b8c\u6574\u72b6\u6001\u4fe1\u606f\u3002<\/li>\n<\/ul>\n
                                      \n

                                      \u6817\u5b50<\/h2>\n
                                        \n
                                      • \u6848\u4f8b\u53c2\u8003torch.set_rng_state()\u7684\u6817\u5b50\u3002<\/li>\n<\/ul>\n

                                        \u6ce8\u610f\u4e8b\u9879<\/h3>\n
                                      • \n

                                        \u4ec5\u9002\u7528\u4e8e CPU：torch.get_rng_state() \u53ea\u8fd4\u56de CPU \u7684\u968f\u673a\u72b6\u6001\u3002 \u5bf9\u4e8eGPU，\u9700\u4f7f\u7528 torch.cuda.get_rng_state() \u548c torch.cuda.set_rng_state()\u3002<\/p>\n<\/li>\n

                                      • \n

                                        \u72b6\u6001\u5f20\u91cf\u7684\u5185\u5bb9\u4e0d\u900f\u660e：\u8fd4\u56de\u7684 ByteTensor \u662f PyTorch \u5185\u90e8\u4f7f\u7528\u7684\u4e8c\u8fdb\u5236\u6570\u636e，\u7528\u6237\u65e0\u9700\u89e3\u6790\u5176\u5185\u5bb9\u3002<\/p>\n<\/li>\n

                                      • \n

                                        \u4e0einitial_seed() \u7684\u533a\u522b：initial_seed() \u8fd4\u56de\u521d\u59cb\u79cd\u5b50\u503c（\u6574\u6570），\u800c get_rng_state() \u8fd4\u56de\u5b8c\u6574\u7684\u8fd0\u884c\u65f6\u72b6\u6001（\u5f20\u91cf）\u3002<\/p>\n<\/li>\n

                                      • \n

                                        \u591a\u7ebf\u7a0b\/\u5206\u5e03\u5f0f\u8bad\u7ec3：\u5728\u5e76\u884c\u73af\u5883\u4e2d，\u6bcf\u4e2a\u7ebf\u7a0b\u7684\u968f\u673a\u72b6\u6001\u72ec\u7acb\u7ba1\u7406，\u9700\u5206\u522b\u4fdd\u5b58\u548c\u6062\u590d\u3002<\/p>\n<\/li>\n

                                        \n

                                        \u5178\u578b\u5e94\u7528\u573a\u666f<\/h3>\n
                                          \n
                                        • \u5b9e\u9a8c\u590d\u73b0：\u5728\u4ee3\u7801\u7684\u7279\u5b9a\u4f4d\u7f6e\u4fdd\u5b58\u968f\u673a\u72b6\u6001，\u786e\u4fdd\u540e\u7eed\u64cd\u4f5c\u53ef\u91cd\u590d\u3002<\/li>\n
                                        • \u8c03\u8bd5\u968f\u673a\u6027：\u6bd4\u8f83\u4e0d\u540c\u9636\u6bb5\u7684\u968f\u673a\u72b6\u6001，\u5b9a\u4f4d\u4e0d\u4e00\u81f4\u95ee\u9898\u3002<\/li>\n
                                        • \u5f3a\u5316\u5b66\u4e60：\u5728\u73af\u5883\u4ea4\u4e92\u8fc7\u7a0b\u4e2d\u51bb\u7ed3\u968f\u673a\u72b6\u6001，\u4fbf\u4e8e\u56de\u653e\u3002<\/li>\n<\/ul>\n
                                          \n

                                          torch.set_rng_state()<\/h2>\n
                                            \n
                                          • \u5728 PyTorch \u4e2d，torch.set_rng_state() \u662f\u4e00\u4e2a\u7528\u4e8e\u6062\u590d\u968f\u673a\u6570\u751f\u6210\u5668（RNG）\u72b6\u6001\u7684\u51fd\u6570，\u901a\u5e38\u4e0e torch.get_rng_state() \u914d\u5408\u4f7f\u7528，\u4ee5\u5b9e\u73b0\u968f\u673a\u72b6\u6001\u7684\u4fdd\u5b58\u548c\u6062\u590d\u3002\u4ee5\u4e0b\u662f\u8be6\u7ec6\u4ecb\u7ecd：<\/li>\n<\/ul>\n
                                            \n
                                              \n
                                            • \u6062\u590d CPU \u968f\u673a\u6570\u751f\u6210\u5668\u7684\u5185\u90e8\u72b6\u6001，\u4f7f\u5176\u56de\u5230\u67d0\u4e2a\u7279\u5b9a\u7684\u968f\u673a\u72b6\u6001\u3002<\/li>\n
                                            • \u901a\u5e38\u4e0e torch.get_rng_state() \u7ed3\u5408\u4f7f\u7528，\u5148\u4fdd\u5b58\u968f\u673a\u72b6\u6001，\u7a0d\u540e\u6062\u590d\u3002<\/li>\n
                                            • \u4ec5\u5f71\u54cd CPU \u7684\u968f\u673a\u72b6\u6001，GPU（CUDA）\u7684\u968f\u673a\u72b6\u6001\u9700\u8981\u4f7f\u7528 torch.cuda.set_rng_state()\u3002<\/li>\n<\/ul>\n
                                              \n

                                              torch.<\/span>set_rng_state(<\/span>new_state)<\/span><\/p>\n

                                                \n
                                              • \u53c2\u6570：\n
                                                  \n
                                                • new_state (torch.ByteTensor)：\u8981\u6062\u590d\u7684\u968f\u673a\u72b6\u6001，\u901a\u5e38\u7531 torch.get_rng_state() \u83b7\u53d6\u3002\n
                                                    \n
                                                  • \u8fd4\u56de\u503c：None（\u4ec5\u4fee\u6539\u5185\u90e8 RNG \u72b6\u6001）\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n
                                                    \n

                                                    \u6817\u5b50<\/h3>\n

                                                    \u793a\u4f8b 1：\u4fdd\u5b58\u548c\u6062\u590d\u968f\u673a\u72b6\u6001<\/h4>\n

                                                    import<\/span> torch<\/p>\n

                                                    # \u751f\u6210\u4e00\u4e9b\u968f\u673a\u6570<\/span>
                                                    \na =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>)<\/span>
                                                    \nprint<\/span>(<\/span>"First random tensor:\\\\n"<\/span>,<\/span> a)<\/span><\/p>\n

                                                    # \u4fdd\u5b58\u5f53\u524d\u968f\u673a\u72b6\u6001<\/span>
                                                    \nsaved_state =<\/span> torch.<\/span>get_rng_state(<\/span>)<\/span><\/p>\n

                                                    # \u751f\u6210\u53e6\u4e00\u7ec4\u968f\u673a\u6570（\u72b6\u6001\u5df2\u53d8\u5316）<\/span>
                                                    \nb =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>)<\/span>
                                                    \nprint<\/span>(<\/span>"Second random tensor (different from a):\\\\n"<\/span>,<\/span> b)<\/span><\/p>\n

                                                    # \u6062\u590d\u4e4b\u524d\u4fdd\u5b58\u7684\u968f\u673a\u72b6\u6001<\/span>
                                                    \ntorch.<\/span>set_rng_state(<\/span>saved_state)<\/span><\/p>\n

                                                    # \u91cd\u65b0\u751f\u6210\u968f\u673a\u6570，\u7ed3\u679c\u5e94\u4e0e\u7b2c\u4e00\u6b21\u76f8\u540c<\/span>
                                                    \nc =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>)<\/span>
                                                    \nprint<\/span>(<\/span>"Restored random tensor (same as a):\\\\n"<\/span>,<\/span> c)<\/span><\/p>\n

                                                    print<\/span>(<\/span>"a == c:"<\/span>,<\/span> torch.<\/span>allclose(<\/span>a,<\/span> c)<\/span>)<\/span> # \u8f93\u51fa: True<\/span><\/p>\n

                                                    \u793a\u4f8b 2：\u7ed3\u5408 manual_seed \u4f7f\u7528<\/h4>\n

                                                    import<\/span> torch<\/p>\n

                                                    # \u8bbe\u7f6e\u56fa\u5b9a\u79cd\u5b50<\/span>
                                                    \ntorch.<\/span>manual_seed(<\/span>42<\/span>)<\/span><\/p>\n

                                                    # \u83b7\u53d6\u521d\u59cb\u968f\u673a\u72b6\u6001<\/span>
                                                    \ninitial_state =<\/span> torch.<\/span>get_rng_state(<\/span>)<\/span><\/p>\n

                                                    # \u751f\u6210\u968f\u673a\u5f20\u91cf<\/span>
                                                    \nx =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>)<\/span>
                                                    \nprint<\/span>(<\/span>"x:\\\\n"<\/span>,<\/span> x)<\/span><\/p>\n

                                                    # \u4fee\u6539\u968f\u673a\u72b6\u6001（\u4f8b\u5982\u8fd0\u884c\u5176\u4ed6\u968f\u673a\u64cd\u4f5c）<\/span>
                                                    \n_ =<\/span> torch.<\/span>rand(<\/span>100<\/span>,<\/span> 100<\/span>)<\/span> # \u6d88\u8017\u968f\u673a\u6570，\u6539\u53d8\u72b6\u6001<\/span><\/p>\n

                                                    # \u6062\u590d\u521d\u59cb\u72b6\u6001<\/span>
                                                    \ntorch.<\/span>set_rng_state(<\/span>initial_state)<\/span><\/p>\n

                                                    # \u91cd\u65b0\u751f\u6210\u5f20\u91cf，\u7ed3\u679c\u5e94\u4e0e x \u76f8\u540c<\/span>
                                                    \ny =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>)<\/span>
                                                    \nprint<\/span>(<\/span>"y == x:"<\/span>,<\/span> torch.<\/span>allclose(<\/span>x,<\/span> y)<\/span>)<\/span> # \u8f93\u51fa: True<\/span><\/p>\n

                                                    \u793a\u4f8b3：\u5b8c\u6574\u4ee3\u7801\u793a\u4f8b（CPU + GPU）<\/h4>\n

                                                    import<\/span> torch<\/p>\n

                                                    # === CPU \u968f\u673a\u72b6\u6001\u7ba1\u7406 ===<\/span>
                                                    \ntorch.<\/span>manual_seed(<\/span>42<\/span>)<\/span>
                                                    \ncpu_state =<\/span> torch.<\/span>get_rng_state(<\/span>)<\/span>
                                                    \na =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>)<\/span>
                                                    \nprint<\/span>(<\/span>"CPU – First sample:\\\\n"<\/span>,<\/span> a)<\/span><\/p>\n

                                                    torch.<\/span>set_rng_state(<\/span>cpu_state)<\/span>
                                                    \nb =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>)<\/span>
                                                    \nprint<\/span>(<\/span>"CPU – Restored sample == first sample:"<\/span>,<\/span> torch.<\/span>allclose(<\/span>a,<\/span> b)<\/span>)<\/span><\/p>\n

                                                    # === GPU \u968f\u673a\u72b6\u6001\u7ba1\u7406 ===<\/span>
                                                    \nif<\/span> torch.<\/span>cuda.<\/span>is_available(<\/span>)<\/span>:<\/span>
                                                    \n torch.<\/span>cuda.<\/span>manual_seed_all(<\/span>42<\/span>)<\/span>
                                                    \n gpu_state =<\/span> torch.<\/span>cuda.<\/span>get_rng_state(<\/span>)<\/span>
                                                    \n c =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>,<\/span> device=<\/span>'cuda'<\/span>)<\/span>
                                                    \n print<\/span>(<\/span>"GPU – First sample:\\\\n"<\/span>,<\/span> c)<\/span><\/p>\n

                                                    torch.<\/span>cuda.<\/span>set_rng_state(<\/span>gpu_state)<\/span>
                                                    \n d =<\/span> torch.<\/span>rand(<\/span>2<\/span>,<\/span> 2<\/span>,<\/span> device=<\/span>'cuda'<\/span>)<\/span>
                                                    \n print<\/span>(<\/span>"GPU – Restored sample == first sample:"<\/span>,<\/span> torch.<\/span>allclose(<\/span>c,<\/span> d)<\/span>)<\/span><\/p>\n

                                                      \n
                                                    • \u901a\u8fc7\u5408\u7406\u4f7f\u7528 get_rng_state() \u548c set_rng_state()，\u53ef\u4ee5\u7cbe\u786e\u63a7\u5236 PyTorch \u4e2d\u7684\u968f\u673a\u6027，\u786e\u4fdd\u5b9e\u9a8c\u7684\u53ef\u63a7\u6027\u548c\u53ef\u590d\u73b0\u6027\u3002<\/li>\n<\/ul>\n
                                                      \n

                                                      \u6ce8\u610f\u4e8b\u9879<\/h3>\n
                                                    • \u4ec5\u9002\u7528\u4e8e CPU :torch.set_rng_state() \u4ec5\u6062\u590d CPU \u7684\u968f\u673a\u72b6\u6001，\u4e0d\u5f71\u54cd GPU（CUDA）\u3002 \u8981\u6062\u590d GPU \u7684\u968f\u673a\u72b6\u6001，\u9700\u4f7f\u7528 torch.cuda.set_rng_state()\u3002<\/li>\n
                                                    • \u72b6\u6001\u5f20\u91cf\u5fc5\u987b\u5339\u914d :new_state \u5fc5\u987b\u662f\u4e00\u4e2a\u6709\u6548\u7684\u968f\u673a\u72b6\u6001\u5f20\u91cf（\u901a\u5e38\u7531 torch.get_rng_state() \u8fd4\u56de），\u5426\u5219\u53ef\u80fd\u5f15\u53d1\u9519\u8bef\u3002<\/li>\n
                                                    • \u4e0e manual_seed \u7684\u533a\u522b\n
                                                        \n
                                                      • torch.manual_seed(seed) \u91cd\u7f6e\u6574\u4e2a RNG \u72b6\u6001（\u57fa\u4e8e\u7ed9\u5b9a\u7684\u79cd\u5b50）\u3002<\/li>\n
                                                      • torch.set_rng_state(state) \u7cbe\u786e\u6062\u590d\u67d0\u4e2a\u5386\u53f2\u72b6\u6001（\u66f4\u7ec6\u7c92\u5ea6\u63a7\u5236）\u3002<\/li>\n<\/ul>\n<\/li>\n
                                                      • \u591a\u7ebf\u7a0b\/\u5206\u5e03\u5f0f\u8bad\u7ec3：\u5728\u5e76\u884c\u8ba1\u7b97\u4e2d，\u6bcf\u4e2a\u7ebf\u7a0b\u9700\u5355\u72ec\u7ba1\u7406\u968f\u673a\u72b6\u6001\u3002<\/li>\n
                                                        \n

                                                        \u5178\u578b\u5e94\u7528\u573a\u666f<\/h3>\n
                                                          \n
                                                        • \u5b9e\u9a8c\u590d\u73b0：\u5728\u4ee3\u7801\u7684\u7279\u5b9a\u4f4d\u7f6e\u4fdd\u5b58\u968f\u673a\u72b6\u6001，\u7a0d\u540e\u6062\u590d\u4ee5\u590d\u73b0\u7ed3\u679c\u3002<\/li>\n
                                                        • \u8c03\u8bd5\u968f\u673a\u6027：\u6bd4\u8f83\u4e0d\u540c\u9636\u6bb5\u7684\u968f\u673a\u72b6\u6001，\u5b9a\u4f4d\u4e0d\u4e00\u81f4\u95ee\u9898\u3002<\/li>\n
                                                        • \u5f3a\u5316\u5b66\u4e60：\u5728\u73af\u5883\u4ea4\u4e92\u8fc7\u7a0b\u4e2d\u51bb\u7ed3\u968f\u673a\u72b6\u6001，\u4fbf\u4e8e\u56de\u653e\u5b9e\u9a8c\u3002<\/li>\n<\/ul>\n
                                                          \n

                                                          \u603b\u7ed3\u5bf9\u6bd4<\/h3>\n\n\u51fd\u6570\u4f5c\u7528\u9002\u7528\u8303\u56f4<\/tr>\n\n\n\n\n\n\n\n\n
                                                          torch.seed()<\/td>\n\u81ea\u52a8\u751f\u6210\u5e76\u8bbe\u7f6e\u968f\u673a\u79cd\u5b50<\/td>\n\u4e3b\u8981\u5f71\u54cd CPU，GPU \u884c\u4e3a\u4f9d\u8d56\u5177\u4f53\u7248\u672c<\/td>\n<\/tr>\n
                                                          torch.initial_seed()<\/td>\n\u67e5\u8be2\u521d\u59cb\u79cd\u5b50\u503c<\/td>\nCPU<\/td>\n<\/tr>\n
                                                          torch.manual_seed()<\/td>\n\u8bbe\u7f6e\u5168\u5c40\u968f\u673a\u79cd\u5b50<\/td>\nCPU\/CUDA<\/td>\n<\/tr>\n
                                                          torch.get_rng_state()<\/td>\n\u83b7\u53d6 CPU \u968f\u673a\u72b6\u6001<\/td>\nCPU<\/td>\n<\/tr>\n
                                                          torch.set_rng_state()<\/td>\n\u6062\u590d CPU \u968f\u673a\u72b6\u6001<\/td>\nCPU<\/td>\n<\/tr>\n
                                                          torch.cuda.get_rng_state()<\/td>\n\u83b7\u53d6 GPU \u968f\u673a\u72b6\u6001<\/td>\nCUDA<\/td>\n<\/tr>\n
                                                          torch.cuda.set_rng_state()<\/td>\n\u6062\u590d GPU \u968f\u673a\u72b6\u6001<\/td>\nCUDA<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
                                                          \n
                                                            \n
                                                          • torch.set_rng_state() \u7528\u4e8e\u7cbe\u786e\u6062\u590d CPU \u7684\u968f\u673a\u72b6\u6001，\u9002\u7528\u4e8e\u9700\u8981\u5b8c\u5168\u63a7\u5236\u968f\u673a\u6027\u7684\u573a\u666f\u3002\u4e0e torch.get_rng_state() \u914d\u5408\u4f7f\u7528，\u53ef\u5b9e\u73b0\u5b9e\u9a8c\u7684\u5b8c\u6574\u590d\u73b0\u3002<\/li>\n
                                                          • GPU \u7684\u968f\u673a\u72b6\u6001\u9700\u901a\u8fc7 torch.cuda.set_rng_state() \u7ba1\u7406\u3002<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"

                                                            \u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb1.1k\u6b21\uff0c\u70b9\u8d5e17\u6b21\uff0c\u6536\u85cf20\u6b21\u3002\u5728\u6df1\u5ea6\u5b66\u4e60\u548c\u79d1\u5b66\u8ba1\u7b97\u4e2d\uff0c\u63a7\u5236\u968f\u673a\u6027\u662f\u4fdd\u8bc1\u5b9e\u9a8c\u7ed3\u679c\u53ef\u590d\u73b0\u7684\u5173\u952e\u3002PyTorch \u63d0\u4f9b\u4e86\u591a\u79cd\u968f\u673a\u6570\u7ba1\u7406\u5de5\u5177\uff0c\u4f46\u5982\u4f55\u6b63\u786e\u4f7f\u7528\u5b83\u4eec\u4ecd\u662f\u8bb8\u591a\u5f00\u53d1\u8005\u5bb9\u6613\u6df7\u6dc6\u7684\u95ee\u9898\u3002\u672c\u6587\u5168\u9762\u89e3\u6790 PyTorch \u7684\u968f\u673a\u6570\u63a7\u5236\u673a\u5236\uff0c\u4ece\u57fa\u7840\u7684 torch.seed() \u548c torch.manual_seed()\uff0c\u5230\u66f4\u5e95\u5c42\u7684 get_rng_state() \u548c set_rng_state()\uff0c\u8be6\u7ec6\u8bf4\u660e\u5176\u4f5c\u7528\u3001\u4f7f\u7528\u573a\u666f\u53ca\u6ce8\u610f\u4e8b\u9879\u3002\u540c\u65f6\uff0c\u6211\u4eec\u4e5f\u4f1a\u63a2\u8ba8 CPU \u548c GPU\uff08CUDA\uff09\u7684\u968f\u673a\u72b6\u6001\u7ba1\u7406\uff0c\u5e76\u63d0\u4f9b\u5b8c\u6574\u7684\u4ee3\u7801\u793a\u4f8b\uff0c\u5e2e\u52a9\u4f60\u5f7b\u5e95\u638c\u63e1 PyTorc<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[81,152,3232,3233,86,3234],"topic":[],"class_list":["post-37362","post","type-post","status-publish","format-standard","hentry","category-server","tag-python","tag-pytorch","tag-rng-","tag-3233","tag-86","tag-3234"],"yoast_head":"\nPyTorch\u968f\u673a\u6570\u63a7\u5236\u5168\u6307\u5357\uff1a\u4ece\u79cd\u5b50\u8bbe\u7f6e\u5230\u72b6\u6001\u7ba1\u7406 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/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:\/\/www.wsisp.com\/helps\/37362.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"PyTorch\u968f\u673a\u6570\u63a7\u5236\u5168\u6307\u5357\uff1a\u4ece\u79cd\u5b50\u8bbe\u7f6e\u5230\u72b6\u6001\u7ba1\u7406 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb1.1k\u6b21\uff0c\u70b9\u8d5e17\u6b21\uff0c\u6536\u85cf20\u6b21\u3002\u5728\u6df1\u5ea6\u5b66\u4e60\u548c\u79d1\u5b66\u8ba1\u7b97\u4e2d\uff0c\u63a7\u5236\u968f\u673a\u6027\u662f\u4fdd\u8bc1\u5b9e\u9a8c\u7ed3\u679c\u53ef\u590d\u73b0\u7684\u5173\u952e\u3002PyTorch \u63d0\u4f9b\u4e86\u591a\u79cd\u968f\u673a\u6570\u7ba1\u7406\u5de5\u5177\uff0c\u4f46\u5982\u4f55\u6b63\u786e\u4f7f\u7528\u5b83\u4eec\u4ecd\u662f\u8bb8\u591a\u5f00\u53d1\u8005\u5bb9\u6613\u6df7\u6dc6\u7684\u95ee\u9898\u3002\u672c\u6587\u5168\u9762\u89e3\u6790 PyTorch \u7684\u968f\u673a\u6570\u63a7\u5236\u673a\u5236\uff0c\u4ece\u57fa\u7840\u7684 torch.seed() \u548c torch.manual_seed()\uff0c\u5230\u66f4\u5e95\u5c42\u7684 get_rng_state() \u548c set_rng_state()\uff0c\u8be6\u7ec6\u8bf4\u660e\u5176\u4f5c\u7528\u3001\u4f7f\u7528\u573a\u666f\u53ca\u6ce8\u610f\u4e8b\u9879\u3002\u540c\u65f6\uff0c\u6211\u4eec\u4e5f\u4f1a\u63a2\u8ba8 CPU \u548c GPU\uff08CUDA\uff09\u7684\u968f\u673a\u72b6\u6001\u7ba1\u7406\uff0c\u5e76\u63d0\u4f9b\u5b8c\u6574\u7684\u4ee3\u7801\u793a\u4f8b\uff0c\u5e2e\u52a9\u4f60\u5f7b\u5e95\u638c\u63e1 PyTorc\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/37362.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2025-05-15T14:27:37+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"5 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/37362.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/37362.html\",\"name\":\"PyTorch\u968f\u673a\u6570\u63a7\u5236\u5168\u6307\u5357\uff1a\u4ece\u79cd\u5b50\u8bbe\u7f6e\u5230\u72b6\u6001\u7ba1\u7406 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2025-05-15T14:27:37+00:00\",\"dateModified\":\"2025-05-15T14:27:37+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/37362.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/37362.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/37362.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"PyTorch\u968f\u673a\u6570\u63a7\u5236\u5168\u6307\u5357\uff1a\u4ece\u79cd\u5b50\u8bbe\u7f6e\u5230\u72b6\u6001\u7ba1\u7406\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"PyTorch\u968f\u673a\u6570\u63a7\u5236\u5168\u6307\u5357\uff1a\u4ece\u79cd\u5b50\u8bbe\u7f6e\u5230\u72b6\u6001\u7ba1\u7406 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","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:\/\/www.wsisp.com\/helps\/37362.html","og_locale":"zh_CN","og_type":"article","og_title":"PyTorch\u968f\u673a\u6570\u63a7\u5236\u5168\u6307\u5357\uff1a\u4ece\u79cd\u5b50\u8bbe\u7f6e\u5230\u72b6\u6001\u7ba1\u7406 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb1.1k\u6b21\uff0c\u70b9\u8d5e17\u6b21\uff0c\u6536\u85cf20\u6b21\u3002\u5728\u6df1\u5ea6\u5b66\u4e60\u548c\u79d1\u5b66\u8ba1\u7b97\u4e2d\uff0c\u63a7\u5236\u968f\u673a\u6027\u662f\u4fdd\u8bc1\u5b9e\u9a8c\u7ed3\u679c\u53ef\u590d\u73b0\u7684\u5173\u952e\u3002PyTorch \u63d0\u4f9b\u4e86\u591a\u79cd\u968f\u673a\u6570\u7ba1\u7406\u5de5\u5177\uff0c\u4f46\u5982\u4f55\u6b63\u786e\u4f7f\u7528\u5b83\u4eec\u4ecd\u662f\u8bb8\u591a\u5f00\u53d1\u8005\u5bb9\u6613\u6df7\u6dc6\u7684\u95ee\u9898\u3002\u672c\u6587\u5168\u9762\u89e3\u6790 PyTorch \u7684\u968f\u673a\u6570\u63a7\u5236\u673a\u5236\uff0c\u4ece\u57fa\u7840\u7684 torch.seed() \u548c torch.manual_seed()\uff0c\u5230\u66f4\u5e95\u5c42\u7684 get_rng_state() \u548c set_rng_state()\uff0c\u8be6\u7ec6\u8bf4\u660e\u5176\u4f5c\u7528\u3001\u4f7f\u7528\u573a\u666f\u53ca\u6ce8\u610f\u4e8b\u9879\u3002\u540c\u65f6\uff0c\u6211\u4eec\u4e5f\u4f1a\u63a2\u8ba8 CPU \u548c GPU\uff08CUDA\uff09\u7684\u968f\u673a\u72b6\u6001\u7ba1\u7406\uff0c\u5e76\u63d0\u4f9b\u5b8c\u6574\u7684\u4ee3\u7801\u793a\u4f8b\uff0c\u5e2e\u52a9\u4f60\u5f7b\u5e95\u638c\u63e1 PyTorc","og_url":"https:\/\/www.wsisp.com\/helps\/37362.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2025-05-15T14:27:37+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"5 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/37362.html","url":"https:\/\/www.wsisp.com\/helps\/37362.html","name":"PyTorch\u968f\u673a\u6570\u63a7\u5236\u5168\u6307\u5357\uff1a\u4ece\u79cd\u5b50\u8bbe\u7f6e\u5230\u72b6\u6001\u7ba1\u7406 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2025-05-15T14:27:37+00:00","dateModified":"2025-05-15T14:27:37+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/37362.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/37362.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/37362.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"PyTorch\u968f\u673a\u6570\u63a7\u5236\u5168\u6307\u5357\uff1a\u4ece\u79cd\u5b50\u8bbe\u7f6e\u5230\u72b6\u6001\u7ba1\u7406"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/37362","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=37362"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/37362\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=37362"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=37362"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=37362"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=37362"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}