{ "version": "https://jsonfeed.org/version/1", "title": "\u673a\u5668\u5b66\u4e60", "description": "\u673a\u5668\u5b66\u4e60\u662f\u4eba\u5de5\u667a\u80fd\u7684\u4e00\u4e2a\u5206\u652f", "home_page_url": "go/ml", "feed_url": "feed/ml.json", "icon": "https://cdn.v2ex.com/navatar/9fe8/593a/678_large.png?m=1628872967", "favicon": "https://cdn.v2ex.com/navatar/9fe8/593a/678_normal.png?m=1628872967", "items": [ { "author": { "url": "member/jununhsu", "name": "jununhsu", "avatar": "https://cdn.v2ex.com/gravatar/04a92ffd7108b44c9d3339c7f714650c?s=73&d=retro" }, "url": "t/1196092", "date_modified": "2026-03-07T05:44:41+00:00", "content_html": "

\u5927\u5806\u7684\u6587\u4ef6\u89e3\u6790\u3001\u65e0\u7528\u6570\u636e\u7684\u8fc7\u6ee4\u7b49\u7b49\u3002\u3002\u3002\u3002\u3002\u3002\u3002\u3002\n\u8fd8\u6709\u66f4\u7e41\u7410\u7684\u5417\uff0c\u8bf4\u51fa\u6765\u8ba9\u6211\u5e73\u8861\u5e73\u8861

\n", "date_published": "2026-03-05T08:30:49+00:00", "title": "llm \u8bad\u7ec3\u6700 dirty work \u7684\u5c31\u662f\u6570\u636e\u5904\u7406\uff01", "id": "t/1196092" }, { "author": { "url": "member/clueglue", "name": "clueglue", "avatar": "https://cdn.v2ex.com/avatar/af62/2dfe/773519_large.png?m=1762480664" }, "url": "t/1185975", "date_modified": "2026-01-15T09:47:17+00:00", "content_html": "

\u5728\u6211\u7684\u672c\u79d1\u9636\u6bb5\uff0c\u6570\u5206\u9ad8\u4ee3\u662f\u7f3a\u5c11\u4e00\u95e8\u8bfe\u7a0b\u6574\u5408\u7684\u3002\u5bfc\u81f4\u5728\u505a\u673a\u5668\u5b66\u4e60\u6df1\u5ea6\u5b66\u4e60\u7684\u65f6\u5019\u53bb\u9605\u8bfb\u4e00\u4e9b\u77e9\u9635\u5fae\u5206\u5f88\u5934\u75bc\uff0c\u53ea\u4f1a component-by-component \u7684\u53bb\u9a8c\u8bc1\uff0c\u6548\u7387\u4f4e\u5230\u7206\u3002

\u76f4\u5230\u6211\u53d1\u73b0\u4e86 MIT \u8fd9\u95e8\u8bfe\uff0c\u624d\u53d1\u73b0\u5176\u5b9e\u81ea\u5df1\u65e9\u5c31\u62e5\u6709\u66f4\u4f18\u96c5\u7684\u53bb\u9605\u8bfb\u8fd9\u4e00\u7c7b\u6587\u7ae0\u7684\u80fd\u529b\uff0c\u53ea\u662f\u7f3a\u5c11\u4e00\u4e2a\u6b63\u786e\u7684\u89c6\u89d2

\u5e76\u4e0d\u662f\u8bc4\u4ef7\u8fd9\u95e8\u8bfe\u6709\u4f18\u79c0\uff0c\u53ea\u662f\u5e0c\u671b\u80fd\u5e2e\u5230\u7f3a\u5c11\u77e9\u9635\u5fae\u79ef\u5206\u77e5\u8bc6\u7684\u4eba\uff0c\u5728\u6b64\u5206\u4eab\u51fa\u6765\u3002 MIT 18.S096

\n", "date_published": "2026-01-15T09:44:36+00:00", "title": "\u505a\u673a\u5668\u5b66\u4e60\u5f3a\u70c8\u63a8\u8350\u770b\u770b MIT 18.S096 \u77e9\u9635\u5fae\u79ef\u5206", "id": "t/1185975" }, { "author": { "url": "member/MuskZhou", "name": "MuskZhou", "avatar": "https://cdn.v2ex.com/gravatar/95bc07d865881eee42210e2ac00fdad3?s=73&d=retro" }, "url": "t/1182644", "date_modified": "2026-01-04T14:33:28+00:00", "content_html": "\u5982\u9898\u6240\u793a\uff0c\u5bfc\u51fa\u4e86\u5927\u6982 12 \u4e07\u6761\u804a\u5929\u8bb0\u5f55\uff0c\u8bf7\u95ee\u5404\u4f4d\u4f6c\u6709\u4ec0\u4e48\u597d\u7684\u529e\u6cd5\u8bad\u7ec3\u8ba9 ai \u6a21\u4eff\u5417", "date_published": "2026-01-02T00:04:29+00:00", "title": "\u6c42\u52a9\u5fae\u4fe1\u804a\u5929\u8bb0\u5f55\u8bad\u7ec3 ai", "id": "t/1182644" }, { "author": { "url": "member/littlemis", "name": "littlemis", "avatar": "https://cdn.v2ex.com/avatar/50c3/3de6/653830_large.png?m=1774763278" }, "url": "t/1180789", "date_modified": "2025-12-25T00:38:57+00:00", "content_html": "", "date_published": "2025-12-24T00:03:53+00:00", "title": "\u8bf7\u95ee AI \u8bad\u7ec3\u5927\u4f6c : \u76ee\u524d\u5927\u5bb6\u4e0d\u5199 blog, stackoverflow \u7b49\uff0c\u5982\u4f55\u8bad\u7ec3\u65b0\u7684\u4f18\u8d28\u6570\u636e\uff0c\u8fd8\u662f\u62ff AI \u6570\u636e\u8bad\u7ec3\u81ea\u5df1?", "id": "t/1180789" }, { "author": { "url": "member/YanSeven", "name": "YanSeven", "avatar": "https://cdn.v2ex.com/avatar/3f22/feff/383743_large.png?m=1775465214" }, "url": "t/1178092", "date_modified": "2025-12-11T00:43:20+00:00", "content_html": "\u5305\u62ec nlp,cv,llm \u8fd9\u4e9b

\u6709\u4e00\u4e2a\u6e05\u5355\u6216\u8005\u4e66\u7c4d\u80fd\u591f\u56ca\u62ec\u8fd9\u4e9b\u6570\u5b66\u5185\u5bb9\u5417\u3002", "date_published": "2025-12-10T09:13:52+00:00", "title": "\u60f3\u8981\u771f\u6b63\u7406\u89e3\u76ee\u524d\u7684\u6df1\u5ea6\u5b66\u4e60(\u4e0d\u4ec5\u662f\u6253\u6742\u7684\u70bc\u4e39\u5e08)\uff0c\u9700\u8981\u600e\u6837\u7684\u6570\u5b66\u57fa\u7840", "id": "t/1178092" }, { "author": { "url": "member/ns2250225", "name": "ns2250225", "avatar": "https://cdn.v2ex.com/avatar/cbc4/6b75/140914_large.png?m=1759846579" }, "url": "t/1177292", "date_modified": "2025-12-07T10:41:49+00:00", "content_html": "

\u6211\u5df2\u7ecf\u95ee\u8fc7 AI \u4e86\uff0c\u5df2\u7ecf\u628a onnx \u6a21\u578b\u8fdb\u884c\u4e86\u526a\u679d\uff0c\u8fdb\u884c\u4e86 int8 \u52a8\u6001\u91cf\u5316\uff0c\u5f00\u542f\u4e86 cpu \u591a\u7ebf\u7a0b\u63a8\u7406\uff0c\u901f\u5ea6\u7531\u539f\u6765\u7684 3s \u4f18\u5316\u5230\u73b0\u5728\u7684 1.3s \uff0c\u611f\u89c9\u5230\u6781\u9650\u4e86\uff0c\u53ef\u662f\u8001\u677f\u7684\u8981\u6c42\u662f 500ms \uff0c\u5927\u4f6c\u5feb\u6551\u6551\u5b69\u5b50

\n", "date_published": "2025-12-06T09:44:23+00:00", "title": "\u600e\u6837\u52a0\u5feb onnx \u6a21\u578b\u5728 cpu \u4e0a\u7684\u63a8\u7406\u901f\u5ea6\u554a", "id": "t/1177292" }, { "author": { "url": "member/intoext", "name": "intoext", "avatar": "https://cdn.v2ex.com/avatar/16dc/e148/243325_large.png?m=1688862810" }, "url": "t/1175862", "title": "AI \u516c\u53f8\u7784\u51c6\u7684\u987a\u5e8f\u662f\u6d4f\u89c8\u5668\u3001\u8f93\u5165\u6cd5\u3001\u64cd\u4f5c\u7cfb\u7edf", "id": "t/1175862", "date_published": "2025-11-29T12:02:16+00:00", "content_html": "

\u6d4f\u89c8\u5668\u662f\u4e92\u8054\u7f51\u5165\u53e3\uff0c\u89c2\u5bdf\u6d4f\u89c8\u5668\u7684\u8bbf\u95ee\u884c\u4e3a\u53ef\u6709\u52a9\u4e8e\u4e86\u89e3\u7528\u6237\u3002\n\u7b49\u4e8e\u7ed9 AI \u516c\u53f8\u5582\u6599\uff08\u6216\u8005\u91c7\u96c6\uff09\u3002

\u4ece\u8c46\u5305\u505a\u8f93\u5165\u6cd5\u6765\u770b\uff0c\u672a\u6765\u4f30\u8ba1\u8981\u4f1a\u805a\u7126\u8f93\u5165\u6cd5\u3002\n\u56e0\u4e3a\u8f93\u5165\u6cd5\u6bd4\u6d4f\u89c8\u5668\u8fd8\u80fd\u6355\u6349\u7528\u6237\u6570\u636e\u3002\u5bf9\u4e8e\u4e0d\u4e0a\u7f51\u7684\u6570\u636e\u3001\u884c\u4e3a\u4e5f\u80fd\u91c7\u96c6\u3002

\u518d\u540e\u6765\uff0c\u4f30\u8ba1\u662f\u8981\u7ade\u4e89\u64cd\u4f5c\u7cfb\u7edf\u4e86\u3002\n\u81ea\u7136\uff0c\u56e0\u4e3a\u6709\u751f\u6001\u539f\u56e0\uff0c\u4e0d\u5927\u4f1a\u51fa\u73b0\u9664 linux \u3001windows \u4e4b\u540e\u7684\u5927\u7cfb\u7edf\u3002\n\u4f46\u662f\u6211\u4f30\u8ba1\u4f7f\u7528\u5f00\u6e90 linux \u9b54\u6539\u7684\u7cfb\u7edf\u662f\u5f88\u591a\u7684--AI \u516c\u53f8\u641e\u70b9\u8fd9\u4e2a\uff0c\u6280\u672f\u4e0a\u7b80\u76f4\u6613\u5982\u53cd\u638c\u3002

\u4e0d\u77e5\u9053\u5927\u5bb6\u600e\u4e48\u770b\uff1f

\n" }, { "author": { "url": "member/rpish", "name": "rpish", "avatar": "https://cdn.v2ex.com/avatar/713c/b677/317324_large.png?m=1762331329" }, "url": "t/1173455", "date_modified": "2025-11-19T01:09:12+00:00", "content_html": "

\u597d\u5947\u9664\u4e86\u5927\u6a21\u578b\uff0c\u56fe\u7247\u751f\u6210\uff0c\u8fd8\u6709\u6ca1\u6709\u4ec0\u4e48\u80fd\u672c\u5730\u8dd1\u7684\uff0c\u597d\u73a9\u7684\u673a\u5668\u5b66\u4e60\u9879\u76ee\uff1f

\n", "date_published": "2025-11-18T03:01:15+00:00", "title": "\u6709\u4ec0\u4e48\u80fd\u672c\u5730\u8dd1\u7684\u673a\u5668\u5b66\u4e60\u9879\u76ee\u5417\uff1f", "id": "t/1173455" }, { "author": { "url": "member/fortytwo", "name": "fortytwo", "avatar": "https://cdn.v2ex.com/avatar/8981/f3e6/742644_large.png?m=1767587046" }, "url": "t/1172404", "date_modified": "2025-11-12T16:45:44+00:00", "content_html": "

\u4f7f\u7528 ddddocr \u81ea\u8bad\u7ec3\u6a21\u578b\u573a\u666f\u4e0b\uff0c\u9047\u5230\u7684\u95ee\u9898

(dddd-trainer) PS C:\\Users\\12460\\Documents\\project\\dddd_trainer> uv run .\\test.py\n2025-11-13 00:35:09.0336981 [E:onnxruntime:, sequential_executor.cc:572 onnxruntime::ExecuteKernel] Non-zero status code returned while running SequenceAt node. Name:'n0_424' Status Message: Invalid sequence index (57) specified for sequence of size (57)\nTraceback (most recent call last):\n File \"C:\\Users\\12460\\Documents\\project\\dddd_trainer\\test.py\", line 14, in <module>\n res = ocr.classification(image)\n ^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"C:\\Users\\12460\\Documents\\project\\dddd_trainer\\.venv\\Lib\\site-packages\\ddddocr\\__init__.py\", line 2643, in classification\n ort_outs = self.__ort_session.run(None, ort_inputs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"C:\\Users\\12460\\Documents\\project\\dddd_trainer\\.venv\\Lib\\site-packages\\onnxruntime\\capi\\onnxruntime_inference_collection.py\", line 287, in run\n return self._sess.run(output_names, input_feed, run_options)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\nonnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Non-zero status code returned while running SequenceAt node. Name:'n0_424' Status Message: Invalid sequence index (57) specified for sequence of size (57)\n\n

\u6d4b\u8bd5\u7684\u8bad\u7ec3\u96c6\u6570\u636e\uff1a\nhttps://wwm.lanzoum.com/itczd0b5z3yj

GitHub\uff1a https://github.com/sml2h3/dddd_trainer.git

# test.py\nimport ddddocr\n\nocr = ddddocr.DdddOcr(\n det=False,\n ocr=False,\n show_ad=False,\n import_onnx_path=r\"C:\\Users\\12460\\Documents\\project\\dddd_trainer\\projects\\test_2\\models\\test_2_1.0_23_6000_2025-11-13-00-06-13.onnx\",\n charsets_path=r\"C:\\Users\\12460\\Documents\\project\\dddd_trainer\\projects\\test_2\\models\\charsets.json\",\n)\n\nwith open(r\"C:\\Users\\12460\\Downloads\\1112\\new\\PKKQ_1578462523867.jpg\", \"rb\") as f:\n image = f.read()\n\nres = ocr.classification(image)\nprint(res)\n

\u64cd\u4f5c\u65e5\u5fd7

(dddd-trainer) PS C:\\Users\\12460\\Documents\\project\\dddd_trainer> uv run .\\app.py create test_3\n2025-11-13 00:29:53.496 | INFO | __main__:__init__:12 -\nHello baby~\n2025-11-13 00:29:53.497 | INFO | __main__:create:15 -\nCreate Project ----> test_3\n2025-11-13 00:29:53.497 | INFO | utils.project_manager:create_project:13 - Creating Directory... ----> C:\\Users\\12460\\Documents\\project\\dddd_trainer\\projects\\test_3\n2025-11-13 00:29:53.497 | INFO | utils.project_manager:create_project:20 - Creating Directory... ----> C:\\Users\\12460\\Documents\\project\\dddd_trainer\\projects\\test_3\\models\n2025-11-13 00:29:53.498 | INFO | utils.project_manager:create_project:24 - Creating Directory... ----> C:\\Users\\12460\\Documents\\project\\dddd_trainer\\projects\\test_3\\cache\n2025-11-13 00:29:53.498 | INFO | utils.project_manager:create_project:28 - Creating Directory... ----> C:\\Users\\12460\\Documents\\project\\dddd_trainer\\projects\\test_3\\checkpoints\n2025-11-13 00:29:53.498 | INFO | utils.project_manager:create_project:32 - Creating CRNN Config File... ----> C:\\Users\\12460\\Documents\\project\\dddd_trainer\\projects\\test_3\\config.yaml\n2025-11-13 00:29:53.500 | INFO | utils.project_manager:create_project:36 - Create Project Success! ----> test_3\n(dddd-trainer) PS C:\\Users\\12460\\Documents\\project\\dddd_trainer> uv run .\\app.py cache test_3 C:\\Users\\12460\\Downloads\\1112\\images\n2025-11-13 00:30:10.911 | INFO | __main__:__init__:12 -\nHello baby~\n2025-11-13 00:30:10.913 | INFO | __main__:cache:20 -\nCaching Data ----> test_3\nPath ----> C:\\Users\\12460\\Downloads\\1112\\images\n2025-11-13 00:30:10.919 | INFO | utils.cache_data:__get_label_from_name:36 -\nFiles number is 8599.\n100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 8599/8599 [00:00<00:00, 1379439.31it/s]\n2025-11-13 00:30:10.926 | INFO | utils.cache_data:__collect_data:92 -\nCoolect labels is [\" \", \"1\", \"D\", \"Q\", \"F\", \"R\", \"P\", \"5\", \"9\", \"6\", \"X\", \"G\", \"H\", \"S\", \"V\", \"Z\", \"3\", \"T\", \"K\", \"J\", \"W\", \"2\", \"8\", \"4\", \"U\", \"Y\", \"E\", \"I\", \"C\", \"B\", \"L\", \"A\", \"7\"]\n2025-11-13 00:30:10.929 | INFO | utils.cache_data:__collect_data:96 -\nWriting Cache Data!\n2025-11-13 00:30:10.929 | INFO | utils.cache_data:__collect_data:98 -\nCache Data Number is 8599\n2025-11-13 00:30:10.929 | INFO | utils.cache_data:__collect_data:99 -\nWriting Train and Val File.\n2025-11-13 00:30:10.931 | INFO | utils.cache_data:__collect_data:116 -\nTrain Data Number is 8342\n2025-11-13 00:30:10.932 | INFO | utils.cache_data:__collect_data:117 -\nVal Data Number is 257\n(dddd-trainer) PS C:\\Users\\12460\\Documents\\project\\dddd_trainer> uv run .\\app.py train test_3\n2025-11-13 00:30:26.382 | INFO | __main__:__init__:12 -\nHello baby~\n2025-11-13 00:30:26.383 | INFO | __main__:train:26 -\nStart Train ----> test_3\n\n2025-11-13 00:30:26.384 | INFO | utils.train:__init__:40 -\nTaget:\nmin_Accuracy: 0.97\nmin_Epoch: 20\nmax_Loss: 0.05\n2025-11-13 00:30:26.384 | INFO | utils.train:__init__:45 -\nUSE GPU ----> 0\n2025-11-13 00:30:26.384 | INFO | utils.train:__init__:52 -\nSearch for history checkpoints...\n2025-11-13 00:30:26.384 | INFO | utils.train:__init__:69 -\nEmpty history checkpoints\n2025-11-13 00:30:26.384 | INFO | utils.train:__init__:71 -\nBuilding Net...\nC:\\Users\\12460\\Documents\\project\\dddd_trainer\\.venv\\Lib\\site-packages\\torch\\nn\\modules\\rnn.py:123: UserWarning: dropout option adds dropout after all but last recurrent layer, so non-zero dropout expects num_layers greater than 1, but got dropout=0.3 and num_layers=1\n warnings.warn(\n2025-11-13 00:30:26.400 | INFO | utils.train:__init__:75 - Net(\n (cnn): DdddOcr(\n (cnn): Sequential(\n (conv0): Conv2d(1, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n (relu0): ReLU(inplace=True)\n (pooling0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n (conv1): Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n (relu1): ReLU(inplace=True)\n (pooling1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n (conv2): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n (batchnorm2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n (relu2): ReLU(inplace=True)\n (conv3): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n (relu3): ReLU(inplace=True)\n (pooling2): MaxPool2d(kernel_size=(2, 2), stride=(2, 1), padding=(0, 1), dilation=1, ceil_mode=False)\n (conv4): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n (batchnorm4): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n (relu4): ReLU(inplace=True)\n (conv5): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n (relu5): ReLU(inplace=True)\n (pooling3): MaxPool2d(kernel_size=(2, 2), stride=(2, 1), padding=(0, 1), dilation=1, ceil_mode=False)\n (conv6): Conv2d(128, 128, kernel_size=(2, 2), stride=(1, 1))\n (batchnorm6): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n (relu6): ReLU(inplace=True)\n )\n )\n (lstm): LSTM(384, 384, dropout=0.3, bidirectiOnal=True)\n (loss): CTCLoss()\n (fc): Linear(in_features=768, out_features=33, bias=True)\n)\n2025-11-13 00:30:26.400 | INFO | utils.train:__init__:76 -\nBuilding End\n2025-11-13 00:30:26.509 | INFO | utils.train:__init__:81 -\nGet Data Loader...\n2025-11-13 00:30:26.510 | INFO | utils.load_cache:__init__:102 -\nCharsets is [\" \", \"1\", \"D\", \"Q\", \"F\", \"R\", \"P\", \"5\", \"9\", \"6\", \"X\", \"G\", \"H\", \"S\", \"V\", \"Z\", \"3\", \"T\", \"K\", \"J\", \"W\", \"2\", \"8\", \"4\", \"U\", \"Y\", \"E\", \"I\", \"C\", \"B\", \"L\", \"A\", \"7\"]\n2025-11-13 00:30:26.510 | INFO | utils.load_cache:__init__:106 -\nImage Resize is [-1, 64]\n2025-11-13 00:30:26.510 | INFO | utils.load_cache:__init__:118 -\nImage Path is C:\\Users\\12460\\Downloads\\1112\\images\n2025-11-13 00:30:26.511 | INFO | utils.load_cache:__init__:25 -\nReading Cache File... ----> C:\\Users\\12460\\Documents\\project\\dddd_trainer\\projects\\test_3\\cache\\cache.train.tmp\n2025-11-13 00:30:26.512 | INFO | utils.load_cache:__init__:30 -\nRead Cache File End! Caches Num is 8342.\n2025-11-13 00:30:26.512 | INFO | utils.load_cache:__init__:25 -\nReading Cache File... ----> C:\\Users\\12460\\Documents\\project\\dddd_trainer\\projects\\test_3\\cache\\cache.val.tmp\n2025-11-13 00:30:26.513 | INFO | utils.load_cache:__init__:30 -\nRead Cache File End! Caches Num is 257.\n2025-11-13 00:30:26.513 | INFO | utils.train:__init__:87 -\nGet Data Loader End!\n2025-11-13 00:30:30.161 | INFO | utils.train:start:108 - [2025-11-13-00_30_30] Epoch: 0 Step: 100 LastLoss: 3.5620920658111572 AvgLoss: 4.684982450008392 Lr: 0.01\n2025-11-13 00:30:33.520 | INFO | utils.train:start:108 - [2025-11-13-00_30_33] Epoch: 0 Step: 200 LastLoss: 3.629438877105713 AvgLoss: 3.591061780452728 Lr: 0.01\n2025-11-13 00:30:36.940 | INFO | utils.train:start:108 - [2025-11-13-00_30_36] Epoch: 1 Step: 300 LastLoss: 3.478365659713745 AvgLoss: 3.574828338623047 Lr: 0.01\n2025-11-13 00:30:40.677 | INFO | utils.train:start:108 - [2025-11-13-00_30_40] Epoch: 1 Step: 400 LastLoss: 3.725684881210327 AvgLoss: 3.5830772018432615 Lr: 0.01\n2025-11-13 00:30:44.054 | INFO | utils.train:start:108 - [2025-11-13-00_30_44] Epoch: 1 Step: 500 LastLoss: 3.6253044605255127 AvgLoss: 3.5884796619415282 Lr: 0.01\n...\n2025-11-13 00:33:39.470 | INFO | utils.train:start:108 - [2025-11-13-00_33_39] Epoch: 19 Step: 5200 LastLoss: 0.0010958763305097818 AvgLoss: 0.0016259117133449763 Lr: 0.009604\n2025-11-13 00:33:44.079 | INFO | utils.train:start:108 - [2025-11-13-00_33_44] Epoch: 20 Step: 5300 LastLoss: 0.0009493848774582148 AvgLoss: 0.0010536348074674606 Lr: 0.009604\n2025-11-13 00:33:48.775 | INFO | utils.train:start:108 - [2025-11-13-00_33_48] Epoch: 20 Step: 5400 LastLoss: 0.0010489150881767273 AvgLoss: 0.0012551895825890823 Lr: 0.009604\n2025-11-13 00:33:53.303 | INFO | utils.train:start:108 - [2025-11-13-00_33_53] Epoch: 21 Step: 5500 LastLoss: 0.0011606556363403797 AvgLoss: 0.000944445063942112 Lr: 0.009604\n2025-11-13 00:33:57.803 | INFO | utils.train:start:108 - [2025-11-13-00_33_57] Epoch: 21 Step: 5600 LastLoss: 0.0007763305329717696 AvgLoss: 0.0009715505503118038 Lr: 0.009604\n2025-11-13 00:34:02.339 | INFO | utils.train:start:108 - [2025-11-13-00_34_02] Epoch: 21 Step: 5700 LastLoss: 0.0012899019056931138 AvgLoss: 0.0015180535794934258 Lr: 0.009604\n2025-11-13 00:34:06.902 | INFO | utils.train:start:108 - [2025-11-13-00_34_06] Epoch: 22 Step: 5800 LastLoss: 0.0009287762804888189 AvgLoss: 0.0015425046352902428 Lr: 0.009604\n2025-11-13 00:34:11.336 | INFO | utils.train:start:108 - [2025-11-13-00_34_11] Epoch: 22 Step: 5900 LastLoss: 0.0007486791582778096 AvgLoss: 0.0011510271998122334 Lr: 0.009604\n2025-11-13 00:34:15.845 | INFO | utils.train:start:137 - [2025-11-13-00_34_15] Epoch: 23 Step: 6000 LastLoss: 0.0006851484067738056 AvgLoss: 0.0009158773045055568 Lr: 0.009604 Acc: 1.0\n2025-11-13 00:34:15.846 | INFO | utils.train:start:143 -\nTraining Finished!Exporting Model...\nC:\\Users\\12460\\Documents\\project\\dddd_trainer\\nets\\__init__.py:216: UserWarning: # 'dynamic_axes' is not recommended when dynamo=True, and may lead to 'torch._dynamo.exc.UserError: Constraints violated.' Supply the 'dynamic_shapes' argument instead if export is unsuccessful.\n torch.onnx.export(net, dummy_input, graph_path, export_params=True, verbose=False,\nW1113 00:34:16.515000 3024 .venv\\Lib\\site-packages\\torch\\onnx\\_internal\\exporter\\_compat.py:114] Setting ONNX exporter to use operator set version 18 because the requested opset_version 12 is a lower version than we have implementations for. Automatic version conversion will be performed, which may not be successful at converting to the requested version. If version conversion is unsuccessful, the opset version of the exported model will be kept at 18. Please consider setting opset_version >=18 to leverage latest ONNX features\nThe model version conversion is not supported by the onnxscript version converter and fallback is enabled. The model will be converted using the onnx C API (target version: 12).\nFailed to convert the model to the target version 12 using the ONNX C API. The model was not modified\nTraceback (most recent call last):\n File \"C:\\Users\\12460\\Documents\\project\\dddd_trainer\\.venv\\Lib\\site-packages\\onnxscript\\version_converter\\__init__.py\", line 127, in call\n converted_proto = _c_api_utils.call_onnx_api(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"C:\\Users\\12460\\Documents\\project\\dddd_trainer\\.venv\\Lib\\site-packages\\onnxscript\\version_converter\\_c_api_utils.py\", line 65, in call_onnx_api\n result = func(proto)\n ^^^^^^^^^^^\n File \"C:\\Users\\12460\\Documents\\project\\dddd_trainer\\.venv\\Lib\\site-packages\\onnxscript\\version_converter\\__init__.py\", line 122, in _partial_convert_version\n return onnx.version_converter.convert_version(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"C:\\Users\\12460\\Documents\\project\\dddd_trainer\\.venv\\Lib\\site-packages\\onnx\\version_converter.py\", line 39, in convert_version\n converted_model_str = C.convert_version(model_str, target_version)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\nRuntimeError: D:\\a\\onnx\\onnx\\onnx/version_converter/BaseConverter.h:68: adapter_lookup: Assertion `false` failed: No Adapter From Version $18 for Split\nSkipping constant folding for op Split with multiple outputs.\nSkipping constant folding for op Split with multiple outputs.\nApplied 3 of general pattern rewrite rules.\nSkipping constant folding for op Split with multiple outputs.\nSkipping constant folding for op Split with multiple outputs.\n2025-11-13 00:34:45.655 | INFO | utils.train:start:159 -\nExport Finished!Using Time: 3.816666666666667min\n(dddd-trainer) PS C:\\Users\\12460\\Documents\\project\\dddd_trainer> uv run .\\test.py\n2025-11-13 00:35:09.0336981 [E:onnxruntime:, sequential_executor.cc:572 onnxruntime::ExecuteKernel] Non-zero status code returned while running SequenceAt node. Name:'n0_424' Status Message: Invalid sequence index (57) specified for sequence of size (57)\nTraceback (most recent call last):\n File \"C:\\Users\\12460\\Documents\\project\\dddd_trainer\\test.py\", line 14, in <module>\n res = ocr.classification(image)\n ^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"C:\\Users\\12460\\Documents\\project\\dddd_trainer\\.venv\\Lib\\site-packages\\ddddocr\\__init__.py\", line 2643, in classification\n ort_outs = self.__ort_session.run(None, ort_inputs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"C:\\Users\\12460\\Documents\\project\\dddd_trainer\\.venv\\Lib\\site-packages\\onnxruntime\\capi\\onnxruntime_inference_collection.py\", line 287, in run\n return self._sess.run(output_names, input_feed, run_options)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\nonnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Non-zero status code returned while running SequenceAt node. Name:'n0_424' Status Message: Invalid sequence index (57) specified for sequence of size (57)\n

\n", "date_published": "2025-11-12T16:45:06+00:00", "title": "\u6c42\u52a9 \u4f7f\u7528 ddddocr \u81ea\u8bad\u7ec3\u6a21\u578b\u573a\u666f\u4e0b\uff0c\u9047\u5230\u4e86 onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument", "id": "t/1172404" }, { "author": { "url": "member/huangyezhufeng", "name": "huangyezhufeng", "avatar": "https://cdn.v2ex.com/avatar/c6bf/ce8d/174807_large.png?m=1732972431" }, "url": "t/1170130", "date_modified": "2025-11-03T03:18:54+00:00", "content_html": "

\n
https://datahonor.com/blog/2025/11/02/bert/
\n

\u76ee\u5f55\uff1a

\u5173\u4e8e\u6211\u4eec\u505a\u5230\u4e86\u54ea\u4e00\u6b65\nBERT \u7b80\u4ecb\n\u6709\u8da3\u7684\u89c2\u5bdf\n \u9884\u8bad\u7ec3\u8fc7\u7a0b\u7684\u6709\u8da3\u73b0\u8c61\n \u6570\u636e\u975e\u5e38\u975e\u5e38\u91cd\u8981\n \u4ece Finetune \u7ba1\u7aa5\u9884\u8bad\u7ec3\u7684\u5a01\u529b\n\u9519\u8bef\u4e0e\u6559\u8bad\n \u5373\u4f7f\u5728\u6781\u5c0f\u6570\u636e\u96c6\u4e0a\u4e5f\u65e0\u6cd5\u8fc7\u62df\u5408\n \u4e00\u6b21\u6027\u9884\u5904\u7406\u5168\u91cf\u6570\u636e\uff0cCPU \u52a0\u73ed\uff0cGPU \u4f11\u5047\n \u8bd5\u56fe\u7528\u4e34\u65f6\u624b\u6413\u7684\u73a9\u610f\u513f\u8fbe\u5230\u5f88\u597d\u7684\u6027\u80fd\n workers \u548c prefetch \u56e0\u5b50\u8bbe\u5f97\u8fc7\u5927\u5bfc\u81f4 OOM\n \u6ca1\u6709\u4f18\u96c5\u5904\u7406 DataLoader worker \u7684\u5f02\u5e38\n \u8bc4\u4f30\u65f6\u4f7f\u7528\u4e86\u9519\u8bef\u7684 tokenizer\n\u6700\u540e\n

\n", "date_published": "2025-11-03T03:18:33+00:00", "title": "BERT \u8bba\u6587\u590d\u73b0: \u7ecf\u9a8c & \u6559\u8bad", "id": "t/1170130" }, { "author": { "url": "member/vocaloid", "name": "vocaloid", "avatar": "https://cdn.v2ex.com/avatar/6134/c51f/392716_large.png?m=1733360228" }, "url": "t/1168798", "date_modified": "2025-10-28T15:23:27+00:00", "content_html": "

\u611f\u89c9\u57fa\u7840\u8fd8\u662f\u4e0d\u592a\u7262\u9760\uff0c\u4e0d\u77e5\u9053\u5404\u4f4d v \u53cb\u6709\u6ca1\u6709\u76f8\u5173\u7684\u4e66\u7c4d\u53ef\u4ee5\u63a8\u8350

\n", "date_published": "2025-10-28T01:11:14+00:00", "title": "\u6c42\u63a8\u8350\u4e00\u4e9b LLM/\u6df1\u5ea6\u5b66\u4e60\u65b9\u9762\u6253\u57fa\u7840\u7684\u4e66\u7c4d\u6216\u8005\u8bfe\u7a0b", "id": "t/1168798" }, { "author": { "url": "member/Suinn", "name": "Suinn", "avatar": "https://cdn.v2ex.com/gravatar/8b8b1a9742d73da3199ba27940769499?s=73&d=retro" }, "url": "t/1167088", "date_modified": "2025-10-21T00:27:01+00:00", "content_html": "

\u5982\u9898\uff0c\u6709\u4e00\u5927\u6279\u6570\u636e\u96c6\uff0c\u4f46\u662f\u6709\u4e00\u6bb5\u65f6\u95f4\u6ca1\u4e86\u89e3\u56fe\u50cf\u5206\u7c7b\u8fd9\u4e2a\u9886\u57df\u4e86\uff0c\u8fd9\u4e2a\u573a\u666f\u7528\u5927\u6a21\u578b\u611f\u89c9\u4e5f\u4e0d\u662f\u5f88\u9002\u5408\uff0c\u6240\u4ee5\u60f3\u95ee\u95ee\u4e07\u80fd\u7684\u7ad9\u53cb\u4eec

\n", "date_published": "2025-10-20T09:12:46+00:00", "title": "2025 \u5e74\u5e95\uff0c\u5982\u679c\u60f3\u505a\u4e00\u4e2a\u56fe\u50cf\u5206\u7c7b\u6a21\u578b\uff0c\u7528\u4ec0\u4e48\u6a21\u578b\u6765\u5fae\u8c03\u6bd4\u8f83\u597d", "id": "t/1167088" }, { "author": { "url": "member/kennen", "name": "kennen", "avatar": "https://cdn.v2ex.com/gravatar/95a9d3dbadd4fa58570e4aa369a91813?s=73&d=retro" }, "url": "t/1166309", "date_modified": "2025-10-18T04:24:58+00:00", "content_html": "

\u76ee\u6807\u662f\u68c0\u6d4b\u4e00\u5f20\u56fe\u7247\u662f\u5426\u5408\u89c4\uff0c\u5c06\u6765\u8981\u6269\u5c55\u5230\u66f4\u591a\u7269\u4f53\u7684\u68c0\u6d4b\uff08\u6bd4\u5982\u4e00\u8f86\u8f66\u6b63\u5728\u65bd\u5de5\uff0c\u524d\u65b9\u4fa7\u65b9\u90fd\u8981\u653e\u7f6e\u5b89\u5168\u9525\u6876\uff09\uff0c\u76ee\u524d\u91c7\u7528\u7684\u662f yolov8 \uff0c\u6709\u6728\u6709\u522b\u7684\u6280\u672f\u65b9\u6848\u53ef\u4ee5\u8003\u8651\uff1f

\n", "date_published": "2025-10-17T01:49:04+00:00", "title": "\u76ee\u524d\u4f7f\u7528 yolov8 \u8bad\u7ec3\u4e86\u4e00\u4e2a\u76ee\u6807\u68c0\u6d4b\u6a21\u578b\uff0c\u4f46\u662f\u4e0a\u7ebf\u4f7f\u7528\u6548\u679c\u4e0d\u662f\u5f88\u597d\uff0c\u600e\u4e48\u4f18\u5316\uff1f", "id": "t/1166309" }, { "author": { "url": "member/quech", "name": "quech", "avatar": "https://cdn.v2ex.com/gravatar/0336289e24d797efe2b64a9b8e04232f?s=73&d=retro" }, "url": "t/1166079", "title": "\u6700\u8fd1\u770b 1-2 \u4e07\u4e3b\u673a\u53d1\u73b0 3090 \u8fd8\u662f\u80fd\u6253\u7684", "id": "t/1166079", "date_published": "2025-10-16T03:38:07+00:00", "content_html": "

\u72d7\u4e1c\u627e\u4e86\u4e00\u4e9b\u4e3b\u673a\u914d\u7f6e\uff0c\u663e\u5361\u96c6\u4e2d\u5728\u4e0b\u9762\u578b\u53f7\uff0cai \u5206\u6790\u5bf9\u6bd4\u4e4b\u540e\u5c45\u7136\u8fd8\u662f\u63a8\u8350 3090\nNvidia RTX4000 Ada 20GB

RTX3090 24G

A5000 24G

2000Ada 16G

RTX4000Ada 20G

RTX5070Ti 16Gb

RTX4000Ada 20G

5080 16G

5060Ti 16G

RTX5060Ti 16G

5070Ti 16G

RTX 5000 24G

\n" }, { "author": { "url": "member/EasonIndie", "name": "EasonIndie", "avatar": "https://cdn.v2ex.com/avatar/6243/806b/708439_large.png?m=1764144065" }, "url": "t/1164848", "date_modified": "2025-10-14T08:17:58+00:00", "content_html": "

V \u53cb\u4eec\u597d\uff0c\u6700\u8fd1\u60f3\u81ea\u5df1\u505a\u4e00\u4e2a \u4e2d\u6587\u6709\u58f0\u5c0f\u8bf4\u9879\u76ee\uff0c\u5e0c\u671b\u547b\u541f\u81ea\u7136\u6d41\u7545\u3001\u8bed\u6c14\u751f\u52a8\u3001\u6709\u505c\u987f\u3001\u7b11\u58f0\u8fd9\u4e9b\u8868\u8fbe\u3002\n\u76ee\u6807\u662f \u514b\u9686\u81ea\u5b9a\u4e49\u58f0\u97f3\uff08\u6211\u81ea\u5df1\u6216\u4e00\u4e9b\u53d7\u6b22\u8fce\u7684\u914d\u97f3\uff09\uff0c\u7136\u540e\u6279\u91cf\u751f\u6210\u5c0f\u8bf4\u97f3\u9891\u3002

\ud83d\udca1 \u6211\u7684\u9700\u6c42\u5927\u81f4\u662f\uff1a

\u5148\u514b\u9686/\u5fae\u8c03\u4e00\u4e2a\u4e2d\u6587\u58f0\u97f3\uff0c\u8ba9\u58f0\u97f3\u81ea\u7136\u3001\u4e0d\u673a\u68b0\u3002
\u98ce\u683c\u5e0c\u671b\u50cf\u76f8\u58f0\u90a3\u79cd\u201c\u8bf4\u5b66\u9017\u5531\u201d\u7684\u8bed\u6c14\u2014\u2014\u8981\u80fd\u63a7\u5236\u8282\u594f\u3001\u505c\u987f\u3001\u60c5\u7eea\u3002
\u540e\u671f\u505a\u6709\u58f0\u5c0f\u8bf4\u8f93\u51fa

\ud83d\udda5\ufe0f \u5f53\u524d\u8bbe\u5907\u914d\u7f6e\uff1a

CPU\uff1aR5 5600G
\u5185\u5b58\uff1a32GB DDR4
\u663e\u5361\uff1a\u6682\u672a\u8d2d\u4e70\uff0c\u8003\u8651 RTX 3060 \uff08 12GB \uff09 \u6216 RTX 3060Ti \uff08 8GB \uff09\u5927\u6982\u5c31\u662f 2000 \u5143\u4ef7\u4f4d\u7684
\u7cfb\u7edf\uff1aWindows / Ubuntu \u90fd\u80fd\u88c5\uff08\u5f00\u53d1\u73af\u5883\u53ef\u5207\u6362\uff09

\ud83d\udd27 \u521d\u6b65\u6280\u672f\u8def\u7ebf\uff1a

\u6211\u76ee\u524d\u67e5\u4e0b\u6765\u4e3b\u8981\u6709\u4e24\u6761\u601d\u8def\uff1a

\n
\u5feb\u901f\u4e0a\u624b\u8def\u7ebf\uff08\u751f\u6210\u7c7b\uff09
\n
- \u4f7f\u7528 Suno Bark \u6216\u7c7b\u4f3c\u5927\u6a21\u578b\uff0c\u76f4\u63a5\u751f\u6210\u591a\u60c5\u611f\u4e2d\u6587\u8bed\u97f3\uff1b
- \u9002\u5408\u5148\u8bd5\u6548\u679c\u3001\u8c03\u8282 prompt \u5b9e\u73b0\u201c\u76f8\u58f0\u8bed\u6c14\u201d\uff1b
- \u7f3a\u70b9\u662f\u58f0\u97f3\u4e0d\u4e00\u5b9a\u7a33\u5b9a\u3001\u4e0d\u53ef\u63a7\u3002
\n
\n
\u9ad8\u53ef\u63a7\u8def\u7ebf\uff08\u8bad\u7ec3\u7c7b\uff09
\n
- \u7ba1\u7ebf\uff1aSpeaker Encoder \u2192 VITS / Coqui-TTS \u2192 HiFi-GAN \uff1b
- \u5f55\u5236 20 \uff5e 60 \u5206\u949f\u9ad8\u8d28\u91cf\u97f3\u9891\u505a\u5fae\u8c03\uff1b
- \u76ee\u6807\u662f\u514b\u9686\u7a33\u5b9a\u97f3\u8272\u3001\u80fd\u751f\u6210\u81ea\u7136\u8bed\u8c03\u7684\u4e2d\u6587\u8bed\u97f3\uff1b
- \u652f\u6301 prosody / style token \u63a7\u5236\u8282\u594f\u8bed\u6c14\u3002
\n

\u8ba1\u5212\u5728\u672c\u5730\u7528 PyTorch + CUDA \u8bad\u7ec3/\u63a8\u7406\uff0c\u663e\u5b58\u6709\u9650\u7684\u8bdd\u51c6\u5907\u4e0a fp16 + LoRA + \u68af\u5ea6\u7d2f\u79ef \u7b49\u4f18\u5316\u3002

\u2753 \u60f3\u8bf7\u6559 V \u53cb\u4eec\uff1a

\n
3060 12GB \u6216 3060Ti 8GB \u80fd\u80dc\u4efb TTS / \u58f0\u97f3\u514b\u9686\u4efb\u52a1\uff1f
\n
- \u542c\u8bf4\u663e\u5b58\u5bb9\u91cf\u5728 TTS \u573a\u666f\u4e0b\u6bd4\u7b97\u529b\u66f4\u5173\u952e\uff0c\u60f3\u542c\u4e0b\u5927\u5bb6\u7684\u5b9e\u6d4b\u7ecf\u9a8c\u3002
\n
\n
\u5982\u679c\u53ea\u505a\u8f7b\u91cf\u5fae\u8c03\uff08\u4e0d\u4ece\u96f6\u8bad\u7ec3\uff09\uff0c3060Ti \u662f\u5426\u8db3\u591f\uff1f
\n
\n
\u6709\u6ca1\u6709\u4eba\u5b9e\u9645\u5728\u672c\u5730\u8dd1\u8fc7 Bark / VITS / Coqui-TTS / HiFi-GAN \u8fd9\u7c7b\u9879\u76ee\uff1f
\n
- \u63a8\u7406\u901f\u5ea6\u548c\u663e\u5b58\u5360\u7528\u5927\u6982\u600e\u6837\uff1f
- \u6709\u6ca1\u6709\u63a8\u8350\u7684\u663e\u5b58\u4f18\u5316\u6280\u5de7\uff1f
\n
\n
\u5bf9\u4e8e\u201c\u76f8\u58f0\u98ce\u683c\u201d\u7684\u8bed\u97f3\uff0c\u6709\u6ca1\u6709\u6210\u719f\u7684\u98ce\u683c\u8fc1\u79fb\u6216 prosody \u63a7\u5236\u65b9\u6cd5\uff1f
\n

\u2705 \u76ee\u6807\uff1a

\u80fd\u5728\u5bb6\u7528\u673a\u4e0a\u7a33\u5b9a\u751f\u6210\u76f8\u58f0\u98ce\u683c\u7684\u6709\u58f0\u5c0f\u8bf4\u97f3\u9891\uff0c\u81ea\u5df1\u505a\u58f0\u97f3\u6a21\u578b\u548c\u540e\u671f\uff0c\u957f\u671f\u8fed\u4ee3\u3002

\u5927\u5bb6\u6709\u505a\u8fc7\u7c7b\u4f3c\u58f0\u97f3\u514b\u9686 / \u672c\u5730 TTS \u9879\u76ee\u7684\uff0c\u6c42\u5206\u4eab\u7ecf\u9a8c\u548c\u663e\u5361\u9009\u578b\u5efa\u8bae \ud83d\ude4f

\n", "date_published": "2025-10-13T06:49:36+00:00", "title": "\u60f3\u505a\u6709\u58f0\u5c0f\u8bf4\u58f0\u97f3\u514b\u9686\uff0c 3060Ti \u80fd\u625b\u5f97\u4f4f\u5417\uff1f", "id": "t/1164848" }, { "author": { "url": "member/huangyezhufeng", "name": "huangyezhufeng", "avatar": "https://cdn.v2ex.com/avatar/c6bf/ce8d/174807_large.png?m=1732972431" }, "url": "t/1136339", "title": "[AI Glimpse \u79d1\u666e\u7cfb\u5217] KV Cache \u7684\u539f\u7406\u63a8\u5bfc\u4e0e\u4ee3\u7801\u5b9e\u73b0(\u57fa\u4e8e GPT2)", "id": "t/1136339", "date_published": "2025-06-04T08:34:06+00:00", "content_html": "

\u535a\u5ba2\u5730\u5740: LLM KV Cache: A Simple Implementation
\u91cd\u70b9\u8bb2\u89e3\u4e86\u4e3a\u4ec0\u4e48\u9700\u8981\u5728\u63a8\u7406\u9636\u6bb5\u505a KV Cache
\u4ee3\u7801\u57fa\u4e8e GPT2 \u5b9e\u73b0\uff0c\u53ef\u4ee5\u5728 16G \u663e\u5361\u4e0a\u8fd0\u884c\u590d\u73b0\uff0c\u4ee3\u7801\u5f00\u6e90\u5728 ai-glimpse/toyllm: ToyLLM: Learning LLM from Scratch

\n" }, { "author": { "url": "member/fenglirookie", "name": "fenglirookie", "avatar": "https://cdn.v2ex.com/gravatar/1d15b5d761612945f51e15c85e50de07?s=73&d=retro" }, "url": "t/1134586", "title": "\u5173\u4e8e OCR \u4ee5\u53ca\u4ee5\u56fe\u627e\u56fe\u8bc6\u522b", "id": "t/1134586", "date_published": "2025-05-27T03:22:16+00:00", "content_html": "

\u6211\u73b0\u5728\u4f7f\u7528 com.google.mlkit:text-recognition:16.0.1 \uff0ccom.google.mlkit:text-recognition-chinese:16.0.1 \u8fd9\u4e2a\u5e93\u8fdb\u884c OCR \u63d0\u53d6\uff0c\u5728\u786e\u8ba4\u4f7f\u7528\u8fd9\u4e2a\u5e93\u7684\u60c5\u51b5\u4e0b\u6709\u4ec0\u4e48\u65b9\u6cd5\u53ef\u4ee5\u63d0\u9ad8 OCR \u8bc6\u522b\u7684\u6b63\u786e\u7387\u5462\uff1f opencv \u7684\u56fe\u7247\u5339\u914d\u662f\u5426\u4e5f\u53ef\u4ee5\u540c\u7406\u63d0\u5347\u6b63\u786e\u7387\uff1f

\n" }, { "author": { "url": "member/jacketma", "name": "jacketma", "avatar": "https://cdn.v2ex.com/gravatar/a941f83ac4192740c2e1f95676ce23c5?s=73&d=retro" }, "url": "t/1132328", "date_modified": "2025-05-16T14:11:42+00:00", "content_html": "

\u8fd9\u4e2a\u6a21\u578b\u751f\u6210\u7684\u89c6\u9891\uff0c\u4eba\u7269\u8868\u60c5\u4e0d\u50f5\u786c\u3001\u76f8\u5f53\u81ea\u7136\uff0c\u54ea\u4e2a\u6a21\u578b\u8c03\u51fa\u8fd9\u6837\u7684\u6548\u679c\uff1f

$\"ai.PNG\"$

\u539f\u89c6\u9891\uff1a https://www.youtube.com/shorts/ygWC9GH3c0A

\u7528\u4e86\u51e0\u4e2a\u56fd\u4ea7\u6a21\u578b\uff0c\u505a\u4e0d\u51fa\u6765\u8fd9\u4e2a\u6548\u679c

\n", "date_published": "2025-05-16T14:08:25+00:00", "title": "\u8fd9\u7c7b\u578b\u7684\u82f1\u8bed\u5b66\u4e60\u89c6\u9891\uff0c\u662f\u7528\u54ea\u4e2a\u6a21\u578b\u751f\u6210\uff1f", "id": "t/1132328" }, { "author": { "url": "member/hmxxmh", "name": "hmxxmh", "avatar": "https://cdn.v2ex.com/gravatar/1a62b759fe6c82112bfb8368b2a2922d?s=73&d=retro" }, "url": "t/1131503", "date_modified": "2025-05-14T13:48:02+00:00", "content_html": "

\u56fe\u7247\u6709\u8fd9\u4e9b\u7279\u5f81\uff1a\u989c\u8272\u3001\u5f62\u72b6\u3001\u6e05\u6670\u5ea6\u3001\u662f\u5426 ps

\n", "date_published": "2025-05-13T09:44:06+00:00", "title": "\u600e\u4e48\u5b9e\u73b0\u63d0\u53d6\u4fe9\u4e2a\u56fe\u7247\u7684\u4e00\u4e9b\u7279\u5f81\uff0c\u7136\u540e\u8fdb\u884c\u7279\u5f81\u6bd4\u5bf9", "id": "t/1131503" }, { "author": { "url": "member/momomoi", "name": "momomoi", "avatar": "https://cdn.v2ex.com/avatar/4f35/eb21/534740_large.png?m=1775729709" }, "url": "t/1131241", "date_modified": "2025-05-14T13:48:08+00:00", "content_html": "

\u573a\u666f\u662f\uff1a\n\u53ef\u89c1\u5149\u6444\u50cf\u5934\u5b9e\u65f6\u62cd\u6444\u4e00\u7247\u533a\u57df\uff0c\u5f88\u5c0f\u90e8\u5206\u662f\u53cc\u76ee(\u5e26\u70ed\u6210\u50cf)\uff0c\u533a\u57df\u5185\u5b58\u5728\u9519\u7efc\u4ea4\u53c9\u7684\u7ba1\u9053(\u5c31\u662f\u7c97\u6c34\u7ba1)

\u73b0\u5728\u9700\u8981\u68c0\u6d4b\u6f0f\u6c34\uff0c\u6c34\u72b6\u6001\u6709 2 \u79cd\uff1a\u7ba1\u9053\u7834\u88c2\u55b7\u518d\u7a7a\u4e2d \u3001\u5730\u9762\u79ef\u6c34\uff0c\u4e3b\u8981\u5c31\u662f\u8fd9\u79cd\u5f02\u5e38\u56fe\u50cf\u592a\u5c11\uff0c\u4e2a\u4f4d\u6570\u7ea7\u522b

\u7528\u4e86 yolo \u68c0\u6d4b\u6a21\u578b+\u56fe\u50cf\u6574\u4e2a\u753b\u9762\u548c\u57fa\u51c6\u56fe\u7684\u5224\u522b\uff0c\u53d1\u73b0\u6f0f\u5224\u4e25\u91cd\uff0c \u4e3b\u8981\u662f\u6444\u50cf\u5934\u4e0d\u65f6\u5bf9\u7126+\u6709\u7a97\u6237\u3001\u5149\u6591\u4f1a\u6620\u5c04\u5230\u5730\u9762\u4e0a+\u5176\u4ed6\u5f02\u5e38\u60c5\u51b5

\u6709\u6ca1\u6709\u5176\u4ed6\u601d\u8def \u6307\u5bfc\u4e00\u4e0b\u3002

\n", "date_published": "2025-05-12T09:55:52+00:00", "title": "\u68c0\u6d4b\u753b\u9762\u4e2d\u5b58\u5728\u6f0f\u6c34\uff0c\u7ed9\u70b9\u601d\u8def", "id": "t/1131241" }, { "author": { "url": "member/soleils", "name": "soleils", "avatar": "https://cdn.v2ex.com/gravatar/6b20eecbba98f121270fa56f0886b2f4?s=73&d=retro" }, "url": "t/1126818", "title": "\u5b66\u4e60 AI, \u8dd1\u4e0b\u6a21\u578b, \u5e94\u8be5\u7528\u4ec0\u4e48 N \u5361?", "id": "t/1126818", "date_published": "2025-04-20T09:38:30+00:00", "content_html": "\u770b\u4e86\u4e0b 5060TI 5070TI 5080 \u90fd\u662f 16G \u663e\u5b58, 4090 \u548c 5090 \u53c8\u592a\u8d35, \u8ba1\u7b97\u901f\u5ea6\u8fd8\u53ef\u4ee5\u5fcd\u53d7\u6162\u4e00\u70b9 , \u8dd1 AI \u5e94\u8be5\u662f\u663e\u5b58\u51b3\u5b9a\u80fd\u4e0d\u80fd\u8dd1\u8d77\u6765\u5427, \u5148\u8dd1\u8d77\u6765\u518d\u8bf4\u7684\u60c5\u51b5\u4e0b, \u4e70\u54ea\u4e2a?" }, { "author": { "url": "member/JoeDH", "name": "JoeDH", "avatar": "https://cdn.v2ex.com/gravatar/7b4910417986ae21be51460897b6b841?s=73&d=retro" }, "url": "t/1126539", "date_modified": "2025-04-19T08:11:12+00:00", "content_html": "

\u6700\u8fd1\u6709\u4e2a\u9700\u6c42\uff0c\u5c31\u662f\u9700\u8981\u90e8\u7f72\u4e00\u5957\u673a\u5668\u5b66\u4e60\u5e73\u53f0\uff0c\u6ee1\u8db3\u5df2\u6709\u516c\u53f8\u5185\u4e00\u4e9b\u6a21\u578b\u7684\u7ba1\u7406\u5565\u7684\uff08\u4e3b\u8981\u662f\u4f20\u7edf\u7684\u56fe\u50cf\u8bc6\u522b\u6a21\u578b\uff09

\u9700\u8981\u652f\u6301\u670d\u52a1\u5668\u7b97\u529b\u3001\u6a21\u578b\u7248\u672c\u3001\u8bad\u7ec3\u3001\u6d4b\u8bd5\u5206\u6790\u3001\u6570\u636e\u96c6\u7b49\u7b49\u8fd9\u7c7b\u7684\u7ba1\u7406

\u770b\u4e86\u4e00\u4e0b kubeflow \u3001\u817e\u8baf\u7684 cube-studio \uff0c\u611f\u89c9\u5185\u5bb9\u6709\u70b9\u5927\u53c8\u590d\u6742

\u56e0\u4e3a\u6211\u4e0d\u662f\u641e AI \u7684\u6240\u4ee5\u4e0d\u592a\u61c2\uff0c\u6709\u5927\u4f6c\u5728\u7528\u7684\u6216\u662f\u4e86\u89e3\u8fd9\u5757\u7684\u5417\uff1f\u662f\u5426\u8fd8\u6709\u5176\u4ed6\u63a8\u8350\u7684\u4e00\u4e9b\u6613\u4e8e\u4f7f\u7528\u7684\u6846\u67b6\u6216\u5e73\u53f0

\n", "date_published": "2025-04-18T09:42:46+00:00", "title": "\u6709\u6ca1\u6709\u5927\u4f6c\u4e86\u89e3\u673a\u5668\u5b66\u4e60\u7684 mlops \u6846\u67b6\u6216\u8005\u5e73\u53f0", "id": "t/1126539" }, { "author": { "url": "member/iOTOi", "name": "iOTOi", "avatar": "https://cdn.v2ex.com/gravatar/4bbde4e566952a88a27ad6e4302988eb?s=73&d=retro" }, "url": "t/1126159", "date_modified": "2025-04-17T12:50:35+00:00", "content_html": "\u5982\u9898\uff0c\u6211\u8fd9\u6709\u5927\u91cf\u7684\u56fe\u7247\u5185\u5bb9\uff0cPDF \uff0cExcel \uff0c\u4ee5\u53ca PPT \u6587\u4ef6\uff0c\u6211\u770b\u5f88\u591a\u5927\u6a21\u578b\u77e5\u8bc6\u5e93\u9700\u8981\u4e0a\u4f20\u6587\u4ef6\uff0c\u6709\u6ca1\u6709\u65b9\u6cd5\u53ef\u4ee5\u4e0d\u4e0a\u4f20\uff0c\u76f4\u63a5\u8bfb\u53d6\u6587\u4ef6\u7136\u540e\u505a\u6210\u5927\u6a21\u578b\u7684\u77e5\u8bc6\u5e93\uff1f", "date_published": "2025-04-17T05:47:58+00:00", "title": "\u5982\u4f55\u628a\u672c\u5730\u6587\u4ef6\u4f5c\u4e3a\u5927\u6a21\u578b\u7684\u77e5\u8bc6\u5e93", "id": "t/1126159" }, { "author": { "url": "member/marklu", "name": "marklu", "avatar": "https://cdn.v2ex.com/gravatar/ee20aa547c70ad09e499e02cd773a663?s=73&d=retro" }, "url": "t/1126133", "date_modified": "2025-04-17T12:50:33+00:00", "content_html": "\u697c\u4e3b\u662f\u505a\u7ecf\u6d4e\u5b66\u7814\u7a76\u7684\uff0c\u8fd1\u671f\u5de5\u4f5c\u6d89\u53ca\u5230\u5229\u7528 LLM \u8fdb\u884c\u6587\u672c\u5206\u7c7b\u3002\u5728\u67e5\u9605\u6587\u732e\uff08\u4e3b\u8981\u662f\u7ecf\u6d4e\u5b66\u4ea4\u53c9\u5e94\u7528\u7c7b\uff09\u65f6\uff0c\u6211\u89c2\u5bdf\u5230\u4e00\u4e2a\u8d8b\u52bf\uff1a\u4f5c\u8005\u4eec\u503e\u5411\u4e8e\u83b7\u53d6\u7279\u5b9a\u4efb\u52a1\u7684\u6807\u6ce8\u6570\u636e\uff0c\u7136\u540e\u5728 BERT \u3001ERNIE \u7b49\u6a21\u578b\u4e0a\u8fdb\u884c\u5fae\u8c03\uff0c\u4ee5\u5b8c\u6210\u5206\u7c7b\u3002

\u7531\u4e8e\u6211\u7684\u6280\u672f\u80cc\u666f\u6709\u9650\uff0c\u5bf9\u4e8e\u8fd9\u79cd\u65b9\u6cd5\u7684\u9009\u62e9\u53ca\u5176\u66ff\u4ee3\u65b9\u6848\u6709\u4e9b\u7591\u95ee\uff0c\u60f3\u542c\u542c\u5927\u5bb6\u7684\u4e13\u4e1a\u610f\u89c1\uff1a

1.\u5fae\u8c03\u7684\u5b9e\u9645\u53ef\u884c\u6027\uff1a \u7528\u4e13\u95e8\u7684\u5206\u7c7b\u6570\u636e\u53bb\u5fae\u8c03\u4e00\u4e2a\u901a\u7528\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u4f1a\u4e0d\u4f1a\u635f\u5931\u6a21\u578b\u57fa\u7840\u6027\u80fd\uff0c\u5bfc\u81f4\u5728\u9047\u5230\u4e0e\u5fae\u8c03\u6570\u636e\u4e0d\u592a\u4e00\u6837\u7684\u6587\u672c\u65f6\uff0c\u8868\u73b0\u53cd\u800c\u53d8\u5dee\uff1f
2.\u662f\u5426\u53ef\u4ee5\u76f4\u63a5\u4f7f\u7528 GPT \u7b49\u6a21\u578b\u66ff\u4ee3\uff1a \u73b0\u5728\u6709\u8bb8\u591a\u80fd\u529b\u975e\u5e38\u5f3a\u5927\u7684 LLM \uff0c\u4f3c\u4e4e\u53ef\u4ee5\u901a\u8fc7\u7ed9\u51fa\u6e05\u6670\u7684\u6307\u4ee4\uff08 Prompt \uff09\u5c31\u80fd\u5b8c\u6210\u5f88\u591a\u4efb\u52a1\u3002\u5bf9\u4e8e\u6587\u672c\u5206\u7c7b\u6765\u8bf4\uff0c\u76f4\u63a5\u4f7f\u7528\u8fd9\u7c7b\u5f3a\u6a21\u578b+\u597d Prompt \uff0c\u76f8\u6bd4\u4e8e\u201c\u8bad\u7ec3\u201d\u4e00\u4e2a\u57fa\u7840\u6a21\u578b\uff0c\u662f\u4e0d\u662f\u4e00\u79cd\u66f4\u9ad8\u6548\uff08\u5f00\u53d1\u65f6\u95f4\u77ed\u3001\u53ef\u80fd\u6548\u679c\u8fd8\u597d\uff09\u7684\u9009\u62e9\uff1f\u5728\u8fd9\u79cd\u60c5\u51b5\u4e0b\uff0c\u5e94\u8be5\u5982\u4f55\u9a8c\u8bc1\u5206\u7c7b\u7684\u51c6\u786e\u6027\uff1f
\u6211\u4e3b\u8981\u60f3\u7406\u89e3\u8fd9\u4e24\u79cd\u6280\u672f\u8def\u5f84\u7684\u9002\u7528\u573a\u666f\u3001\u4f18\u7f3a\u70b9\u4ee5\u53ca\u5b9e\u9645\u64cd\u4f5c\u4e2d\u7684\u8003\u91cf\u3002\u4efb\u4f55\u7ecf\u9a8c\u5206\u4eab\u6216\u5efa\u8bae\u90fd\u5c06\u5bf9\u6211\u975e\u5e38\u6709\u5e2e\u52a9\uff01\u8c22\u8c22\u5927\u5bb6\uff01", "date_published": "2025-04-17T04:00:32+00:00", "title": "\u4f7f\u7528\u5927\u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u6587\u672c\u5206\u7c7b\u4efb\u52a1\u9700\u8981\u5fae\u8c03\u5417\uff1f", "id": "t/1126133" }, { "author": { "url": "member/julyclyde", "name": "julyclyde", "avatar": "https://cdn.v2ex.com/gravatar/4dae02ea3d84a8c34fec215b261bfb5f?s=73&d=retro" }, "url": "t/1125074", "date_modified": "2025-04-15T00:35:37+00:00", "content_html": "

\u602a\u4e0d\u5f97\u770b\u8d77\u6765\u90a3\u4e48\u76f8\u4f3c\u5462

https://lima-vm.io/docs/faq/colima/\nColima is a third-party project that wraps Lima to provide an alternative user experience for launching containers.

\n", "date_published": "2025-04-13T04:06:31+00:00", "title": "lima \u548c colima \u5c45\u7136\u662f\u6709\u5173\u7cfb\u7684", "id": "t/1125074" }, { "author": { "url": "member/lthero", "name": "lthero", "avatar": "https://cdn.v2ex.com/gravatar/77b27233e67963633225abbccf560168?s=73&d=retro" }, "url": "t/1124516", "date_modified": "2025-04-11T08:38:23+00:00", "content_html": "

gpt4o \u56fe\u50cf\u751f\u6210\u7684\u7279\u70b9\u662f\uff0c\u751f\u6210\u65f6\u4ece\u4e0a\u5230\u4e0b\u9010\u6e10\u6e05\u6670\u5316\uff08\u5e76\u4e0d\u53ea\u662f\u663e\u793a\u6280\u5de7\uff09

$\"s1\"$

\u5982\u679c\u4f7f\u7528 diffusion \u8fdb\u884c\u751f\u6210\uff0c\u5b83\u7684\u8fc7\u7a0b\u53ef\u80fd\u662f\u8fd9\u6837\u7684

$\"s2\"$

\u4f46\u5df2\u77e5\u7684\u662f gpt4o \u56fe\u50cf\u751f\u6210\uff08\u4f3c\u4e4e\uff09\u5df2\u7ecf\u8f6c\u5411 autoregressive(\u81ea\u56de\u5f52\u6a21\u578b)+transformer

$\"s3\"$

\u76ee\u524d\u5916\u7f51\u4e5f\u5bf9 gpt4o \u7684\u6280\u672f\u8fdb\u884c\u4e86\u731c\u6d4b\uff0c\u4f46\u4e5f\u6ca1\u8ba8\u8bba\u51fa\u4e2a\u7ed3\u679c\u6765\uff08\u5927\u591a\u662f\u8ba4\u540c\u8f6c\u5411\u4e86 ar \u6a21\u578b\uff09

\u81ea\u56de\u5f52\u6a21\u578b\u662f\u8981\u6253\u8d25 diffusion \uff0c\u5e76\u5728\u591a\u6a21\u6001\u9886\u57df\u53c8\u597d\u7528\u8d77\u6765\u4e86\u5417\uff1f

\u53e6\u5916\uff0c\u76ee\u524d\u5f00\u6e90\u754c\u4f3c\u4e4e\u8fd8\u6ca1\u6709\u4ec0\u4e48\u52a8\u9759\uff0c\u56fd\u5185\u7684\u5b57\u8282\u8df3\u52a8\u5728 ar \u7684\u56fe\u50cf\u751f\u6210\u9886\u57df\u63a2\u7d22\u5f97\u8fd8\u633a\u591a\uff08\u53d1\u4e86\u4e0d\u5c11 paper \uff09

\n", "date_published": "2025-04-10T08:30:32+00:00", "title": "gpt4o \u56fe\u50cf\u751f\u6210\u7684\u6280\u672f\u8ba8\u8bba(\u81ea\u56de\u5f52\u6a21\u578b\u53c8\u597d\u8d77\u6765\u4e86?)", "id": "t/1124516" }, { "author": { "url": "member/huangyezhufeng", "name": "huangyezhufeng", "avatar": "https://cdn.v2ex.com/avatar/c6bf/ce8d/174807_large.png?m=1732972431" }, "url": "t/1123968", "date_modified": "2025-04-08T07:04:02+00:00", "content_html": "

TLDR

\n
WHAT: \u4ecb\u7ecd\u5e76\u590d\u73b0 DeepMind \u7684\u4e00\u7bc7\u5173\u4e8e LLM Speculative Sampling \u7684\u8bba\u6587\uff1aAccelerating large language model decoding with speculative sampling. \u6211\u4eec\u5c06\u7528\u4e0d\u5230 100 \u884c\u4ee3\u7801\u6765\u590d\u73b0\u8fd9\u7bc7\u8bba\u6587\uff0c\u5e76\u5f97\u5230 2 \u500d\u4ee5\u4e0a\u7684\u901f\u5ea6\u63d0\u5347\u3002
\n
\n
\u4eae\u70b9\uff1a\u57fa\u4e8e GPT2, \u4ee3\u7801\uff0c\u6a21\u578b\u6743\u91cd\u5168\u90e8\u53ef\u4ee5\u4e0b\u8f7d\u5e76\u672c\u5730\u8fd0\u884c\uff1b\u53ea\u9700\u8981 16GB \u7684\u663e\u5b58\u5373\u53ef\u5b8c\u6574\u672c\u5730\u590d\u73b0\u3002
\n
\n
\u535a\u5ba2: https://datahonor.com/blog/2025/03/08/llm_sps
\n
\n
\u516c\u4f17\u53f7\u6587\u7ae0(\u5185\u5bb9\u540c\u535a\u5ba2\uff0c\u4fbf\u4e8e\u6536\u85cf)\uff1a https://mp.weixin.qq.com/s/3rFk8cgJuxjW30A4-emhEA
\n
\n
\u4ee3\u7801: https://github.com/ai-glimpse/toyllm/tree/master/toyllm/sps
\n

\u5177\u4f53\u5b9e\u73b0

$\"\"$

\n", "date_published": "2025-04-08T07:00:48+00:00", "title": "[AI Glimpse \u79d1\u666e\u7cfb\u5217] \u4e0d\u5230 100 \u884c Python \u4ee3\u7801\u4ece\u96f6\u5b9e\u73b0 LLM \u7684\u63a8\u6d4b\u91c7\u6837\u7b97\u6cd5", "id": "t/1123968" }, { "author": { "url": "member/albertxavier001", "name": "albertxavier001", "avatar": "https://cdn.v2ex.com/gravatar/4533978ef5a9324fdc15f47229b69d5e?s=73&d=retro" }, "url": "t/1123593", "title": "\u4ec0\u4e48\uff1f\u5520\u5520\u55d1\u5c31\u80fd\u751f\u6210\u6e38\u620f\u5173\u5361! Unreal Engine + MCP \u6765\u4e86\uff01", "id": "t/1123593", "date_published": "2025-04-06T23:08:10+00:00", "content_html": "

$\"\u8bf7\u6dfb\u52a0\u56fe\u7247\u63cf\u8ff0\"$

\n
\u5373\u4f7f\u6ca1\u6709\u7f16\u7a0b\u57fa\u7840\uff0c\u53ea\u8981\u6000\u63e3\u72ec\u7279\u521b\u610f\uff0cMCP+Unreal \u4e5f\u80fd\u52a9\u4f60\u5c06\u60f3\u8c61\u53d8\u4e3a\u89e6\u624b\u53ef\u53ca\u7684\u7cbe\u54c1\u826f\u4f5c
\n

\u60f3\u5fc5\u5173\u6ce8\u6e38\u620f\u5f00\u53d1\u9886\u57df\u7684\u72ec\u7acb\u6e38\u620f\u5f00\u53d1\u8005\u5df2\u7ecf\u6ce8\u610f\u5230\u4e86 blender-mcp \u8fd9\u4e2a\u9879\u76ee \u5b83\u5141\u8bb8 Blender \u8fde\u63a5\u5230 Claude AI \uff0c\u5141\u8bb8 Claude \u76f4\u63a5\u4e0e Blender \u4ea4\u4e92\u548c\u63a7\u5236\uff0c\u4f7f\u5373\u65f6\u8f85\u52a9 3D \u5efa\u6a21\u3001\u573a\u666f\u521b\u5efa\u548c\u64cd\u4f5c\u6210\u4e3a\u53ef\u80fd\u3002

\u73b0\u5728\uff0c\u901a\u8fc7 UnrealMCP \u63d2\u4ef6\u548c Python Editor Script \u63d2\u4ef6\uff0c\u63a7\u5236\u865a\u5e7b\u5f15\u64ce\u5236\u4f5c\u6e38\u620f\u573a\u666f POC \u4e5f\u6210\u4e3a\u4e86\u53ef\u80fd\u3002

\u6548\u679c\u5c55\u793a\ud83e\udd73

\n
\u5520\u5520\u55d1\u5c31\u80fd\u751f\u6210\u6e38\u620f\u5173\u5361\uff0c\u8c01\u80fd\u4e0d\u7231\ud83d\ude18\uff1f
\n

$\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$ \n $\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\u767d\u8272\u7684\u5c0f\u4e0d\u70b9\u662f\u5c0f\u72d7\ud83d\udc15\u54e6

$\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\u4ece\u5730\u9762\u4e0b\u770b\u770b~

\u914d\u7f6e\u6b65\u9aa4\ud83e\udd16

01 \u786e\u4fdd Python Editor Script \u63d2\u4ef6\u5df2\u542f\u7528

\u6253\u5f00 Settings/Plugins

$\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\u641c\u7d22\u5e76\u52fe\u9009 Python Editor Script Plugin\n $\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

02 \u5b89\u88c5 UnrealMCP \u63d2\u4ef6

\n
UnrealMCP \u662f\u4e00\u4e2a\u975e\u5b98\u65b9\u7684\u865a\u5e7b\u5f15\u64ce\u63d2\u4ef6\uff0c\u65e8\u5728\u901a\u8fc7\u4eba\u5de5\u667a\u80fd\u5de5\u5177\u63a7\u5236\u865a\u5e7b\u5f15\u64ce\u3002\u5b83\u5728\u865a\u5e7b\u5f15\u64ce\u5185\u90e8\u5b9e\u73b0\u4e86\u4e00\u79cd\u673a\u5668\u63a7\u5236\u534f\u8bae (MCP)\uff0c\u5141\u8bb8\u5916\u90e8\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\u4ee5\u7f16\u7a0b\u65b9\u5f0f\u4e0e\u865a\u5e7b\u73af\u5883\u8fdb\u884c\u4ea4\u4e92\u548c\u64cd\u4f5c
\n

\u5728\u9879\u76ee\u6839\u76ee\u5f55\u4e0b\u521b\u5efa Plugins \u63d2\u4ef6\n $\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\u5728 Plugin \u76ee\u5f55\u4e0b\u6253\u5f00\u547d\u4ee4\u884c\u5e76\u8fd0\u884c

git clone <https://github.com/kvick-games/UnrealMCP>

\u786e\u4fdd\u5728 Plugins\\UnrealMCP \u76ee\u5f55\u4e0b\u5305\u542b GitHub \u4e0a\u7684\u6587\u4ef6

$\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\u91cd\u542f\u865a\u5e7b\u5f15\u64ce\u7f16\u8f91\u5668\uff0c\u5728 Settings/Plugins \u4e2d\u641c\u7d22\u5e76\u52fe\u9009 UnrealMCP \u63d2\u4ef6\uff08\u540c\u7b2c\u4e00\u6b65\uff09

03 \u914d\u7f6e UnrealMCP Server

Plugins\\UnrealMCP\\MCP \u4e2d\u8fd0\u884c setup_unreal_mcp.bat \u811a\u672c

04 \u5c06\u9879\u76ee\u8f6c\u6362\u6210 C++\u9879\u76ee

\n
\u8fd9\u4e00\u6b65\u4e3b\u8981\u662f\u4e3a\u4e86\u7f16\u8bd1\u7b2c\u4e8c\u6b65\u4e0b\u8f7d\u7684\u63d2\u4ef6\uff0c\u4e0d\u9700\u8981\u771f\u7684\u53bb\u5199 C++
\n

\u65b0\u5efa\u4e00\u4e2a C++ class \u5373\u53ef\uff0c\u540e\u7eed\u6839\u636e UI \u521b\u5efa\u4e00\u4e2a\u9ed8\u8ba4\u7c7b\u5373\u53ef\n $\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\u6253\u5f00 Visual Studio

$\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\u5173\u95ed\u865a\u5e7b\u5f15\u64ce\u7f16\u8f91\u5668\uff0c\u7f16\u8bd1\u9879\u76ee\n $\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\u91cd\u65b0\u6253\u5f00\u865a\u5e7b\u5f15\u64ce\u7f16\u8f91\u5668\uff0c\u70b9\u51fb\u5de5\u5177\u680f\u6700\u53f3\u4fa7\u7684\u56fe\u6807\u6253\u5f00 MCP Server Control Panel

\u70b9\u51fb Start Server\n $\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

Server Status \u53d8\u6210 Running \u8bf4\u660e Unreal MCP Server \u53ef\u4ee5\u8fd0\u884c\u4e86\n $\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

05 \u5728 Cursor \u4e2d\u6dfb\u52a0 UnrealMCP Server

\u6253\u5f00 Cursor Settings\n $\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\u70b9\u51fb + Add new global MCP server

$\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\u5728\u201cmcpServer\u201d\u4e2d\u52a0\u5165\u914d\u7f6e

\"unreal\": {\n \"command\": \"powershell\",\n \"args\": [\"<YOUR_GAME_ROOT>/Plugins/UnrealMCP/MCP/run_unreal_mcp.bat\"]\n}\n

\u5728 Cursor Settings \u4e2d\u51fa\u73b0\u4e0b\u56fe\u8bf4\u660e\u6dfb\u52a0\u6210\u529f\n $\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\u4e4b\u540e\u5c31\u53ef\u4ee5\u6109\u5feb\u7684\u5728 Chat \u4e2d\u6109\u5feb\u7684\u8ba9 AI \u5e2e\u6211\u4eec\u5728\u865a\u5e7b\u5f15\u64ce\u4e2d\u521b\u5efa POC \u573a\u666f\u5566~~~

$\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\u9879\u76ee\u9650\u5236\ud83e\udd72

\u7531\u4e8e UnrealMCP \u63d2\u4ef6\u9879\u76ee\u8fd8\u5728\u975e\u5e38\u65e9\u671f\u7684\u9636\u6bb5\uff0c\u73b0\u5728\u4ec5\u53ef\u652f\u6301\u6709\u9650\u7684\u57fa\u672c\u64cd\u4f5c\uff1a\u83b7\u53d6\u573a\u666f\u57fa\u672c\u4fe1\u606f\uff0cPython \u811a\u672c\u8fd0\u884c\uff0c\u57fa\u7840\u6750\u8d28\u64cd\u4f5c\u7b49\u3002\u800c\u4e14\u5c31\u7b97\u4f7f\u7528\u4e86\u9876\u5c16\u7684\u5927\u8bed\u8a00\u6a21\u578b\u4e5f\u4e0d\u80fd\u4e00\u6b21\u6027\u751f\u6210\u6b63\u786e\u7684 MCP Server \u8c03\u7528\uff0c\u9700\u8981\u53cd\u590d\u4fee\u6b63\u3002

\u5c55\u671b\u672a\u6765\ud83d\ude0d

\u867d\u7136\u76ee\u524d UnrealMCP \u63d2\u4ef6\u8fd8\u6709\u8bf8\u591a\u9650\u5236\uff0c\u4f46\u662f MCP \u8fd8\u662f\u4e3a\u865a\u5e7b\u5f15\u64ce\u6253\u5f00\u4e86\u4e00\u53e5\u8bdd\u751f\u6210\u6e38\u620f\u7684\u5927\u95e8\uff01

\u8bd5\u60f3\u4e00\u4e0b\uff0c\u672a\u6765 UnrealMCP \u652f\u6301\u4e86\u66f4\u591a\u7684 Unreal \u64cd\u4f5c\uff1a\u52a8\u753b\u3001\u5730\u5f62\u3001AI \u3001PCG \u7b49\u7b49\uff0c\u5e76\u4e14\u80fd\u652f\u6301\u84dd\u56fe\u6216\u7b2c\u4e09\u65b9\u6e38\u620f\u5f00\u53d1\u53ef\u70ed\u66f4\u811a\u672c\uff08\u4f8b\u5982\u817e\u8baf\u7684 PuerTS \uff09\uff0c\u90a3\u4e48\u4e0d\u8bba\u662f\u6e38\u620f\u573a\u666f\u5236\u4f5c\u8fd8\u662f gameplay \u903b\u8f91\u7f16\u5199\uff0c\u90fd\u53ef\u4ee5\u901a\u8fc7\u5728 Cursor/Cluade/Windsurf/VSCode \u7b49\u7f16\u8f91\u5668\u4e2d\u901a\u8fc7\u81ea\u7136\u8bed\u8a00\u63cf\u8ff0\u751f\u6210\u6e38\u620f\u3002

\u8fd9\u4e0d\u4ec5\u80fd\u5c06\u6e38\u620f\u5f00\u53d1\u3001\u539f\u578b\u5236\u4f5c\u7684\u6548\u7387\u5927\u5927\u63d0\u9ad8\uff0c\u66f4\u80fd\u5c06\u6e38\u620f\u5f00\u53d1\u95e8\u69db\u5927\u5927\u964d\u4f4e\uff01

\u4e5f\u8bb8\u5728\u672a\u6765\u67d0\u4e00\u5929\uff0c\u8fd9\u6837\u7684\u573a\u666f\u4f1a\u53d8\u6210\u73b0\u5b9e\uff1a\u501f\u52a9 MCP+\u6e38\u620f\u5f15\u64ce\uff0c\u96f6\u57fa\u7840\u521b\u4f5c\u8005\u4e5f\u80fd\u5c06\u7075\u611f\u8f7b\u677e\u8f6c\u5316\u4e3a\u4e13\u4e1a\u7ea7\u54c1\u8d28\u7684\u6e38\u620f\u4f5c\u54c1\u3002

\u6709\u8da3\u6e38\u620f\u8d44\u8baf\ud83d\udc7e\u5f00\u53d1\u5206\u4eab\ud83d\udda5\ufe0f\u5c3d\u5728\u6e38\u620f\u78b0\u78b0\ud83c\udfae

\u5fae\u4fe1\u53f7\uff1a\u6e38\u620f\u78b0\u78b0\n\u626b\u7801\u5173\u6ce8 \u4e86\u89e3\u66f4\u591a

$\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"$

\n" }, { "author": { "url": "member/happyn", "name": "happyn", "avatar": "https://cdn.v2ex.com/avatar/5e3c/57ae/549037_large.png?m=1699836390" }, "url": "t/1121905", "date_modified": "2025-03-31T01:14:45+00:00", "content_html": "\u6211\u7684\u7406\u89e3\u5c31\u662f\u5927\u6a21\u578b\u662f\u4e00\u4e2a Token \u9884\u6d4b\u7684\u7f51\u7edc\uff0c\u901a\u8fc7\u5927\u91cf\u6570\u636e\u7684\u9884\u8bad\u7ec3\uff0c\u4ece\u8f93\u5165\u7684 context \u4e2d\u9884\u6d4b\u4e0b\u4e00\u4e2a\u6700\u6709\u53ef\u80fd\u7684 token \uff1b\u90a3\u6211\u6709\u4e00\u4e2a\u5f88\u5c0f\u767d\u7684\u95ee\u9898\uff0c\u8bf8\u5982\u6570\u5b66\u8fd0\u7b97\uff0c\u6bd4\u5982 673 X 3666 = ? \u8fd9\u79cd\u8fd0\u7b97\uff0c\u5927\u6a21\u578b\u662f\u5982\u4f55\u901a\u8fc7 token \u9884\u6d4b\u5f97\u51fa\u7b54\u6848\u7684\uff1f\u800c\u4e14\u4e4b\u524d\u4e5f\u6709\u4e00\u4e9b\u5f88\u660e\u663e\u7684\u5e94\u7528\uff0c\u6bd4\u5982\u8ba9\u5927\u6a21\u578b\u626e\u6f14\u4e00\u4e2a\u7f16\u8bd1\u5668\u7b49\u7b49\u6267\u884c\u975e\u5e38\u7cbe\u786e\u7684\u8fd0\u7b97\uff0c\u6211\u5b9e\u5728\u60f3\u4e0d\u901a\uff0c\u5982\u679c\u4e0d\u662f\u7531\u4e13\u6709\u7684\u7a0b\u5e8f\uff0c\u4ec5\u4ec5\u4f9d\u9760\u795e\u7ecf\u7f51\u7edc\uff0c\u5c31\u80fd\u6267\u884c\u8fd9\u79cd\u7c7b\u4f3c\u7684\u8fd0\u7b97\u5417\uff1f

\u6211\u8ddf Gemini \u804a\u4e86\u534a\u5929\uff0c\u53d1\u73b0\u4e00\u4e2a\u975e\u5e38\u6709\u610f\u601d\u7684\u4e8b\u60c5\uff0c\u6bd4\u5982\u6211\u95ee\u4ed6 \"\u8bf7\u544a\u8bc9\u6211\u5706\u5468\u7387\u5c0f\u6570\u70b9\u540e x \u4f4d\u7684\u6570\u5b57\"\uff0c\u5f53 10 \u4f4d\uff0c30 \u4f4d\uff0c50 \u4f4d\u7684\u65f6\u5019\uff0c\u90fd\u6ca1\u6709\u95ee\u9898\uff0c\u4f46\u662f\u8d85\u8fc7\u4e00\u5b9a\u91cf\uff0c\u6bd4\u5982 1000 \u4f4d\uff0c\u5b83\u5c31\u4f1a\u5b95\u673a; \u5982\u679c\u7528 deepseek \u7684\u63a8\u7406\u6a21\u5f0f\uff0c\u4ed6\u5c31\u4f1a\u81ea\u5df1\u8ba1\u7b97\uff1b\u6240\u4ee5\u6211\u7684\u7406\u89e3\u662f\uff1a

1. \u5927\u6a21\u578b\u7406\u89e3\u95ee\u9898\uff0c\u662f\u9760\u795e\u7ecf\u7f51\u7edc\u8fdb\u884c token \u9884\u6d4b\u7684
2. \u5927\u6a21\u578b\u89e3\u51b3\u6570\u5b66\u7c7b\u7684\u7cbe\u786e\u95ee\u9898\uff0c\u5fc5\u987b\u662f\u6df7\u5408\u6a21\u578b(MoE)\uff0c\u8c03\u7528\u4e13\u95e8\u5904\u7406\u7cbe\u786e\u8ba1\u7b97\u7684\u90a3\u4e2a\u90e8\u5206\uff0c\u624d\u80fd\u5f97\u51fa\u6b63\u786e\u7b54\u6848\uff1f

\u8fd9\u4e2a\u7406\u89e3\u5bf9\u5417\uff1f", "date_published": "2025-03-29T02:46:08+00:00", "title": "\u5927\u6a21\u578b\u662f\u5982\u4f55\u6267\u884c\u50cf\u6570\u5b66\u8fd0\u7b97\u3001\u7f16\u7a0b\u7b49\u7cbe\u786e\u6027\u8981\u6c42\u6bd4\u8f83\u9ad8\u7684\u4efb\u52a1\u7684\uff1f", "id": "t/1121905" }, { "author": { "url": "member/ojh", "name": "ojh", "avatar": "https://cdn.v2ex.com/gravatar/b2649469f9eeccba1a86613033307c0f?s=73&d=retro" }, "url": "t/1121545", "date_modified": "2025-03-28T01:16:43+00:00", "content_html": "

\u6211\u60f3\u7cfb\u7edf\u5b66\u4e60\u4e00\u4e0b LLM \u7684\u4f7f\u7528\u6bd4\u5982\u53d1\u8d77\u8bf7\u6c42\u3001\u4e00\u4e9b\u57fa\u672c\u6982\u5ff5\u6bd4\u5982\u6e29\u5ea6\u3001\u4e0a\u4e0b\u6587\u7a97\u53e3\u3001MCP \u7b49\uff0c\u4f46\u4e0d\u9700\u8981\u6df1\u5165\u5b66\u4ec0\u4e48\u4eba\u5de5\u667a\u80fd\u65b9\u9762\u4e13\u4e1a\u7684\u77e5\u8bc6\uff0c\u7f51\u53cb\u4eec\u6709\u6ca1\u6709\u4ec0\u4e48\u597d\u7684\u6559\u7a0b\u6216\u8005\u4e66\u63a8\u8350

\n", "date_published": "2025-03-27T08:04:12+00:00", "title": "\u6709\u6ca1\u6709 LLM \u5165\u95e8\u7684\u7cfb\u7edf\u5b66\u4e60\u6559\u7a0b", "id": "t/1121545" }, { "author": { "url": "member/sugarkeek", "name": "sugarkeek", "avatar": "https://cdn.v2ex.com/avatar/dbe9/73b9/172054_large.png?m=1745726832" }, "url": "t/1121113", "title": "mtranserver \u7684\u7ffb\u8bd1\u6548\u679c\u8fd8\u662f\u5dee\u4e00\u70b9", "id": "t/1121113", "date_published": "2025-03-26T02:20:50+00:00", "content_html": "\u5de6\u8fb9 google \uff0c\u53f3\u8fb9 mtranserver\uff1a

" }, { "author": { "url": "member/9527kf", "name": "9527kf", "avatar": "https://cdn.v2ex.com/gravatar/968b1c0586f01ad2f58e87a08bc87c15?s=73&d=retro" }, "url": "t/1118001", "date_modified": "2025-03-13T00:35:51+00:00", "content_html": "\u6211\u6709\u4e00\u4e9b\u8bbe\u8ba1\u56fe\uff0c\u60f3\u7528\u89c6\u89c9\u5927\u6a21\u578b\u5c06\u56fe\u4e2d\u6709\u54ea\u4e9b\u6a21\u5757\uff0c\u5404\u4e2a\u6a21\u5757\u7684\u4f4d\u7f6e\u4fe1\u606f\u7ed9\u6807\u51fa\u6765\uff08\u5de6\u4e0a\u89d2\u4f4d\u7f6e\u548c\u53f3\u4e0b\u89d2\u4f4d\u7f6e\uff09\u3002
\u8bd5\u4e86\u51e0\u4e2a\u6a21\u578b\uff0c\u80fd\u8bc6\u522b\u51fa\u6765\u6709\u51e0\u4e2a\u6a21\u5757\uff0c\u4f46\u4e0d\u80fd\u51c6\u786e\u6807\u51fa\u4f4d\u7f6e\u4fe1\u606f\u3002
\u5927\u5bb6\u6709\u4e86\u89e3\u8fd9\u5757\u7684\u5417", "date_published": "2025-03-13T00:34:22+00:00", "title": "\u6c42\u52a9\uff0c\u6709\u6ca1\u6709\u89c6\u89c9\u5927\u6a21\u578b\u53ef\u4ee5\u8bc6\u522b\u4e00\u4e2a UI \u8bbe\u8ba1\u56fe\u4e2d\u5404\u4e2a\u6a21\u5757\u7684\u5212\u5206", "id": "t/1118001" }, { "author": { "url": "member/565656", "name": "565656", "avatar": "https://cdn.v2ex.com/gravatar/8bd1911a5b05e20fd1233392ebbcf4dc?s=73&d=retro" }, "url": "t/1115294", "date_modified": "2025-03-03T00:36:36+00:00", "content_html": "

\u56fe\u6587\u5927\u6a21\u578b\u5f00\u6e90\u7684\u4e00\u822c\u5c31 3b \uff0c7b \uff0c72b \u3002\u5e94\u8be5\u662f\u67d0\u79cd\u539f\u56e0\u9650\u5236\u4e86\u591a\u6a21\u6001\u6a21\u578b\u7684\u5927\u5c0f\uff1f\u6bd4\u5982\u56fe\u7247\u4e0d\u50cf\u6587\u5b57\u4e00\u6837\u81ea\u56de\u5f52\u9884\u8bad\u7ec3\u5927\u6a21\u578b\uff1f\u5355\u6587\u672c\u7684 600b \u90fd\u6709\uff0c\u8fd9\u662f\u4e0d\u662f\u610f\u5473\u7740 gpt4o \u5176\u5b9e\u6a21\u578b\u5e76\u6ca1\u6709 gpt4 \u5927\uff1f\u90a3\u5bf9\u6bd4\u7684\u65f6\u5019\u8fd9\u4e9b\u591a\u6a21\u6001\u6a21\u578b\u5e94\u8be5\u6253\u4e0d\u8fc7\u5355\u6587\u5b57\u6a21\u578b\u5427\uff1f

\n", "date_published": "2025-03-02T12:31:42+00:00", "title": "\u591a\u6a21\u6001\u5927\u6a21\u578b\u7684\u5927\u5c0f\u8fdc\u4f4e\u4e8e\u5355\u6587\u672c\u6a21\u578b\u554a\u3002", "id": "t/1115294" }, { "author": { "url": "member/rateltalk", "name": "rateltalk", "avatar": "https://cdn.v2ex.com/gravatar/f475e9ec3926841965209f62cc90a1d4?s=73&d=retro" }, "url": "t/1115191", "date_modified": "2025-03-02T02:08:03+00:00", "content_html": "

\u65b0\u95fb\u94fe\u63a5\uff1a https://mp.weixin.qq.com/s/p3msf_1p4uwTdZFTH8JNdA

Deepseek \u662f\u6587\u672c\u751f\u6210\u7684\uff0c\u4ed6\u662f\u600e\u4e48\u7ed3\u5408\u533b\u5b66\u5f71\u50cf\u627e\u51fa\u7ed3\u8282\u7684\uff1f

\n", "date_published": "2025-03-02T01:39:26+00:00", "title": "\u5229\u7528 deepseek \u5206\u6790\u533b\u5b66\u5f71\u50cf\u627e\u51fa\u7ed3\u8282\u8fd9\u662f\u4ec0\u4e48\u6280\u672f\u539f\u7406\uff1f", "id": "t/1115191" }, { "author": { "url": "member/huangyezhufeng", "name": "huangyezhufeng", "avatar": "https://cdn.v2ex.com/avatar/c6bf/ce8d/174807_large.png?m=1732972431" }, "url": "t/1113214", "title": "\u5927\u8bed\u8a00\u6a21\u578b\u4e0e\u6df1\u5ea6\u5b66\u4e60\u4e66\u7c4d\u63a8\u8350", "id": "t/1113214", "date_published": "2025-02-21T05:26:30+00:00", "content_html": "

TLDR

LLM \u76f8\u5173

Build a Large Language Model (From Scratch)
Super Study Guide: Transformers & Large Language Models
Natural Language Processing with Transformers

DL \u76f8\u5173

Neural Networks and Deep Learning(NNDL)
Neural Networks from Scratch in Python(NNFS)
Dive into Deep Learning(D2L)
Grokking Deep Learning(\u524d 6 \u7ae0)

\u8be6\u60c5

\u535a\u5ba2\u539f\u6587: \u6709\u7b80\u5355\u4e66\u8bc4\u548c\u63a8\u8350\u8bed
\u516c\u4f17\u53f7\uff1a\u4e3b\u8981\u662f\u65b9\u4fbf\u5927\u5bb6\u6536\u85cf\u67e5\u9605(\u56e0\u4e3a\u5f88\u591a\u670b\u53cb\u5176\u5b9e\u4e0d\u592a\u4e60\u60ef\u8ba2\u9605\u535a\u5ba2)

\n" }, { "author": { "url": "member/xuchkang171", "name": "xuchkang171", "avatar": "https://cdn.v2ex.com/avatar/aab2/f15c/65377_large.png?m=1712948533" }, "url": "t/1112497", "date_modified": "2025-02-20T02:37:20+00:00", "content_html": "

", "date_published": "2025-02-19T01:24:13+00:00", "title": "\u4e3a\u4ec0\u4e48 LLM \u4e0d\u64c5\u957f\u6570\u997a\u5b50\uff1f", "id": "t/1112497" }, { "author": { "url": "member/pinocc012", "name": "pinocc012", "avatar": "https://cdn.v2ex.com/gravatar/b648af74d4390346f34ee7b23dc8651a?s=73&d=retro" }, "url": "t/1112229", "title": "\u7a81\u7136\u5192\u51fa\u4e00\u4e2a\u70b9\u5b50\uff0c\u5229\u7528\u73b0\u5728\u70ed\u95e8\u7684 AI \u6765\u505a web \u5916\u6302\u5de5\u5177", "id": "t/1112229", "date_published": "2025-02-18T02:46:46+00:00", "content_html": "

\u73b0\u5728\u5f88\u591a\u65e5\u5e38\u5de5\u4f5c\uff0c\u662f\u5728\u5404\u4e2a web \u5e94\u7528\u4e4b\u95f4 copy paste \uff0c\u7406\u8bba\u4e0a\u5e94\u7528\u4e4b\u95f4\u6570\u636e\u6253\u901a\u53ef\u80fd\u5c31\u6ca1\u8fd9\u4e9b\u7834\u4e8b\u4e86\uff0c\u4f46\u662f\u73b0\u5b9e\u4e2d\u8fd8\u662f\u4e0d\u5c11\u9700\u8981\u4eba\u53bb\u5e72\n\u6240\u4ee5\u5982\u679c\u80fd\u8bad\u7ec3 AI \u6765\u5e72\u8fd9\u4e2a\u4e8b\u60c5\u6211\u4eec\u5c31\u80fd\u591f\u66f4\u6109\u5feb\u7684\u6478\u9c7c\n\u8981\u505a\u7684\u5f00\u53d1\u5927\u6982\u662f\u6cb9\u7334\u811a\u672c\u5bf9\u63a5\u5927\u6a21\u578b\uff1f

\n" }, { "author": { "url": "member/ixixi", "name": "ixixi", "avatar": "https://cdn.v2ex.com/avatar/6879/5a6d/456967_large.png?m=1768011429" }, "url": "t/1110797", "date_modified": "2025-02-13T02:14:26+00:00", "content_html": "

\u76ee\u524d\u6709 macmini m4 \u548c i7-12700 + amd \u7684\u663e\u5361 \u60f3\u4e70\u4e2a n \u5361\u73a9\u73a9 ai

\u6c42\u73a9\u8fc7\u7684\u8001\u54e5\u8bf4\u4e0b\u7ecf\u9a8c

\n", "date_published": "2025-02-11T16:07:05+00:00", "title": "\u8bf7\u95ee\u8bad\u7ec3\u6216\u8005\u8c03\u8bd5 yolo \u7528\u4ec0\u4e48\u914d\u7f6e\u6027\u4ef7\u6bd4\u8f83\u9ad8\u3002", "id": "t/1110797" }, { "author": { "url": "member/lurui45", "name": "lurui45", "avatar": "https://cdn.v2ex.com/avatar/e858/0876/266961_large.png?m=1741526607" }, "url": "t/1107038", "date_modified": "2025-01-23T01:05:38+00:00", "content_html": "

\u5927\u6a21\u578b\u80fd\u8bb0\u4f4f\u5b83\u8bad\u7ec3\u8fc7\u7684\u6570\u636e\u5417\uff1f

\u5982\u679c\u4f60\u8ba9 GPT \u80cc\u8bf5\u7ea2\u697c\u68a6\u5168\u6587\uff0c\u5b83\u4f1a\u56de\u7b54\u5b83\u65e0\u6cd5\u9010\u5b57\u9010\u53e5\u80cc\u8bf5\uff0c\u56e0\u4e3a\u8d85\u8fc7\u5b83\u7684\u751f\u6210\u5bb9\u91cf\u9650\u5236\uff0c\u4f46\u662f\u5982\u679c\u4f60\u9700\u8981\u67d0\u4e00\u90e8\u5206\u7684\u5185\u5bb9\uff08\u6bd4\u5982\u67d0\u4e00\u56de\u7684\u7247\u6bb5\u3001\u67d0\u4e2a\u7ecf\u5178\u573a\u666f\u6216\u5bf9\u8bdd\uff09\uff0c\u5b83\u53ef\u4ee5\u6839\u636e\u4f60\u7684\u9700\u6c42\u751f\u6210\u76f8\u5e94\u7684\u6bb5\u843d\u6216\u7ae0\u8282\u3002

\u6240\u4ee5\u5b83\u7684\u6570\u636e\u5e93\u91cc\u9762\u662f\u771f\u7684\u5f55\u8fdb\u53bb\u5168\u6587\u4e86\u5417\uff1f\u7ea2\u697c\u68a6\u5168\u6587 80 \u4e07\u5b57\uff0c\u5f53\u6211\u4eec\u8bf4\u5230\u67d0\u4e2a\u6a21\u578b\u8981\u591a\u5c11 Billion \u53c2\u6570\u65f6\uff0c\u8fd9 80 \u4e07\u5b57\u80fd\u5426\u7406\u89e3\u4e3a\u5176\u4e2d\u7684\u4e00\u90e8\u5206\u53c2\u6570\uff1f

\u6211\u95ee\u4e86 GPT \uff0c\u5b83\u56de\u7b54\u5b83\u901a\u8fc7\u5b66\u4e60\u8fd9\u4e9b\u6587\u672c\u7684\u89c4\u5f8b\u548c\u6a21\u5f0f\uff0c\u80fd\u591f\u5728\u9700\u8981\u65f6\u751f\u6210\u76f8\u5e94\u7684\u5185\u5bb9\u3002\u603b\u4e4b\u65e0\u6cd5\u7406\u89e3...

\n", "date_published": "2025-01-22T04:19:15+00:00", "title": "\u5927\u6a21\u578b\u80fd\u8bb0\u4f4f\u6240\u6709\u5b83\u8bad\u7ec3\u8fc7\u7684\u6570\u636e\u5417\uff1f", "id": "t/1107038" }, { "author": { "url": "member/lenqu", "name": "lenqu", "avatar": "https://cdn.v2ex.com/gravatar/44fb46f46fdc8605ec3bf7b22952e752?s=73&d=retro" }, "url": "t/1104401", "date_modified": "2025-01-17T13:29:55+00:00", "content_html": "

\u6211\u6700\u8fd1\u95f2\u6765\u65e0\u4e8b\uff0c\u60f3\u7cfb\u7edf\u6027\u7684\u5b66\u4e60\u4e00\u4e0b\u795e\u7ecf\u7f51\u7edc\uff0c\u4e4b\u524d\u770b\u5230\u7684\u90fd\u5f88\u7247\u9762\uff0c\u5e0c\u671b\u6709\u61c2\u7684\u63a8\u8350\u4e00\u4e0b

\n", "date_published": "2025-01-11T13:10:05+00:00", "title": "\u5982\u4f55\u4ece\u5934\u6784\u5efa\u4e00\u4e2a\u81ea\u5df1\u7684\u5927\u6a21\u578b\u5462\uff1f\u4ece\u5e95\u5c42\u6700\u57fa\u7840\u7684\u795e\u7ecf\u7f51\u7edc\u5f00\u59cb\u5b9e\u73b0", "id": "t/1104401" }, { "author": { "url": "member/huc2", "name": "huc2", "avatar": "https://cdn.v2ex.com/avatar/c8ee/a584/614681_large.png?m=1765604377" }, "url": "t/1104386", "date_modified": "2025-01-17T13:28:51+00:00", "content_html": "

\u95ee\u9898\u6765\u81ea\u5728\u770b\u8fd9\u7bc7\u535a\u5ba2\u65f6\u770b\u5230\u7684\u56fe\u7247

$\"img1\"$

\u8fd9\u5f20\u56fe\u7247\u4e2d\u662f\u9759\u6001 batch \u7684\u793a\u610f\u56fe\uff0c\u4f46\u6211\u7684\u7406\u89e3\u4f3c\u4e4e\u6709\u6240\u504f\u5dee\uff0c\u5e0c\u671b\u6709\u5927\u4f6c\u80fd\u7b54\u7591\u89e3\u60d1\u3002

\u95ee\u9898 1\uff1a\u5bf9\u4e8e\u9759\u6001 batch \u573a\u666f\uff0c\u540c\u4e00\u4e2a batch \u4e2d\u4e0d\u540c\u7684 sample \u7684 prefill \u662f\u540c\u65f6\u5b8c\u6210\u7684\u5417\uff1f

\u6211\u7684\u7406\u89e3\uff1a\n\u5bf9\u4e8e\u4e00\u4e2a\u5355\u72ec\u7684\u77ed prompt \uff0cprefill \u9636\u6bb5\u80af\u5b9a\u662f\u4e00\u4e2a\u6bd4\u5f88\u957f\u7684 prompt \u5feb\u7684\u3002\u4f46\u662f\u5f53\u8fd9 2 \u4e2a\u957f\u5ea6\u4e0d\u540c\u7684 sample \u7ecf\u8fc7 padding \u7136\u540e\u62fc\u6210\u4e86\u4e00\u4e2a\u9759\u6001 batch \u4e4b\u540e\uff0c\u4e5f\u5c31\u662f\u7ef4\u5ea6\u53d8\u6210\u4e86 [batch_size, ..] \uff08\u5f53\u7136\u8fd9\u91cc\u53ef\u80fd\u4e0d\u6b62 2 \u4e2a sample \uff09\u3002\u5728 prefill \u9636\u6bb5\uff0c\u4ed6\u4eec\u80af\u5b9a\u662f\u540c\u65f6\u5f00\u59cb\u7684\uff0c\u56e0\u4e3a transformers \u5185\u90e8\u662f\u5f88\u591a\u7684\u77e9\u9635\u4e58\u6cd5\u3002\u5e76\u4e14\u8981\u7ecf\u8fc7\u5f88\u591a\u5c42\uff0c\u6bd4\u5982\uff1a

emb = layer1(emb)\nemb = layer2(emb)\n...\n

\u4e0a\u9762\u7684 emb \u7684\u7ef4\u5ea6\u5e94\u8be5\u4e5f\u90fd\u4f1a\u662f [batch_size, ....]

\u867d\u7136\u540c\u4e00\u4e2a batch \u4e4b\u95f4\u4e0d\u540c sample \u5355\u72ec\u505a prefill \u9700\u8981\u7684\u65f6\u95f4\u4e0d\u540c\uff0c\u4f46\u662f\u5f53\u4ed6\u4eec\u6210\u4e3a\u4e00\u4e2a batch \u4e4b\u540e\uff0c\u53d8\u6210\u4e86\u4e00\u4e2a\u5927\u77e9\u9635\uff0c\u4ed6\u4eec\u5728\u7ecf\u8fc7\u4e0d\u540c\u7684 layer \u7684\u65f6\u5019\uff0c\u90fd\u662f\u4e00\u8d77\u4e00\u5c42\u4e00\u5c42\u8fc7\u7684\uff0c\u4e5f\u5c31\u76f8\u5f53\u4e8e\u4ed6\u4eec\u540c\u65f6\u5f00\u59cb prefill \u9636\u6bb5\uff0c\u7136\u540e\u540c\u65f6\u5b8c\u6210 layer1, \u540c\u65f6\u5b8c\u6210 layer2, ......\u3002\u6700\u540e\u4e00\u8d77\u5b8c\u6210\u6700\u540e\u4e00\u5c42\uff0c\u83b7\u5f97\u7b2c\u4e00\u4e2a\u9884\u6d4b\u7684 new_tokens \uff08 batch \u4e2d\u6bcf\u4e2a sample \u90fd\u6709\u4e00\u4e2a new_token)\u3002\u6240\u4ee5\uff0c\u5728\u6211\u7684\u7406\u89e3\u4e2d\uff0cprefill \u9636\u6bb5\u5e94\u8be5\u662f\u5de6\u53f3\u5bf9\u9f50\u7684\u3002

\u95ee\u9898 2: \u4ee5\u4e0b\u6211\u5bf9 Continuous Batch \u7684\u7406\u89e3\u662f\u5426\u6b63\u786e\uff1f

\u6211\u7684\u7406\u89e3\uff1a

\u9996\u5148\u6a21\u578b\u6bcf\u6b21 forward \u90fd\u662f\u751f\u6210\u4e00\u4e2a\u65b0\u7684 token \uff08\u65e0\u8bba\u662f prefill \u8fd8\u662f decoder \u9636\u6bb5\uff09\u3002prefill \u5b8c\u6210\u751f\u6210\u7b2c\u4e00\u4e2a new_token \u540e\u63a5\u7740\u81ea\u56de\u5f52\u3002
\u5982\u679c\u662f continous batch \u914d\u7f6e\u60c5\u51b5\u4e0b\uff0c\u5982\u679c\u6709 sample \u8f93\u51fa end_token \uff0c\u90a3 batch_size \u5c31\u5c11\u4e00\u4e2a\uff0c\u53ef\u80fd\u5c31\u53ef\u4ee5\u518d\u653e\u4e00\u4e2a sample \u8fdb\u6765\u3002\u8fd9\u6837\u5b50\u7684\u8bdd\u8fd9\u4e2a\u65b0 sample \u8981\u505a prefill \uff08\u76ee\u7684\u4e5f\u662f\u751f\u6210 new token \uff09\uff0c\u522b\u7684 sample \u662f\u7ee7\u7eed\u5728 decoder \u9636\u6bb5\u3002\u4e0d\u8fc7\u522b\u7684 sample \u90fd\u662f\u6709\u524d\u9762\u7684 kvcache \u7684\uff0c\u6240\u4ee5\u8fd9\u4e2a\u65f6\u5019\u8fd9\u4e2a\u65b0 sample \u7684 prefill \u5c31\u62d6\u7d2f\u4e86\u65e7\u7684 sample \u4e2d\u7684 decoder \u8fc7\u7a0b\uff0c\u6bd5\u7adf\u5927\u5bb6\u90fd\u8fd8\u662f\u5728\u4e00\u4e2a batch \u4e2d\uff0c\u5927\u5bb6\u6700\u7ec8\u8fd8\u662f\u4e00\u8d77\u4e00\u5c42 layer \u4e00\u5c42 layer \u8fc7\u3002
\u5982\u679c\u540c\u4e00\u4e2a batch \u90fd\u5728 decoder \u9636\u6bb5\uff0c\u56e0\u4e3a kvcache \u7684\u5b58\u5728\u5927\u5bb6\u90fd\u5f88\u5feb\u3002\u4f46\u662f\u53ea\u8981\u6709\u4e00\u4e2a sample \u8f93\u51fa\u4e86 end_token, \u5c31\u53ef\u80fd\u5141\u8bb8\u65b0\u7684 sample \u8fdb\u6765\uff0c\u8fd9\u4e2a\u65b0 sample \u7684 prefill \u8fc7\u7a0b\u56e0\u4e3a\u53ef\u80fd\u6ca1\u6709 kvcache \u4f1a\u62d6\u7d2f\u5176\u4ed6\u7684 sample \u3002\u6240\u4ee5\u6709\u4e00\u4e9b\u5de5\u4f5c\u63d0\u51fa\u8981\u5c06 prefill \u548c decoder \u5206\u79bb\u3002

\u4ee5\u4e0a\u662f\u6211\u7684\u7406\u89e3\uff0c\u4e0d\u77e5\u9053\u6709\u591a\u5c11\u9519\u8bef\u7684\u5185\u5bb9\uff0c\u5e0c\u671b\u6709\u5927\u4f6c\u6307\u6b63\u3002\u6211\u7684\u63cf\u8ff0\u53ef\u80fd\u6709\u70b9\u5570\u55e6\uff0c\u62b1\u6b49\u3002

\n", "date_published": "2025-01-11T11:50:44+00:00", "title": "LLM \u9759\u6001\u6279\u5904\u7406\u548c Continuous Batch \u76f8\u5173\u7591\u95ee\u7684\u6c42\u89e3", "id": "t/1104386" }, { "author": { "url": "member/YaD2x", "name": "YaD2x", "avatar": "https://cdn.v2ex.com/gravatar/32af3ffe83354a0b5845859c611e2278?s=73&d=retro" }, "url": "t/1103390", "date_modified": "2025-01-08T11:15:53+00:00", "content_html": "

\u6700\u597d\u662f\u57fa\u4e8e AI \u6a21\u578b\u7684\uff0c\u901a\u8fc7\u7b14\u8bb0\u672c\u524d\u7f6e\u6444\u50cf\u5934\uff0c\u6216\u5176\u5b83\u666e\u901a\u6444\u50cf\u5934\u68c0\u6d4b\u6d3b\u4f53\u957f\u65f6\u95f4\u5750\u5728\u7535\u8111\u524d\u7684\u3002\u5b9e\u73b0\u4e45\u5750 1h \u5f39\u51fa\u63d0\u793a\u7684\u3002

\n", "date_published": "2025-01-08T01:49:06+00:00", "title": "\u6709\u6ca1\u6709\u4e45\u5750\u63d0\u9192\u7c7b\u670d\u52a1\uff0c\u6700\u597d\u662f\u57fa\u4e8e\u5565\u5565\u5565\u6a21\u578b\u7684\u90a3\u79cd\u6d3b\u4f53\u68c0\u6d4b\u3002", "id": "t/1103390" }, { "author": { "url": "member/CNYoki", "name": "CNYoki", "avatar": "https://cdn.v2ex.com/gravatar/4ffdf90759f1220f203f79d71751a91e?s=73&d=retro" }, "url": "t/1102993", "date_modified": "2025-01-07T04:35:44+00:00", "content_html": "\u73b0\u5728\u5927\u6982\u6709\u4e09\u7ec4\u5171 100 \u591a\u4e2a\u6807\u7b7e\uff0c\u7528\u6765\u5bf9\u591a\u6bb5\u6587\u672c\u8fdb\u884c\u5206\u7c7b\uff0c\u662f\u5e94\u8be5\u8bbe\u7f6e\u4e09\u4e2a prompt \u5206\u4e09\u6b21\u8fdb\u884c\u751f\u6210\uff0c\u8fd8\u662f\u7528\u4e00\u4e2a\u957f prompt \uff1f

\u53e6\u5916\u73b0\u5728\u901a\u7528\u578b LLMs \u597d\u50cf\u5bf9\u6807\u7b7e\u5de5\u4f5c\u90fd\u4e0d\u592a\u5b8c\u7f8e\uff0c\u7ecf\u5e38\u51fa\u73b0\u6807\u51fa\u5176\u4ed6\u4e0d\u5728\u5b57\u5178\u7684\u6807\u7b7e\uff0c\u6216\u662f\u6807\u7b7e\u8fc7\u591a\u4e4b\u540e\u5206\u7c7b\u6548\u679c\u4e0d\u4f73\u3002

\u6709\u6ca1\u6709\u670b\u53cb\u5728\u505a\u76f8\u5173\u5de5\u4f5c\u6709\u7ecf\u9a8c\u7684\uff0c\u53ef\u4ee5\u5206\u4eab\u4e0b\u65b9\u6cd5\u561b", "date_published": "2025-01-06T10:58:59+00:00", "title": "LLMs \u6587\u672c\u6807\u7b7e\uff08\u5206\u7c7b\uff09\u4efb\u52a1\u600e\u4e48\u505a\u6bd4\u8f83\u597d\uff1f", "id": "t/1102993" }, { "author": { "url": "member/ttgo", "name": "ttgo", "avatar": "https://cdn.v2ex.com/avatar/54a9/40cc/276192_large.png?m=1739270718" }, "url": "t/1102896", "date_modified": "2025-01-07T04:35:42+00:00", "content_html": "\u65e9\u5e74\u95f4\u901a\u8fc7\u5434\u6069\u8fbe\u7684\u673a\u5668\u5b66\u4e60\u6559\u7a0b\u5165\u7684\u95e8\u3002\u6700\u8fd1\u51e0\u5e74 transformer/vit/llm \u4ec0\u4e48\u7684\u51fa\u6765\u4ee5\u540e\uff0c\u77e5\u8bc6\u4e0d\u591f\u7528\u4e86\uff0c\u611f\u89c9\u9700\u8981\u8ddf\u7740\u4f8b\u5b50\u4e0a\u624b\u8dd1\u4e00\u8dd1\u624d\u884c\u3002\u6709\u63a8\u8350\u7684\u8bfe\u7a0b\u4e48\uff1f\u504f\u6280\u672f\u3001\u504f\u5b9e\u6218\u7684\u3002\u4e0d\u662f\u5439\u725b\u903c\u3001\u5199\u63d0\u793a\u8bcd\u90a3\u79cd\u3002", "date_published": "2025-01-06T06:09:22+00:00", "title": "\u80fd\u63a8\u8350\u4e2a\u504f\u5b9e\u6218\u7684 tranformer/LLM \u7684\u8bfe\u7a0b\u4e48\uff1f", "id": "t/1102896" }, { "author": { "url": "member/anivie", "name": "anivie", "avatar": "https://cdn.v2ex.com/avatar/5aa1/52a3/628407_large.png?m=1773838354" }, "url": "t/1101819", "title": "\u6a21\u578b\u90e8\u7f72\u7684\u59ff\u52bf", "id": "t/1101819", "date_published": "2025-01-01T10:07:19+00:00", "content_html": "

\u5c0f\u767d\u60f3\u95ee\u95ee\u73b0\u5728\u7684\u6a21\u578b\u4e00\u822c\u90fd\u662f\u600e\u4e48\u90e8\u7f72\u7684\uff1f\u76f4\u63a5\u7528 python \u7684\u6846\u67b6\u90e8\u7f72\u4e0a\u53bb\u5417\uff1f\u8fd8\u662f\u5bfc\u51fa\u7ed9 c++\u90e8\u7f72\uff1f\u770b\u5230\u597d\u50cf onnx \u90e8\u7f72\u7684\u6587\u7ae0\u597d\u50cf\u5f88\u7a00\u5c11\uff0c\u5927\u4f19\u90fd\u662f\u76f4\u63a5\u7528 python \u90e8\u7f72\u7684\u5417\uff1f

\n" }, { "author": { "url": "member/jiashuaibei", "name": "jiashuaibei", "avatar": "https://cdn.v2ex.com/avatar/fc06/6100/624833_large.png?m=1766543554" }, "url": "t/1101050", "date_modified": "2024-12-30T00:10:50+00:00", "content_html": "\u6bd5\u4e1a\u8bbe\u8ba1\u7684\u8bfe\u9898\u8bad\u7ec3\u7684\u6a21\u578b\u6570\u636e\u91cf\u4e0d\u90a3\u4e48\u5927\uff0c\u60f3\u5728\u81ea\u5df1\u7535\u8111\u4e0a\u8bad\u7ec3\u3002
\u4e0d\u662f\u5f88\u559c\u6b22\u7269\u7406\u673a\u88c5 Ubuntu \uff0c\u6bd5\u7adf\u8fd8\u9700\u8981\u5199\u8bba\u6587\u4ee5\u53ca\u884c\u4e1a\u8f6f\u4ef6\u53ea\u652f\u6301 Windows \u3002
\u90a3\u4e48\u5728 WSL \u8dd1\u6df1\u5ea6\u5b66\u4e60\u663e\u5361\u76f4\u901a\u635f\u8017\u5927\u6982\u591a\u5c11\uff1f\u53ef\u884c\u5417\uff1f\u6709\u8fc7\u6765\u4eba\u5417\uff1f", "date_published": "2024-12-29T08:09:02+00:00", "title": "\u6df1\u5ea6\u5b66\u4e60\u5728\u81ea\u5df1\u7535\u8111\u4e0a\u8dd1\uff0c Linux \u73af\u5883\uff0c\u9009\u62e9 WSL \u8fd8\u662f\u7269\u7406\u673a Ubuntu\uff1f", "id": "t/1101050" }, { "author": { "url": "member/jiashuaibei", "name": "jiashuaibei", "avatar": "https://cdn.v2ex.com/avatar/fc06/6100/624833_large.png?m=1766543554" }, "url": "t/1100947", "date_modified": "2024-12-30T00:10:21+00:00", "content_html": "", "date_published": "2024-12-28T11:28:50+00:00", "title": "CUDA \u641e\u6df1\u5ea6\u5b66\u4e60\uff0c 1TB \u7684\u4e00\u5757\u786c\u76d8\uff0c C \u76d8\u7559\u591a\u5927\u5408\u9002\uff1f", "id": "t/1100947" }, { "author": { "url": "member/James369", "name": "James369", "avatar": "https://cdn.v2ex.com/gravatar/8b64ecd4b42da39f0025e467ce8497d6?s=73&d=retro" }, "url": "t/1100332", "date_modified": "2024-12-26T23:58:32+00:00", "content_html": "\u5bf9\u4e8e\u7b80\u5355\u7684\u7f51\u7edc\uff08\u7f51\u7edc\u7ed3\u6784\u662f\u56fa\u5b9a\u4e0d\u53d8\u7684\uff0c\u9759\u6001\u7684\uff09\uff0c\u6bd4\u5982\u5168\u8fde\u63a5\u7684\u6211\u53ef\u4ee5\u7406\u89e3\uff0c\u5c31\u662f\u8bad\u7ec3\u4e00\u4e2a\u51fd\u6570\uff1a
y = f(x, \u03b8)
\u8fd9\u91cc\u51fd\u6570 f \u7684\u5b9a\u4e49\u662f\u56fa\u5b9a\u7684\u3002\u90a3\u4e48\uff0c\u53ea\u8981\u8f93\u5165\u4e00\u7ec4 x \u548c y \uff0c\u5c31\u53ef\u4ee5\u8bad\u7ec3\u51fa\u03b8\u3002

\u4f46\u662f\u5bf9\u4e8e RNN \uff0c\u6211\u4e0d\u5927\u7406\u89e3\u7684\u662f\uff1a
1. \u5b83\u7684\u7f51\u7edc\u7ed3\u6784\u662f\u4e0d\u5b9a\u7684\uff0c\u591a\u91cd\u7684\u3002\u5047\u5982\u5faa\u73af\u4e86 n \u6b21\uff0c\u76f8\u5f53\u4e8e\u8981\u8bad\u7ec3 n \u4e2a\u51fd\u6570: f1(x,\u03b8), f2(x,\u03b8),,, fn(x,\u03b8).
2. \u5bf9\u4e8e\u8fd9\u6837\u591a\u91cd\u7684\u7f51\u7edc\u7ed3\u6784\uff0c\u5b83\u662f\u600e\u4e48\u8bad\u7ec3\u6536\u655b\u7684\uff1f
3. \u540e\u9762\u8f93\u5165\u8fdb\u6765\u8bad\u7ec3\u7684 x \uff0c\u662f\u5426\u4f1a\u5bf9\u524d\u9762\u5df2\u7ecf\u8bad\u7ec3\u597d\u7684\u03b8\u4ea7\u751f\u5f71\u54cd\uff08\u7c7b\u4f3c\u707e\u96be\u6027\u9057\u5fd8\uff09", "date_published": "2024-12-25T22:46:01+00:00", "title": "\u600e\u4e48\u7406\u89e3 RNN \u5faa\u73af\u795e\u7ecf\u7f51\u7edc\u5de5\u4f5c\u539f\u7406\uff1f", "id": "t/1100332" }, { "author": { "url": "member/lisisi", "name": "lisisi", "avatar": "https://cdn.v2ex.com/avatar/235b/4426/245623_large.png?m=1730104552" }, "url": "t/1099417", "date_modified": "2024-12-22T09:49:46+00:00", "content_html": "

\u795e\u7ecf\u7f51\u7edc\u6bcf\u4e00\u5c42\u8282\u70b9\u8f93\u51fa\u503c\uff0c\u90fd\u8981\u5957\u4e00\u5c42\u975e\u7ebf\u6027\u6fc0\u6d3b\u51fd\u6570\u3002\u73b0\u5728\u4f7f\u7528\u8f83\u591a\u7684\u6fc0\u6d3b\u51fd\u6570 ReLU \u6fc0\u6d3b\u51fd\u6570\u5728 x \uff1e 0 \u7684\u65f6\u5019\u4e5f\u662f\u7ebf\u6027\u7684\uff0c\u5e76\u4e14\u662f y=x \uff0c\u8fd9\u80fd\u8d77\u5230\u201c\u7ebf\u6027\u201d=>\u201c\u975e\u7ebf\u6027\u201d\u7684\u4f5c\u7528\u5417\uff1f\u4e0d\u592a\u7406\u89e3 ReLU \u662f\u600e\u4e48\u5b9e\u73b0\u6fc0\u6d3b\u4f5c\u7528\u7684

\n", "date_published": "2024-12-22T09:42:32+00:00", "title": "ReLU \u51fd\u6570\u5728 x\uff1e 0 \u7684\u65f6\u5019\u4e5f\u662f\u7ebf\u6027\u7684\uff0c\u5957 ReLU \u51fd\u6570\u662f\u600e\u4e48\u5b9e\u73b0\u6fc0\u6d3b\u4f5c\u7528\u7684\uff1f", "id": "t/1099417" }, { "author": { "url": "member/cowcomic", "name": "cowcomic", "avatar": "https://cdn.v2ex.com/avatar/4939/ffd8/73524_large.png?m=1727658639" }, "url": "t/1093265", "title": "\u963f\u91cc\u65b0\u5f00\u6e90\u7684 QwQ-32B-Preview \u6a21\u578b\u6709\u5565\u7279\u6b8a\u6027\uff1f", "id": "t/1093265", "date_published": "2024-11-28T01:13:56+00:00", "content_html": "

\u8bf4\u662f\u6570\u5b66\u548c\u7f16\u7a0b\u9886\u57df\u8868\u73b0\u51fa\u8272\uff0c\u4f46\u4ed6\u4e0d\u662f\u6709\u4e86 coder \u548c math \u4e24\u4e2a\u5355\u72ec\u7684\u6a21\u578b\u4e86\u4e48