{"id":647864,"date":"2025-10-10T15:53:49","date_gmt":"2025-10-10T14:53:49","guid":{"rendered":"https:\/\/blog.jetbrains.com\/?post_type=pycharm&#038;p=647864"},"modified":"2025-10-10T15:53:54","modified_gmt":"2025-10-10T14:53:54","slug":"fine-tuning-and-deploying-gpt-models-using-hugging-face-transformers","status":"publish","type":"pycharm","link":"https:\/\/blog.jetbrains.com\/zh-hans\/pycharm\/2025\/10\/fine-tuning-and-deploying-gpt-models-using-hugging-face-transformers\/","title":{"rendered":"\u4f7f\u7528 Hugging Face Transformers \u5fae\u8c03\u548c\u90e8\u7f72 GPT \u6a21\u578b"},"content":{"rendered":"<p>\u5bf9\u4e8e\u673a\u5668\u5b66\u4e60\u7814\u7a76\u5458\u548c\u7231\u597d\u8005\u800c\u8a00\uff0cHugging Face \u5982\u4eca\u5df2\u662f\u4e00\u4e2a\u5bb6\u55bb\u6237\u6653\u7684\u540d\u5b57\u3002 \u4ed6\u4eec\u6700\u663e\u8457\u7684\u6210\u679c\u4e4b\u4e00\u4fbf\u662f <a href=\"https:\/\/huggingface.co\/docs\/transformers\/en\/index\" target=\"_blank\" rel=\"noopener\">Transformers<\/a>\uff0c\u8fd9\u662f\u4e00\u4e2a\u9002\u7528\u4e8e\u6587\u672c\u3001\u8ba1\u7b97\u673a\u89c6\u89c9\u3001\u97f3\u9891\u548c\u89c6\u9891\u9886\u57df\u673a\u5668\u5b66\u4e60\u6a21\u578b\u7684\u6a21\u578b\u5b9a\u4e49\u6846\u67b6\u3002 \u7531\u4e8e <a href=\"https:\/\/huggingface.co\/models\" target=\"_blank\" rel=\"noopener\">Hugging Face Hub<\/a> \u4e0a\u62e5\u6709\u6d77\u91cf\u6700\u5148\u8fdb\u7684\u673a\u5668\u5b66\u4e60\u6a21\u578b\uff0c\u5e76\u4e14 Transformers \u80fd\u4e0e\u7edd\u5927\u591a\u6570\u8bad\u7ec3\u6846\u67b6\u517c\u5bb9\uff0c\u8be5\u6846\u67b6\u88ab\u5e7f\u6cdb\u5e94\u7528\u4e8e\u63a8\u7406\u548c\u6a21\u578b\u8bad\u7ec3\u3002<\/p>\n<h2 class=\"wp-block-heading\">\u4e3a\u4ec0\u4e48\u8981\u5bf9 AI \u6a21\u578b\u8fdb\u884c\u5fae\u8c03\uff1f<\/h2>\n<p>\u5bf9 AI \u6a21\u578b\u8fdb\u884c\u5fae\u8c03\u81f3\u5173\u91cd\u8981\uff0c\u5176\u6838\u5fc3\u4f5c\u7528\u662f\u8ba9\u6a21\u578b\u6027\u80fd\u9002\u914d\u7279\u5b9a\u4efb\u52a1\u4e0e\u6570\u636e\u96c6\uff0c\u76f8\u8f83\u4e8e\u4f7f\u7528\u901a\u7528\u6a21\u578b\uff0c\u7ecf\u5fae\u8c03\u7684\u6a21\u578b\u80fd\u5b9e\u73b0\u66f4\u9ad8\u7684\u51c6\u786e\u6027\u4e0e\u6548\u7387\u3002 \u901a\u8fc7\u5bf9\u9884\u8bad\u7ec3\u6a21\u578b\u8fdb\u884c\u9002\u914d\uff0c\u5fae\u8c03\u65e0\u9700\u4ece\u5934\u5f00\u59cb\u8bad\u7ec3\u6a21\u578b\uff0c\u5927\u5e45\u8282\u7701\u4e86\u65f6\u95f4\u4e0e\u8d44\u6e90\u3002 \u540c\u65f6\uff0c\u5b83\u8fd8\u80fd\u8ba9\u6a21\u578b\u66f4\u597d\u5730\u5904\u7406\u7279\u5b9a\u9886\u57df\u5185\u7684\u4e13\u5c5e\u683c\u5f0f\u3001\u7ec6\u5fae\u5dee\u5f02\u548c\u8fb9\u7f18\u7528\u4f8b\uff0c\u8fdb\u800c\u751f\u6210\u66f4\u53ef\u9760\u3001\u66f4\u8d34\u5408\u9700\u6c42\u7684\u8f93\u51fa\u3002<\/p>\n<p>\u5728\u8fd9\u7bc7\u535a\u6587\u4e2d\uff0c\u6211\u4eec\u5c06\u5bf9 GPT \u6a21\u578b\u7684\u6570\u5b66\u63a8\u7406\u80fd\u529b\u8fdb\u884c\u5fae\u8c03\uff0c\u4f7f\u5176\u80fd\u66f4\u51fa\u8272\u5730\u5904\u7406\u6570\u5b66\u95ee\u9898\u3002<\/p>\n<h2 class=\"wp-block-heading\">\u4f7f\u7528\u6765\u81ea Hugging Face \u7684\u6a21\u578b<\/h2>\n<p>\u4f7f\u7528 PyCharm \u65f6\uff0c\u6211\u4eec\u53ef\u4ee5\u8f7b\u677e\u6d4f\u89c8\u5e76\u6dfb\u52a0\u6765\u81ea Hugging Face \u7684\u4efb\u4f55\u6a21\u578b\u3002 \u5728\u65b0\u5efa\u7684 Python \u6587\u4ef6\u4e2d\uff0c\u4ece\u9876\u90e8\u7684 <em>Code<\/em>\uff08\u4ee3\u7801\uff09\u83dc\u5355\u4e2d\u9009\u62e9 <em>Insert HF Model<\/em>\uff08\u63d2\u5165 HF \u6a21\u578b\uff09\u3002<\/p>\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" loading=\"lazy\" class=\"wp-image-594074\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/08\/image-40.png\" alt=\"\u4f7f\u7528\u6765\u81ea Hugging Face \u7684\u6a21\u578b\" width=\"946\" height=\"1070\" \/><\/figure>\n<p>\u5728\u5f39\u51fa\u7684\u83dc\u5355\u4e2d\uff0c\u60a8\u65e2\u53ef\u4ee5\u6309\u7c7b\u522b\u6d4f\u89c8\u6a21\u578b\uff0c\u4e5f\u53ef\u4ee5\u5728\u9876\u90e8\u7684\u641c\u7d22\u680f\u4e2d\u8f93\u5165\u5185\u5bb9\u8fdb\u884c\u641c\u7d22\u3002 \u9009\u4e2d\u67d0\u4e2a\u6a21\u578b\u540e\uff0c\u60a8\u53ef\u4ee5\u5728\u53f3\u4fa7\u67e5\u770b\u6a21\u578b\u8bf4\u660e\u3002<\/p>\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" loading=\"lazy\" class=\"wp-image-594085\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/08\/image-41.png\" alt=\"\u63a2\u7d22 Hugging Face \u7684\u6a21\u578b\" width=\"1600\" height=\"923\" \/><\/figure>\n<p>\u70b9\u51fb <em>Use Model<\/em>\uff08\u4f7f\u7528\u6a21\u578b\uff09\u540e\uff0c\u60a8\u4f1a\u770b\u5230\u4e00\u6bb5\u4ee3\u7801\u5df2\u6dfb\u52a0\u5230\u6587\u4ef6\u4e2d\u3002 \u81f3\u6b64\u64cd\u4f5c\u5373\u5b8c\u6210\uff0c\u60a8\u5df2\u51c6\u5907\u597d\u5f00\u59cb\u4f7f\u7528 Hugging Face \u6a21\u578b\u3002<\/p>\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" loading=\"lazy\" class=\"wp-image-594096\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/08\/image-42.png\" alt=\"\u5728 PyCharm \u4e2d\u4f7f\u7528 Hugging Face \u6a21\u578b\" width=\"1600\" height=\"312\" \/><\/figure>\n<h2 class=\"wp-block-heading\">GPT\uff08\u751f\u6210\u5f0f\u9884\u8bad\u7ec3 Transformer\uff09\u6a21\u578b<\/h2>\n<p>\u5728 <a href=\"https:\/\/huggingface.co\/models\" target=\"_blank\" rel=\"noopener\">Hugging Face Hub<\/a> \u4e0a\uff0cGPT \u6a21\u578b\u975e\u5e38\u53d7\u6b22\u8fce\uff0c\u4f46 GPT \u6a21\u578b\u7a76\u7adf\u662f\u4ec0\u4e48\u5462\uff1f GPT \u662f\u4e00\u7c7b\u7ecf\u8fc7\u8bad\u7ec3\u7684\u6a21\u578b\uff0c\u80fd\u591f\u7406\u89e3\u81ea\u7136\u8bed\u8a00\u5e76\u751f\u6210\u9ad8\u8d28\u91cf\u6587\u672c\u3002 \u5b83\u4eec\u4e3b\u8981\u5e94\u7528\u4e8e\u6587\u672c\u8574\u542b\u3001\u95ee\u7b54\u3001\u8bed\u4e49\u76f8\u4f3c\u5ea6\u5224\u65ad\u53ca\u6587\u6863\u5206\u7c7b\u7b49\u76f8\u5173\u4efb\u52a1\u3002 \u5176\u4e2d\u6700\u77e5\u540d\u7684\u6a21\u578b\u4fbf\u662f <a href=\"https:\/\/openai.com\/index\/chatgpt\/\" target=\"_blank\" rel=\"noopener\">OpenAI \u5f00\u53d1\u7684 ChatGPT<\/a>\u3002<\/p>\n<p>\u76ee\u524d\uff0c<a href=\"https:\/\/huggingface.co\/models\" target=\"_blank\" rel=\"noopener\">Hugging Face Hub<\/a> \u4e0a\u63d0\u4f9b\u4e86\u591a\u6b3e OpenAI GPT \u6a21\u578b\uff0c\u5728\u540e\u7eed\u5185\u5bb9\u4e2d\uff0c\u6211\u4eec\u5c06\u5b66\u4e60\u5982\u4f55\u901a\u8fc7 Transformers \u4f7f\u7528\u8fd9\u4e9b\u6a21\u578b\u3001\u5982\u4f55\u5229\u7528\u81ea\u6709\u6570\u636e\u5bf9\u5176\u8fdb\u884c\u5fae\u8c03\uff0c\u4ee5\u53ca\u5982\u4f55\u5728\u5e94\u7528\u7a0b\u5e8f\u4e2d\u90e8\u7f72\u8fd9\u4e9b\u6a21\u578b\u3002<\/p>\n<h2 class=\"wp-block-heading\">\u4f7f\u7528 Transformers \u7684\u4f18\u52bf<\/h2>\n<p>Transformers \u4e0e Hugging Face \u63d0\u4f9b\u7684\u5176\u4ed6\u5de5\u5177\u76f8\u7ed3\u5408\uff0c\u4e3a\u5fae\u8c03\u5404\u7c7b\u590d\u6742\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u63d0\u4f9b\u4e86\u9ad8\u9636\u5de5\u5177\u3002 \u8fd9\u7c7b\u5de5\u5177\u65e0\u9700\u60a8\u5b8c\u5168\u7406\u89e3\u67d0\u4e00\u6a21\u578b\u7684\u67b6\u6784\u4e0e\u8bcd\u4f8b\u5316\u65b9\u6cd5\uff0c\u5c31\u80fd\u5e2e\u52a9\u8ba9\u6a21\u578b\u5b9e\u73b0\u201c\u5373\u63d2\u5373\u7528\u201d\uff0c\u9002\u914d\u4efb\u610f\u517c\u5bb9\u7684\u8bad\u7ec3\u6570\u636e\uff0c\u540c\u65f6\u5728\u8bcd\u4f8b\u5316\u4e0e\u8bad\u7ec3\u73af\u8282\u8fd8\u63d0\u4f9b\u4e86\u4e30\u5bcc\u7684\u81ea\u5b9a\u4e49\u7a7a\u95f4\u3002<\/p>\n<h2 class=\"wp-block-heading\">Transformers \u5b9e\u9645\u8fd0\u4f5c<\/h2>\n<p>\u4e3a\u4e86\u66f4\u76f4\u89c2\u5730\u4e86\u89e3 Transformers \u7684\u5b9e\u9645\u8fd0\u4f5c\u65b9\u5f0f\uff0c\u6211\u4eec\u6765\u770b\u770b\u5982\u4f55\u901a\u8fc7\u5b83\u4e0e GPT \u6a21\u578b\u8fdb\u884c\u4ea4\u4e92\u3002<\/p>\n<h3 class=\"wp-block-heading\">\u4f7f\u7528\u9884\u8bad\u7ec3\u6a21\u578b\u901a\u8fc7\u7ba1\u9053\u8fdb\u884c\u63a8\u7406<\/h3>\n<p>\u9009\u62e9 OpenAI GPT-2 \u6a21\u578b\u5e76\u5c06\u5176\u6dfb\u52a0\u5230\u4ee3\u7801\u4e2d\u540e\uff0c\u6211\u4eec\u4f1a\u5f97\u5230\u5982\u4e0b\u5185\u5bb9\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">from transformers import pipeline\n\n\npipe = pipeline(\"text-generation\", model=\"openai-community\/gpt2\")<\/pre>\n<p>\u4f7f\u7528\u524d\uff0c\u6211\u4eec\u9700\u8981\u505a\u4e00\u4e9b\u51c6\u5907\u5de5\u4f5c\u3002 \u9996\u5148\uff0c\u6211\u4eec\u9700\u8981\u5b89\u88c5\u4e00\u4e2a\u673a\u5668\u5b66\u4e60\u6846\u67b6\u3002 \u5728\u672c\u4f8b\u4e2d\uff0c\u6211\u4eec\u9009\u62e9\u4e86 <a href=\"https:\/\/pytorch.org\/get-started\/locally\/\" target=\"_blank\" rel=\"noopener\">PyTorch<\/a>\u3002 \u60a8\u53ef\u4ee5\u901a\u8fc7 PyCharm \u4e2d\u7684 <em>Python Packages<\/em>\uff08Python \u8f6f\u4ef6\u5305\uff09\u7a97\u53e3\u8f7b\u677e\u5b8c\u6210\u5b89\u88c5\u3002<\/p>\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" loading=\"lazy\" class=\"wp-image-594107\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/08\/image-43.png\" alt=\"\u5728 PyCharm \u4e2d\u5b89\u88c5 PyTorch\" width=\"920\" height=\"654\" \/><\/figure>\n<p>\u63a5\u4e0b\u6765\uff0c\u6211\u4eec\u9700\u8981\u4f7f\u7528 `torch` \u9009\u9879\u5b89\u88c5 Transformers\u3002 \u60a8\u53ef\u4ee5\u901a\u8fc7\u7ec8\u7aef\u8fdb\u884c\u6b64\u64cd\u4f5c \u2013 \u70b9\u51fb\u5de6\u4fa7\u7684\u6309\u94ae\u6253\u5f00\u7ec8\u7aef\uff0c\u6216\u8005\u4f7f\u7528 <em>\u2325 F12 <\/em>(macOS) \u6216 <em>Alt + F12<\/em> (Windows) \u70ed\u952e\u3002<\/p>\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" loading=\"lazy\" class=\"wp-image-594118\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/08\/image-44.png\" alt=\"\u5728 PyCharm \u7684\u7ec8\u7aef\u4e2d\u5b89\u88c5 Transformers\" width=\"838\" height=\"502\" \/><\/figure>\n<p>\u5728\u7ec8\u7aef\u4e2d\uff0c\u7531\u4e8e\u6211\u4eec\u6b63\u5728\u4f7f\u7528 uv\uff0c\u9700\u8981\u4f7f\u7528\u4ee5\u4e0b\u547d\u4ee4\u5c06\u5176\u6dfb\u52a0\u4e3a\u4f9d\u8d56\u9879\u5e76\u5b8c\u6210\u5b89\u88c5\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"bash\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">uv add \u201ctransformers[torch]\u201d\nuv sync<\/pre>\n<p>\u5982\u679c\u60a8\u4f7f\u7528 pip\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"bash\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">pip install \u201ctransformers[torch]\u201d<\/pre>\n<p>\u6211\u4eec\u8fd8\u5c06\u5b89\u88c5\u540e\u7eed\u4f1a\u7528\u5230\u7684\u53e6\u5916\u51e0\u4e2a\u5e93\uff0c\u5305\u62ec python-dotenv\u3001datasets\u3001notebook \u4ee5\u53ca ipywidgets\u3002 \u60a8\u53ef\u4ee5\u901a\u8fc7\u4e0a\u8ff0\u4efb\u610f\u4e00\u79cd\u65b9\u6cd5\u5b89\u88c5\u8fd9\u4e9b\u5e93\u3002<br \/>\u4e4b\u540e\uff0c\u6700\u597d\u6dfb\u52a0\u4e00\u4e2a GPU \u8bbe\u5907\u6765\u52a0\u901f\u6a21\u578b\u3002 \u60a8\u53ef\u4ee5\u901a\u8fc7\u5728\u7ba1\u9053\u4e2d\u8bbe\u7f6e device \u53c2\u6570\u6765\u6dfb\u52a0\uff0c\u5177\u4f53\u53d6\u51b3\u4e8e\u8bbe\u5907\u914d\u7f6e\u3002 \u7531\u4e8e\u6211\u4f7f\u7528\u7684\u662f Mac M2 \u8bbe\u5907\uff0c\u6211\u53ef\u4ee5\u50cf\u8fd9\u6837\u8bbe\u7f6e<code class=\"EnlighterJSRAW\" data-enlighter-language=\"python\"> device=\"mps\"<\/code>\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">pipe = pipeline(\"text-generation\", model=\"openai-community\/gpt2\", device=\"mps\")<\/pre>\n<p>\u5982\u679c\u60a8\u6709 CUDA GPU\uff0c\u4e5f\u53ef\u4ee5\u8bbe\u7f6e <code class=\"EnlighterJSRAW\" data-enlighter-language=\"python\">device=\"cuda\"<\/code>\u3002<\/p>\n<p>\u73b0\u5728\uff0c\u6211\u4eec\u5df2\u7ecf\u8bbe\u7f6e\u597d\u4e86\u7ba1\u9053\uff0c\u4e0b\u9762\u7528\u4e00\u6761\u7b80\u5355\u7684\u63d0\u793a\u6765\u8bd5\u4e00\u4e0b\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">from transformers import pipeline\n\n\npipe = pipeline(\"text-generation\", model=\"openai-community\/gpt2\", device=\"mps\")\n\n\nprint(pipe(\"A rectangle has a perimeter of 20 cm. If the length is 6 cm, what is the width?\", max_new_tokens=200))<\/pre>\n<p>\u70b9\u51fb\u9876\u90e8\u7684 <em>Run<\/em>\uff08\u8fd0\u884c\uff09\u6309\u94ae (<img decoding=\"async\" loading=\"lazy\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/09\/AD_4nXf6ZDm7vSGyFlO0DzXegK6WP9JxsStUiJA-bkRZ0mwPsUsmn8M70emV5Sr8f17-fEK6z9V1EQKWEm3RPHdT8n8uqG18faVmQn5y09psVInQLU0CZQKXAEg2q7m7AOsh4hPU7G8gcQ.png\" width=\"30\" height=\"23\" \/>)\uff1a<\/p>\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" loading=\"lazy\" class=\"wp-image-594129\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/08\/image-45.png\" alt=\"\u5728 PyCharm \u4e2d\u8fd0\u884c\u811a\u672c\" width=\"1050\" height=\"282\" \/><\/figure>\n<p>\u8fd0\u884c\u7ed3\u679c\u5927\u81f4\u5982\u4e0b\u6240\u793a\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">[{'generated_text': 'A rectangle has a perimeter of 20 cm. If the length is 6 cm, what is the width?nnA rectangle has a perimeter of 20 cm. If the length is 6 cm, what is the width? A rectangle has a perimeter of 20 cm. If the width is 6 cm, what is the width? A rectangle has a perimeter of 20 cm. If the width is 6 cm, what is the width? A rectangle has a perimeter of 20 cm. If the width is 6 cm, what is the width?nnA rectangle has a perimeter of 20 cm. If the width is 6 cm, what is the width? A rectangle has a perimeter of 20 cm. If the width is 6 cm, what is the width? A rectangle has a perimeter of 20 cm. If the width is 6 cm, what is the width? A rectangle has a perimeter of 20 cm. If the width is 6 cm, what is the width?nnA rectangle has a perimeter of 20 cm. If the width is 6 cm, what is the width? A rectangle has a perimeter'}]<\/pre>\n<p>\u8fd9\u6bb5\u8f93\u51fa\u5b8c\u5168\u6ca1\u6709\u4ec0\u4e48\u903b\u8f91\u53ef\u8a00\uff0c\u53ea\u662f\u4e00\u5806\u65e0\u610f\u4e49\u7684\u5185\u5bb9\u3002\u00a0<\/p>\n<p>\u60a8\u53ef\u80fd\u8fd8\u4f1a\u770b\u5230\u5982\u4e0b\u8b66\u544a\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"bash\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">Setting `pad_token_id` to `eos_token_id`:50256 for open-end generation.<\/pre>\n<p>\u8fd9\u662f\u9ed8\u8ba4\u8bbe\u7f6e\u3002\u60a8\u4e5f\u53ef\u4ee5\u6309\u4ee5\u4e0b\u65b9\u5f0f\u624b\u52a8\u6dfb\u52a0\uff0c\u4ee5\u4fbf\u6d88\u9664\u6b64\u8b66\u544a\uff0c\u4f46\u5728\u5f53\u524d\u9636\u6bb5\uff0c\u6211\u4eec\u65e0\u9700\u8fc7\u5206\u5173\u6ce8\u6b64\u95ee\u9898\u3002<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">print(pipe(\"A rectangle has a perimeter of 20 cm. If the length is 6 cm, what is the width?\", max_new_tokens=200, pad_token_id=pipe.tokenizer.eos_token_id))<\/pre>\n<p>\u6211\u4eec\u73b0\u5728\u5df2\u7ecf\u4e86\u89e3\u4e86 GPT-2 \u9ed8\u8ba4\u72b6\u6001\u4e0b\u7684\u8868\u73b0\uff0c\u63a5\u4e0b\u6765\u770b\u770b\u80fd\u5426\u901a\u8fc7\u4e00\u4e9b\u5fae\u8c03\uff0c\u8ba9\u5b83\u5728\u6570\u5b66\u63a8\u7406\u65b9\u9762\u7684\u80fd\u529b\u6709\u6240\u63d0\u5347\u3002<\/p>\n<h3 class=\"wp-block-heading\">\u4ece Hugging Face Hub \u52a0\u8f7d\u5e76\u51c6\u5907\u6570\u636e\u96c6<\/h3>\n<p>\u5728\u5904\u7406 GPT \u6a21\u578b\u4e4b\u524d\uff0c\u6211\u4eec\u9996\u5148\u9700\u8981\u8bad\u7ec3\u6570\u636e\u3002 \u4e0b\u9762\u6765\u770b\u770b\u5982\u4f55\u4ece Hugging Face Hub \u83b7\u53d6\u6570\u636e\u96c6\u3002<\/p>\n<p>\u5982\u679c\u5c1a\u672a\u6ce8\u518c\uff0c\u8bf7\u5148\u6ce8\u518c\u4e00\u4e2a Hugging Face \u5e10\u6237\u5e76<a href=\"https:\/\/huggingface.co\/docs\/hub\/security-tokens#user-access-tokens\" target=\"_blank\" rel=\"noopener\">\u521b\u5efa\u8bbf\u95ee\u4ee4\u724c<\/a>\u3002 \u76ee\u524d\u6211\u4eec\u53ea\u9700\u8981\u4e00\u4e2a `read` \u4ee4\u724c\u3002 \u5c06\u4ee4\u724c\u5b58\u50a8\u5728 `.env` \u6587\u4ef6\u4e2d\uff0c\u5982\u4e0b\u6240\u793a\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"bash\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">HF_TOKEN=your-hugging-face-access-token<\/pre>\n<p>\u6211\u4eec\u5c06\u4f7f\u7528\u8fd9\u4e2a<a href=\"https:\/\/huggingface.co\/datasets\/Cheukting\/math-meta-reasoning-cleaned\" target=\"_blank\" rel=\"noopener\">\u6570\u5b66\u63a8\u7406\u6570\u636e\u96c6<\/a>\uff0c\u5176\u4e2d\u5305\u542b\u63cf\u8ff0\u4e00\u4e9b\u6570\u5b66\u63a8\u7406\u7684\u6587\u672c\u3002 \u6211\u4eec\u4f1a\u7528\u8fd9\u4e2a\u6570\u636e\u96c6\u5bf9 GPT \u6a21\u578b\u8fdb\u884c\u5fae\u8c03\uff0c\u4f7f\u5176\u80fd\u66f4\u6709\u6548\u5730\u89e3\u51b3\u6570\u5b66\u96be\u9898\u3002<\/p>\n<p>\u6211\u4eec\u6765\u65b0\u5efa\u4e00\u4e2a Jupyter Notebook \u7528\u4e8e\u8fdb\u884c\u5fae\u8c03\uff0c\u56e0\u4e3a\u5b83\u80fd\u8ba9\u6211\u4eec\u9010\u4e2a\u8fd0\u884c\u4e0d\u540c\u7684\u4ee3\u7801\u6bb5\u5e76\u76d1\u63a7\u8fdb\u5ea6\u3002<\/p>\n<p>\u5728\u7b2c\u4e00\u4e2a\u5355\u5143\u4e2d\uff0c\u6211\u4eec\u4f7f\u7528\u4ee5\u4e0b\u811a\u672c\u4ece Hugging Face Hub \u52a0\u8f7d\u6570\u636e\u96c6\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">from datasets import load_dataset\nfrom dotenv import load_dotenv\nimport os\n\n\nload_dotenv()\ndataset = load_dataset(\"Cheukting\/math-meta-reasoning-cleaned\", token=os.getenv(\"HF_TOKEN\"))\ndataset<\/pre>\n<p>\u8fd0\u884c\u6b64\u5355\u5143\uff08\u6240\u9700\u65f6\u95f4\u53ef\u80fd\u8f83\u957f\uff0c\u5177\u4f53\u53d6\u51b3\u4e8e\u7f51\u7edc\u901f\u5ea6\uff09\uff0c\u5b83\u4f1a\u4e0b\u8f7d\u6570\u636e\u96c6\u3002 \u5b8c\u6210\u540e\uff0c\u6211\u4eec\u53ef\u4ee5\u67e5\u770b\u7ed3\u679c\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">DatasetDict({\n    train: Dataset({\n        features: ['id', 'text', 'token_count'],\n        num_rows: 987485\n    })\n})\n<\/pre>\n<p>\u5982\u679c\u60a8\u597d\u5947\u5e76\u60f3\u5feb\u901f\u67e5\u770b\u4e00\u4e0b\u6570\u636e\uff0c\u53ef\u4ee5\u5728 PyCharm \u4e2d\u64cd\u4f5c\u3002 \u70b9\u51fb\u53f3\u4fa7\u7684\u6309\u94ae\uff0c\u6253\u5f00 <em>Jupyter Variables<\/em>\uff08Jupyter \u53d8\u91cf\uff09\u7a97\u53e3\uff1a<\/p>\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" loading=\"lazy\" class=\"wp-image-594140\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/08\/image-46.png\" alt=\"\u5728 PyCharm \u4e2d\u6253\u5f00 Jupyter Variables\uff08Jupyter \u53d8\u91cf\uff09\" width=\"1052\" height=\"740\" \/><\/figure>\n<p>\u5c55\u5f00 <em>dataset<\/em>\uff0c\u60a8\u4f1a\u5728 <em>dataset[\u2018train\u2019]<\/em> \u65c1\u8fb9\u770b\u5230 <em>View as DataFrame<\/em>\uff08\u4f5c\u4e3a DataFrame \u67e5\u770b\uff09\u9009\u9879\u3002<\/p>\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" loading=\"lazy\" class=\"wp-image-594152\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/08\/image-47.png\" alt=\"PyCharm \u4e2d\u7684 Jupyter Variables\uff08Jupyter \u53d8\u91cf\uff09\" width=\"980\" height=\"882\" \/><\/figure>\n<p>\u70b9\u51fb\u8be5\u9009\u9879\uff0c\u5373\u53ef\u5728 <em>Data View<\/em>\uff08\u6570\u636e\u89c6\u56fe\uff09\u5de5\u5177\u7a97\u53e3\u4e2d\u67e5\u770b\u6570\u636e\uff1a<\/p>\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" loading=\"lazy\" class=\"wp-image-594163\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/08\/image-48.png\" alt=\"PyCharm \u4e2d\u7684 Data View\uff08\u6570\u636e\u89c6\u56fe\uff09\u5de5\u5177\" width=\"980\" height=\"1102\" \/><\/figure>\n<p>\u63a5\u4e0b\u6765\uff0c\u6211\u4eec\u5c06\u5bf9\u6570\u636e\u96c6\u4e2d\u7684\u6587\u672c\u8fdb\u884c\u8bcd\u4f8b\u5316\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">from transformers import GPT2Tokenizer\n\n\ntokenizer = GPT2Tokenizer.from_pretrained(\"openai-community\/gpt2\")\ntokenizer.pad_token = tokenizer.eos_token\n\n\ndef tokenize_function(examples):\n   return tokenizer(examples['text'], truncation=True, padding='max_length', max_length=512)\n\n\ntokenized_datasets = dataset.map(tokenize_function, batched=True)<\/pre>\n<p>\u8fd9\u91cc\u6211\u4eec\u5c06\u4f7f\u7528 GPT-2 \u5206\u8bcd\u5668\uff0c\u5e76\u5c06 <code>pad_token<\/code> \u8bbe\u7f6e\u4e3a <code>eos_token<\/code>\uff08\u5373\u8868\u793a\u884c\u5c3e\u7684 token\uff09\u3002 \u4e4b\u540e\uff0c\u6211\u4eec\u4f1a\u901a\u8fc7\u4e00\u4e2a\u51fd\u6570\u5bf9\u6587\u672c\u8fdb\u884c\u8bcd\u4f8b\u5316\u3002 \u9996\u6b21\u8fd0\u884c\u65f6\u53ef\u80fd\u9700\u8981\u4e00\u6bb5\u65f6\u95f4\uff0c\u4f46\u4e4b\u540e\u4f1a\u8fdb\u884c\u7f13\u5b58\uff0c\u5982\u679c\u9700\u8981\u518d\u6b21\u8fd0\u884c\u8be5\u5355\u5143\uff0c\u901f\u5ea6\u4f1a\u52a0\u5feb\u3002<\/p>\n<p>\u8be5\u6570\u636e\u96c6\u5305\u542b\u8fd1 100 \u4e07\u884c\u8bad\u7ec3\u6570\u636e\u3002 \u5982\u679c\u60a8\u7684\u7b97\u529b\u8db3\u4ee5\u5904\u7406\u6240\u6709\u6570\u636e\uff0c\u53ef\u4ee5\u5168\u90e8\u90fd\u7528\u3002 \u4e0d\u8fc7\uff0c\u5728\u672c\u6f14\u793a\u4e2d\uff0c\u6211\u4eec\u662f\u5728\u672c\u5730\u7b14\u8bb0\u672c\u7535\u8111\u4e0a\u8fdb\u884c\u8bad\u7ec3\uff0c\u56e0\u6b64\u6700\u597d\u53ea\u4f7f\u7528\u4e00\u5c0f\u90e8\u5206\u6570\u636e\uff01<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">tokenized_datasets_split = tokenized_datasets[\"train\"].shard(num_shards=100, index=0).train_test_split(test_size=0.2, shuffle=True)\ntokenized_datasets_split<\/pre>\n<p>\u8fd9\u91cc\u6211\u53ea\u9009\u53d6 1% \u7684\u6570\u636e\uff0c\u7136\u540e\u6267\u884c <code>train_test_split<\/code> \u6765\u5c06\u6570\u636e\u96c6\u5206\u4e3a\u4e24\u90e8\u5206\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">DatasetDict({\n    train: Dataset({\n        features: ['id', 'text', 'token_count', 'input_ids', 'attention_mask'],\n        num_rows: 7900\n    })\n    test: Dataset({\n        features: ['id', 'text', 'token_count', 'input_ids', 'attention_mask'],\n        num_rows: 1975\n    })\n})\n<\/pre>\n<p>\u73b0\u5728\uff0c\u6211\u4eec\u5df2\u51c6\u5907\u597d\u5bf9 GPT-2 \u6a21\u578b\u8fdb\u884c\u5fae\u8c03\u3002<\/p>\n<h3 class=\"wp-block-heading\">\u5fae\u8c03 GPT \u6a21\u578b<\/h3>\n<p>\u5728\u63a5\u4e0b\u6765\u7684\u7a7a\u767d\u5355\u5143\u4e2d\uff0c\u6211\u4eec\u5c06\u8bbe\u7f6e\u8bad\u7ec3\u5b9e\u53c2\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">from transformers import TrainingArguments\ntraining_args = TrainingArguments(\n   output_dir='.\/results',\n   num_train_epochs=5,\n   per_device_train_batch_size=8,\n   per_device_eval_batch_size=8,\n   warmup_steps=100,\n   weight_decay=0.01,\n   save_steps = 500,\n   logging_steps=100,\n   dataloader_pin_memory=False\n)<\/pre>\n<p>\u8fd9\u4e9b\u5927\u591a\u662f\u5fae\u8c03\u6a21\u578b\u7684\u6807\u51c6\u914d\u7f6e\u3002 \u4e0d\u8fc7\uff0c\u6839\u636e\u60a8\u7684\u8ba1\u7b97\u673a\u914d\u7f6e\uff0c\u60a8\u53ef\u80fd\u9700\u8981\u8c03\u6574\u4ee5\u4e0b\u51e0\u9879\uff1a<\/p>\n<ul>\n<li>\u6279\u6b21\u5927\u5c0f \u2013 \u627e\u5230\u6700\u4f18\u7684\u6279\u5904\u7406\u5927\u5c0f\u5f88\u91cd\u8981\uff0c\u56e0\u4e3a\u6279\u6b21\u5927\u5c0f\u8d8a\u5927\uff0c\u8bad\u7ec3\u901f\u5ea6\u8d8a\u5feb\u3002 \u4e0d\u8fc7\uff0c\u53d7\u9650\u4e8e CPU \u6216 GPU \u7684\u5185\u5b58\u5bb9\u91cf\uff0c\u4f1a\u5b58\u5728\u4e00\u4e2a\u4e0a\u9650\u9608\u503c\u3002<\/li>\n<li>\u5468\u671f \u2013 \u5468\u671f\u8d8a\u591a\uff0c\u8bad\u7ec3\u8017\u65f6\u8d8a\u957f\u3002 \u60a8\u53ef\u4ee5\u6839\u636e\u9700\u6c42\u786e\u5b9a\u5468\u671f\u6570\u91cf\u3002<\/li>\n<li>\u4fdd\u5b58\u6b65\u6570 \u2013 \u8be5\u53c2\u6570\u51b3\u5b9a\u68c0\u67e5\u70b9\u4fdd\u5b58\u5230\u78c1\u76d8\u7684\u9891\u7387\u3002 \u5982\u679c\u8bad\u7ec3\u901f\u5ea6\u8f83\u6162\uff0c\u4e14\u5b58\u5728\u610f\u5916\u4e2d\u65ad\u7684\u53ef\u80fd\uff0c\u5219\u5efa\u8bae\u60a8\u589e\u52a0\u4fdd\u5b58\u9891\u6b21\uff08\u5c06\u6b64\u503c\u8bbe\u5c0f\u4e00\u4e9b\uff09\u3002<\/li>\n<\/ul>\n<p>\u00a0\u914d\u7f6e\u597d\u8fd9\u4e9b\u8bbe\u7f6e\u540e\uff0c\u6211\u4eec\u5c06\u5728\u4e0b\u4e00\u4e2a\u5355\u5143\u4e2d\u6574\u5408\u8bad\u7ec3\u7a0b\u5e8f\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">from transformers import Trainer, DataCollatorForLanguageModeling\n\n\nmodel = GPT2LMHeadModel.from_pretrained(\"openai-community\/gpt2\")\ndata_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False)\n\n\ntrainer = Trainer(\n   model=model,\n   args=training_args,\n   train_dataset=tokenized_datasets_split['train'],\n   eval_dataset=tokenized_datasets_split['test'],\n   data_collator=data_collator,\n)\n\n\ntrainer.train(resume_from_checkpoint=False)<\/pre>\n<p>\u6211\u4eec\u8bbe\u7f6e `resume_from_checkpoint=False`\uff0c\u4f46\u5982\u679c\u8bad\u7ec3\u4e2d\u65ad\uff0c\u60a8\u53ef\u4ee5\u5c06\u5176\u8bbe\u4e3a `True`\uff0c\u4ee5\u4fbf\u4ece\u4e0a\u6b21\u7684\u68c0\u67e5\u70b9\u7ee7\u7eed\u8bad\u7ec3\u3002<\/p>\n<p>\u8bad\u7ec3\u5b8c\u6210\u540e\uff0c\u6211\u4eec\u5c06\u8bc4\u4f30\u5e76\u4fdd\u5b58\u6a21\u578b\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">trainer.evaluate(tokenized_datasets_split['test'])\ntrainer.save_model(\".\/trained_model\")<\/pre>\n<p>\u73b0\u5728\uff0c\u6211\u4eec\u4fbf\u53ef\u4ee5\u5728\u7ba1\u9053\u4e2d\u4f7f\u7528\u8fd9\u4e2a\u8bad\u7ec3\u597d\u7684\u6a21\u578b\u4e86\u3002 \u8ba9\u6211\u4eec\u5207\u6362\u56de `model.py`\uff0c\u6b64\u524d\u6211\u4eec\u5728\u8be5\u6587\u4ef6\u4e2d\u4f7f\u7528\u4e86\u5305\u542b\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u7ba1\u9053\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">from transformers import pipeline\n\n\npipe = pipeline(\"text-generation\", model=\"openai-community\/gpt2\", device=\"mps\")\n\n\nprint(pipe(\"A rectangle has a perimeter of 20 cm. If the length is 6 cm, what is the width?\", max_new_tokens=200, pad_token_id=pipe.tokenizer.eos_token_id))<\/pre>\n<p>\u73b0\u5728\uff0c\u6211\u4eec\u5c06 `model=\u201dopenai-community\/gpt2\u2033` \u6539\u6210 `model=\u201d.\/trained_model\u201d`\uff0c\u770b\u770b\u4f1a\u5f97\u5230\u4ec0\u4e48\u7ed3\u679c\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">[{'generated_text': \"A rectangle has a perimeter of 20 cm. If the length is 6 cm, what is the width?nAlright, let me try to solve this problem as a student, and I'll let my thinking naturally fall into the common pitfall as described.nn---nn**Step 1: Attempting the Problem (falling into the pitfall)**nnWe have a rectangle with perimeter 20 cm. The length is 6 cm. We want the width.nnFirst, I need to find the area under the rectangle.nnLet\u2019s set ( A = 20 - 12 ), where ( A ) is the perimeter.nn**Area under a rectangle:**  n[nA = (20-12)^2 + ((-12)^2)^2 = 20^2 + 12^2 = 24n]nnSo, ( 24 = (20-12)^2 = 27 ).nnNow, I\u2019ll just divide both sides by 6 to find the area under the rectangle.n\"}]<\/pre>\n<p>\u9057\u61be\u7684\u662f\uff0c\u5b83\u4ecd\u7136\u65e0\u6cd5\u89e3\u51b3\u8fd9\u4e2a\u96be\u9898\u3002 \u4e0d\u8fc7\uff0c\u5b83\u786e\u5b9e\u751f\u6210\u4e86\u4e00\u4e9b\u4e4b\u524d\u6ca1\u7528\u8fc7\u7684\u6570\u5b66\u516c\u5f0f\u548c\u63a8\u7406\u3002 \u5982\u679c\u613f\u610f\uff0c\u60a8\u53ef\u4ee5\u5c1d\u8bd5\u7528\u6211\u4eec\u672a\u4f7f\u7528\u7684\u6570\u636e\u5bf9\u6a21\u578b\u8fdb\u884c\u66f4\u591a\u5fae\u8c03\u3002<\/p>\n<p>\u5728\u4e0b\u4e00\u8282\u4e2d\uff0c\u6211\u4eec\u5c06\u4e86\u89e3\u5982\u4f55\u7ed3\u5408 Hugging Face \u63d0\u4f9b\u7684\u5de5\u5177\u548c FastAPI\uff0c\u5c06\u5fae\u8c03\u540e\u7684\u6a21\u578b\u90e8\u7f72\u5230 API \u7aef\u70b9\u3002<\/p>\n<h2 class=\"wp-block-heading\">\u90e8\u7f72\u5fae\u8c03\u540e\u7684\u6a21\u578b<\/h2>\n<p>\u5728\u670d\u52a1\u5668\u540e\u7aef\u90e8\u7f72\u6a21\u578b\u7684\u6700\u7b80\u5355\u65b9\u5f0f\u662f\u4f7f\u7528 FastAPI\u3002 \u6211\u4e4b\u524d\u5199\u8fc7\u4e00\u7bc7\u5173\u4e8e\u4f7f\u7528 FastAPI \u90e8\u7f72\u673a\u5668\u5b66\u4e60\u6a21\u578b\u7684<a href=\"https:\/\/blog.jetbrains.com\/pycharm\/2024\/09\/how-to-use-fastapi-for-machine-learning\/\">\u535a\u6587<\/a>\u3002 \u6211\u4eec\u5728\u8fd9\u91cc\u4e0d\u4f1a\u6df1\u5165\u5230\u540c\u6837\u7684\u7ec6\u8282\u7a0b\u5ea6\uff0c\u4f46\u4f1a\u4ecb\u7ecd\u5982\u4f55\u90e8\u7f72\u5fae\u8c03\u540e\u7684\u6a21\u578b\u3002<\/p>\n<p>\u6211\u4eec\u501f\u52a9 <a href=\"https:\/\/www.jetbrains.com.cn\/junie\/\" target=\"_blank\" rel=\"noopener\">Junie<\/a> \u521b\u5efa\u4e86\u4e00\u4e9b\u811a\u672c\uff0c\u5927\u5bb6\u53ef\u4ee5\u5728<a href=\"https:\/\/github.com\/Cheukting\/fine-tune-gpt2\/tree\/main\/app\" target=\"_blank\" rel=\"noopener\">\u8fd9\u91cc<\/a>\u770b\u5230\u3002 \u6211\u4eec\u53ef\u4ee5\u5229\u7528\u8fd9\u4e9b\u811a\u672c\u90e8\u7f72\u5e26\u6709 FastAPI \u7aef\u70b9\u7684\u670d\u52a1\u5668\u540e\u7aef\u3002\u00a0<\/p>\n<p>\u6211\u4eec\u9700\u8981\u6dfb\u52a0\u4e00\u4e9b\u65b0\u7684\u4f9d\u8d56\u9879\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"bash\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">uv add fastapi pydantic uvicorn\nuv sync<\/pre>\n<p>\u6211\u4eec\u4ee5 `main.py` \u4e3a\u4f8b\uff0c\u6765\u770b\u4e00\u4e0b\u811a\u672c\u4e2d\u4e00\u4e9b\u503c\u5f97\u5173\u6ce8\u7684\u5730\u65b9\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\"># Initialize FastAPI app\napp = FastAPI(\n   title=\"Text Generation API\",\n   description=\"API for generating text using a fine-tuned model\",\n   version=\"1.0.0\"\n)\n\n\n# Initialize the model pipeline\ntry:\n   pipe = pipeline(\"text-generation\", model=\"..\/trained_model\", device=\"mps\")\nexcept Exception as e:\n   # Fallback to CPU if MPS is not available\n   try:\n       pipe = pipeline(\"text-generation\", model=\"..\/trained_model\", device=\"cpu\")\n   except Exception as e:\n       print(f\"Error loading model: {e}\")\n       pipe = None<\/pre>\n<p>\u521d\u59cb\u5316\u5e94\u7528\u540e\uff0c\u811a\u672c\u4f1a\u5c1d\u8bd5\u5c06\u6a21\u578b\u52a0\u8f7d\u5230\u7ba1\u9053\u4e2d\u3002 \u5982\u679c\u6ca1\u6709 Metal GPU \u53ef\u7528\uff0c\u5c31\u4f1a\u56de\u9000\u5230\u4f7f\u7528 CPU\u3002 \u5982\u679c\u60a8\u4f7f\u7528\u7684\u662f CUDA GPU \u800c\u975e Metal GPU\uff0c\u53ef\u4ee5\u5c06 `mps` \u6539\u4e3a `cuda`\u3002<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\"># Request model\nclass TextGenerationRequest(BaseModel):\n   prompt: str\n   max_new_tokens: int = 200\n  \n# Response model\nclass TextGenerationResponse(BaseModel):\n   generated_text: str<\/pre>\n<p>\u521b\u5efa\u4e86\u4e24\u4e2a\u65b0\u7c7b\uff0c\u5747\u7ee7\u627f\u81ea Pydantic \u7684 `BaseModel`\u3002<\/p>\n<p>\u6211\u4eec\u8fd8\u53ef\u4ee5\u901a\u8fc7 <em>Endpoints <\/em>\uff08\u7aef\u70b9\uff09\u5de5\u5177\u7a97\u53e3\u68c0\u67e5\u6240\u6709\u7aef\u70b9\u3002 \u70b9\u51fb\u7b2c 11 \u884c\u4e2d\u7684 `app = FastAPI` \u65c1\u8fb9\u7684\u5730\u7403\u56fe\u6807\uff0c\u7136\u540e\u9009\u62e9 <em>Show All Endpoints<\/em>\uff08\u663e\u793a\u6240\u6709\u7aef\u70b9\uff09\u3002<\/p>\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" loading=\"lazy\" class=\"wp-image-594174\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/08\/image-49.png\" alt=\"\u5728 PyCharm \u4e2d\u663e\u793a\u6240\u6709\u7aef\u70b9\" width=\"1600\" height=\"833\" \/><\/figure>\n<p>\u6211\u4eec\u6709\u4e09\u4e2a\u7aef\u70b9\u3002 \u7531\u4e8e\u6839\u7aef\u70b9\u53ea\u662f\u6b22\u8fce\u4fe1\u606f\uff0c\u6211\u4eec\u5c06\u91cd\u70b9\u67e5\u770b\u53e6\u5916\u4e24\u4e2a\u7aef\u70b9\u3002<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">@app.post(\"\/generate\", response_model=TextGenerationResponse)\nasync def generate_text(request: TextGenerationRequest):\n   \"\"\"\n   Generate text based on the provided prompt.\n  \n   Args:\n       request: TextGenerationRequest containing the prompt and generation parameters\n      \n   Returns:\n       TextGenerationResponse with the generated text\n   \"\"\"\n   if pipe is None:\n       raise HTTPException(status_code=500, detail=\"Model not loaded properly\")\n  \n   try:\n       result = pipe(\n           request.prompt,\n           max_new_tokens=request.max_new_tokens,\n           pad_token_id=pipe.tokenizer.eos_token_id\n       )\n      \n       # Extract the generated text from the result\n       generated_text = result[0]['generated_text']\n      \n       return TextGenerationResponse(generated_text=generated_text)\n   except Exception as e:\n       raise HTTPException(status_code=500, detail=f\"Error generating text: {str(e)}\")\n<\/pre>\n<p>`\/generate` \u7aef\u70b9\u4f1a\u6536\u96c6\u8bf7\u6c42\u63d0\u793a\uff0c\u5e76\u901a\u8fc7\u6a21\u578b\u751f\u6210\u56de\u7b54\u6587\u672c\u3002<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">@app.get(\"\/health\")\nasync def health_check():\n   \"\"\"Check if the API and model are working properly.\"\"\"\n   if pipe is None:\n       raise HTTPException(status_code=500, detail=\"Model not loaded\")\n   return {\"status\": \"healthy\", \"model_loaded\": True}<\/pre>\n<p>`\/health` \u7aef\u70b9\u7528\u4e8e\u68c0\u67e5\u6a21\u578b\u662f\u5426\u5df2\u6b63\u786e\u52a0\u8f7d\u3002 \u5982\u679c\u5ba2\u6237\u7aef\u5e94\u7528\u7a0b\u5e8f\u9700\u8981\u5148\u68c0\u67e5\u6a21\u578b\u72b6\u6001\uff0c\u7136\u540e\u5728\u5176 UI \u4e2d\u5f00\u653e\u5176\u4ed6\u7aef\u70b9\uff0c\u6b64\u7aef\u70b9\u4f1a\u975e\u5e38\u5b9e\u7528\u3002<\/p>\n<p>\u5728 `run.py` \u4e2d\uff0c\u6211\u4eec\u4f7f\u7528 <a href=\"https:\/\/www.uvicorn.org\/\" target=\"_blank\" rel=\"noopener\">uvicorn<\/a> \u6765\u8fd0\u884c\u670d\u52a1\u5668\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"python\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">import uvicorn\n\n\nif __name__ == \"__main__\":\n   uvicorn.run(\"main:app\", host=\"0.0.0.0\", port=8000, reload=True)<\/pre>\n<p>\u8fd0\u884c\u6b64\u811a\u672c\u65f6\uff0c\u670d\u52a1\u5668\u5c06\u5728 <a href=\"http:\/\/0.0.0.0:8000\/\" target=\"_blank\" rel=\"noopener\">http:\/\/0.0.0.0:8000\/<\/a> \u4e0b\u542f\u52a8\u3002<\/p>\n<p>\u5f00\u59cb\u8fd0\u884c\u670d\u52a1\u5668\u540e\uff0c\u6211\u4eec\u53ef\u4ee5\u8bbf\u95ee <a href=\"http:\/\/0.0.0.0:8000\/docs\" target=\"_blank\" rel=\"noopener\">http:\/\/0.0.0.0:8000\/docs<\/a> \u6765\u6d4b\u8bd5\u8fd9\u4e9b\u7aef\u70b9\u3002\u00a0<\/p>\n<h2 class=\"wp-block-heading\"><img decoding=\"async\" loading=\"lazy\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/09\/AD_4nXf6PXwz_Vz7VEQoyZs20NJ9TsO36oWJPf0w4iMjwHZ_EBP1Pk9c_8aWR2ybGE-wsmArM1zAQl1s8jHEr09I0g1A3boGD1Kt4i4CemufHZTHnATjIWrJ8x2ZUYg4Q7E4b3tc2XDmmg.png\" width=\"624\" height=\"269\" \/><\/h2>\n<p>\u6211\u4eec\u53ef\u4ee5\u901a\u8fc7 `\/generate` \u7aef\u70b9\u6765\u5c1d\u8bd5\u4e00\u4e0b\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">{\n  \"prompt\": \"5 people give each other a present. How many presents are given altogether?\",\n  \"max_new_tokens\": 300\n}<\/pre>\n<p>\u6211\u4eec\u5f97\u5230\u7684\u54cd\u5e94\u5982\u4e0b\u6240\u793a\uff1a<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">{\n  \"generated_text\": \"5 people give each other a present. How many presents are given altogether?nAlright, let's try to solve the problem:nn**Problem**  n1. Each person gives each other a present. How many presents are given altogether?n2. How many \"gift\" are given altogether?nn**Common pitfall**  nAssuming that each present is a \"gift\" without considering the implications of the original condition.nn---nn### Step 1: Attempting the problem (falling into the pitfall)nnOkay, so I have two people giving each other a present, and I want to know how many are present. I remember that there are three types of gifts\u2014gifts, gins, and ginses.nnLet me try to count how many of these:nn- Gifts: Let\u2019s say there are three people giving each other a present.n- Gins: Let\u2019s say there are three people giving each other a present.n- Ginses: Let\u2019s say there are three people giving each other a present.nnSo, total gins and ginses would be:nn- Gins: ( 2 times 3 = 1 ), ( 2 times 1 = 2 ), ( 1 times 1 = 1 ), ( 1 times 2 = 2 ), so ( 2 times 3 = 4 ).n- Ginses: ( 2 times 3 = 6 ), (\"\n}\n<\/pre>\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" loading=\"lazy\" class=\"wp-image-594185\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/08\/image-50.png\" alt=\"\" width=\"1600\" height=\"873\" \/><\/figure>\n<p>\u60a8\u53ef\u4ee5\u5c1d\u8bd5\u53d1\u9001\u5176\u4ed6\u8bf7\u6c42\u8fdb\u884c\u6d4b\u8bd5\u3002<\/p>\n<h2 class=\"wp-block-heading\">\u7ed3\u8bba\u548c\u540e\u7eed\u884c\u52a8<\/h2>\n<p>\u73b0\u5728\uff0c\u5927\u5bb6\u5df2\u7ecf\u6210\u529f\u4f7f\u7528\u6570\u5b66\u63a8\u7406\u6570\u636e\u96c6\u5fae\u8c03\u4e86\u50cf GPT-2 \u8fd9\u6837\u7684\u5927\u8bed\u8a00\u6a21\u578b\uff0c\u5e76\u901a\u8fc7 FastAPI \u5b8c\u6210\u4e86\u90e8\u7f72\u3002\u540e\u7eed\u5927\u5bb6\u53ef\u4ee5\u5bf9 Hugging Face Hub \u4e0a\u63d0\u4f9b\u7684\u66f4\u591a\u5f00\u6e90\u5927\u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u5fae\u8c03\u3002 \u60a8\u65e2\u53ef\u4ee5\u7528\u5e73\u53f0\u4e0a\u7684\u5f00\u6e90\u6570\u636e\uff0c\u4e5f\u53ef\u4ee5\u7528\u81ea\u5df1\u7684\u6570\u636e\u96c6\uff0c\u5c1d\u8bd5\u5fae\u8c03\u5176\u4ed6\u5927\u8bed\u8a00\u6a21\u578b\u3002 \u5982\u679c\u60a8\u60f3\uff08\u5e76\u4e14\u539f\u59cb\u6a21\u578b\u7684\u8bb8\u53ef\u8bc1\u4e5f\u652f\u6301\uff09\uff0c\u8fd8\u53ef\u4ee5\u5c06\u5fae\u8c03\u540e\u7684\u6a21\u578b\u4e0a\u4f20\u5230 Hugging Face Hub\u3002 \u5177\u4f53\u64cd\u4f5c\u53ef\u4ee5\u53c2\u8003\u8be5\u5e73\u53f0\u7684<a href=\"https:\/\/huggingface.co\/docs\/transformers\/v4.53.3\/en\/main_classes\/trainer#transformers.Trainer.push_to_hub\" target=\"_blank\" rel=\"noopener\">\u6587\u6863<\/a>\u3002<\/p>\n<p>\u5173\u4e8e\u4f7f\u7528 Hugging Face Hub \u4e0a\u7684\u8d44\u6e90\uff0c\u6216\u57fa\u4e8e\u8fd9\u4e9b\u8d44\u6e90\u5fae\u8c03\u6a21\u578b\uff0c\u6700\u540e\u8fd8\u6709\u4e00\u70b9\u9700\u8981\u6ce8\u610f\uff0c\u52a1\u5fc5\u9605\u8bfb\u60a8\u6240\u4f7f\u7528\u7684\u4efb\u4f55\u6a21\u578b\u6216\u6570\u636e\u96c6\u7684\u8bb8\u53ef\u8bc1\uff0c\u660e\u786e\u4f7f\u7528\u8fd9\u4e9b\u8d44\u6e90\u7684\u76f8\u5173\u6761\u4ef6\u3002 \u662f\u5426\u5141\u8bb8\u5546\u4e1a\u7528\u9014\uff1f \u662f\u5426\u9700\u8981\u6ce8\u660e\u6240\u4f7f\u7528\u8d44\u6e90\u7684\u6765\u6e90\uff1f<\/p>\n<p>\u5728\u540e\u7eed\u7684\u535a\u6587\u4e2d\uff0c\u6211\u4eec\u5c06\u7ee7\u7eed\u63a2\u7d22\u66f4\u591a\u6d89\u53ca Python\u3001AI\u3001\u673a\u5668\u5b66\u4e60\u4ee5\u53ca\u6570\u636e\u53ef\u89c6\u5316\u7684\u4ee3\u7801\u793a\u4f8b\u3002<\/p>\n<p>\u5728\u6211\u770b\u6765\uff0c<a href=\"https:\/\/www.jetbrains.com.cn\/pycharm\/\" target=\"_blank\" rel=\"noopener\">PyCharm<\/a> \u63d0\u4f9b\u4e86\u4e1a\u754c\u9876\u5c16\u7684 Python \u652f\u6301\uff0c\u80fd\u591f\u540c\u65f6\u4fdd\u8bc1\u901f\u5ea6\u4e0e\u51c6\u786e\u6027\u3002 \u60a8\u53ef\u4ee5\u5145\u5206\u5229\u7528\u5176\u6700\u667a\u80fd\u7684\u4ee3\u7801\u8865\u5168\u3001PEP 8 \u5408\u89c4\u6027\u68c0\u67e5\u3001\u667a\u80fd\u91cd\u6784\uff0c\u4ee5\u53ca\u5404\u7c7b\u68c0\u67e5\uff0c\u6ee1\u8db3\u60a8\u7684\u6240\u6709\u7f16\u7801\u9700\u6c42\u3002 \u6b63\u5982\u8fd9\u7bc7\u535a\u6587\u6240\u5c55\u793a\u7684\uff0cPyCharm \u8fd8\u96c6\u6210\u4e86 Hugging Face Hub \u529f\u80fd\uff0c\u8ba9\u60a8\u65e0\u9700\u79bb\u5f00 IDE\uff0c\u5373\u53ef\u6d4f\u89c8\u5e76\u4f7f\u7528\u6a21\u578b\u3002 \u8fd9\u4f7f\u5176\u975e\u5e38\u9002\u5408\u5f00\u5c55\u5404\u7c7b AI \u548c\u5927\u8bed\u8a00\u6a21\u578b\u5fae\u8c03\u9879\u76ee\u3002<\/p>\n<div class=\"buttons\">\n<div class=\"buttons__row\"><a class=\"btn\" href=\"https:\/\/www.jetbrains.com.cn\/pycharm\/\" target=\"\" rel=\"noopener\">\u7acb\u5373\u4e0b\u8f7d PyCharm<\/a><\/div>\n<\/div>\n\n\n<p><\/p>\n\n\n\n<p><\/p>\n\n\n\n<p>\u672c\u535a\u6587\u82f1\u6587\u539f\u4f5c\u8005\uff1a<\/p>\n\n\n    <div class=\"about-author \">\n        <div class=\"about-author__box\">\n            <div class=\"row\">\n                <div class=\"about-author__box-img\">\n                    <img decoding=\"async\" src=\"https:\/\/blog.jetbrains.com\/wp-content\/uploads\/2025\/01\/CheukTingHo-Kimono-e1738750639162-200x200.jpg\" width=\"200\" height=\"200\" alt=\"Cheuk Ting Ho\" loading=\"lazy\"  class=\"avatar avatar-200 wp-user-avatar wp-user-avatar-200 photo avatar-default\">\n                <\/div>\n                <div class=\"about-author__box-text\">\n                                            <h4>Cheuk Ting Ho<\/h4>\n                                                        <\/div>\n            <\/div>\n        <\/div>\n    <\/div>\n","protected":false},"author":1297,"featured_media":647865,"comment_status":"closed","ping_status":"closed","template":"","categories":[952,1401],"tags":[8900,8428,3252],"cross-post-tag":[8851],"acf":[],"_links":{"self":[{"href":"https:\/\/blog.jetbrains.com\/zh-hans\/wp-json\/wp\/v2\/pycharm\/647864"}],"collection":[{"href":"https:\/\/blog.jetbrains.com\/zh-hans\/wp-json\/wp\/v2\/pycharm"}],"about":[{"href":"https:\/\/blog.jetbrains.com\/zh-hans\/wp-json\/wp\/v2\/types\/pycharm"}],"author":[{"embeddable":true,"href":"https:\/\/blog.jetbrains.com\/zh-hans\/wp-json\/wp\/v2\/users\/1297"}],"replies":[{"embeddable":true,"href":"https:\/\/blog.jetbrains.com\/zh-hans\/wp-json\/wp\/v2\/comments?post=647864"}],"version-history":[{"count":2,"href":"https:\/\/blog.jetbrains.com\/zh-hans\/wp-json\/wp\/v2\/pycharm\/647864\/revisions"}],"predecessor-version":[{"id":647878,"href":"https:\/\/blog.jetbrains.com\/zh-hans\/wp-json\/wp\/v2\/pycharm\/647864\/revisions\/647878"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/blog.jetbrains.com\/zh-hans\/wp-json\/wp\/v2\/media\/647865"}],"wp:attachment":[{"href":"https:\/\/blog.jetbrains.com\/zh-hans\/wp-json\/wp\/v2\/media?parent=647864"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.jetbrains.com\/zh-hans\/wp-json\/wp\/v2\/categories?post=647864"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.jetbrains.com\/zh-hans\/wp-json\/wp\/v2\/tags?post=647864"},{"taxonomy":"cross-post-tag","embeddable":true,"href":"https:\/\/blog.jetbrains.com\/zh-hans\/wp-json\/wp\/v2\/cross-post-tag?post=647864"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}