{"id":10572,"date":"2026-06-10T02:19:24","date_gmt":"2026-06-09T18:19:24","guid":{"rendered":"https:\/\/ab123.xyz\/?p=10572"},"modified":"2026-06-10T02:19:24","modified_gmt":"2026-06-09T18:19:24","slug":"hugging-face-transformers-%e5%be%ae%e8%b0%83%e6%8c%87%e5%8d%97%ef%bc%9a%e5%ae%9e%e7%8e%b0%e6%83%85%e6%84%9f%e5%88%86%e6%9e%90%e6%a8%a1%e5%9e%8b%e7%9a%84%e9%ab%98%e6%95%88%e9%83%a8%e7%bd%b2","status":"publish","type":"post","link":"https:\/\/ab123.xyz\/?p=10572","title":{"rendered":"Hugging Face Transformers \u5fae\u8c03\u6307\u5357\uff1a\u5b9e\u73b0\u60c5\u611f\u5206\u6790\u6a21\u578b\u7684\u9ad8\u6548\u90e8\u7f72"},"content":{"rendered":"

\u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\u9886\u57df\uff0c\u60c5\u611f\u5206\u6790\u662f\u6700\u5e38\u89c1\u4e14\u5546\u4e1a\u4ef7\u503c\u6781\u9ad8\u7684\u5e94\u7528\u4e4b\u4e00\u3002\u5229\u7528\u9884\u8bad\u7ec3\u6a21\u578b\u8fdb\u884c\u5fae\u8c03\uff08Fine-Tuning\uff09\u53ef\u4ee5\u663e\u8457\u964d\u4f4e\u6570\u636e\u9700\u6c42\u5e76\u63d0\u5347\u51c6\u786e\u7387\u3002Hugging Face Transformers \u5b98\u65b9\u6587\u6863<\/a> \u63d0\u4f9b\u4e86\u6700\u5b8c\u6574\u3001\u6700\u6613\u7528\u7684\u5fae\u8c03\u5de5\u5177\u94fe\uff0c\u8ba9\u5f00\u53d1\u8005\u80fd\u591f\u5feb\u901f\u5c06 BERT\u3001RoBERTa \u7b49\u6a21\u578b\u9002\u914d\u5230\u7279\u5b9a\u60c5\u611f\u5206\u7c7b\u4efb\u52a1\u4e2d\u3002<\/p>\n

\u6838\u5fc3\u529f\u80fd\u4e0e\u5185\u7f6e\u7ec4\u4ef6<\/h2>\n

Hugging Face Transformers \u5e93\u6574\u5408\u4e86 Trainer API\u3001AutoModelForSequenceClassification \u4ee5\u53ca Data Collator \u7b49\u6a21\u5757\u3002\u5f00\u53d1\u8005\u53ea\u9700\u51c6\u5907 CSV \u6216 JSON \u683c\u5f0f\u7684\u6807\u6ce8\u6570\u636e\uff0c\u8c03\u7528 from_pretrained \u52a0\u8f7d\u9884\u8bad\u7ec3\u6743\u91cd\uff0c\u7136\u540e\u8c03\u7528 Trainer.train() \u5373\u53ef\u542f\u52a8\u5fae\u8c03\u3002\u5e93\u5185\u81ea\u52a8\u5904\u7406\u5b66\u4e60\u7387\u8c03\u5ea6\u3001\u68af\u5ea6\u7d2f\u79ef\u3001\u6df7\u5408\u7cbe\u5ea6\u8bad\u7ec3\uff0c\u5e76\u652f\u6301 TensorBoard \u53ef\u89c6\u5316\u3002<\/p>\n

\u6570\u636e\u96c6\u5904\u7406\u4e0e Tokenizer \u96c6\u6210<\/h3>\n

\u5e93\u63d0\u4f9b\u7684 Tokenizer \u80fd\u591f\u81ea\u52a8\u5c06\u4e2d\u6587\u6216\u82f1\u6587\u6587\u672c\u5207\u5206\u4e3a subword \u5e8f\u5217\uff0c\u5e76\u586b\u5145\u6216\u622a\u65ad\u4e3a\u7edf\u4e00\u957f\u5ea6\u3002\u914d\u5408 datasets \u5e93\uff0c\u53ef\u4ee5\u9ad8\u6548\u52a0\u8f7d IMDB\u3001Yelp \u6216\u81ea\u5b9a\u4e49\u8bc4\u8bba\u6587\u672c\uff0c\u65e0\u9700\u624b\u52a8\u7f16\u5199\u6570\u636e\u7ba1\u9053\u3002<\/p>\n

\u6838\u5fc3\u4f18\u52bf\uff1a\u9884\u8bad\u7ec3\u8fc1\u79fb\u4e0e\u751f\u6001\u517c\u5bb9<\/h2>\n

\u76f8\u6bd4\u4ece\u96f6\u8bad\u7ec3\uff0c\u5fae\u8c03\u53ea\u9700\u5c11\u91cf\u6807\u6ce8\u6570\u636e\uff08\u4f8b\u5982\u6bcf\u7c7b 100 \u6761\uff09\u3002\u6b64\u5916\uff0cHugging Face \u6a21\u578b\u4ed3\u5e93\u62e5\u6709\u8d85\u8fc7 10 \u4e07\u4e2a\u793e\u533a\u8d21\u732e\u7684 checkpoint\uff0c\u4e2d\u6587\u9886\u57df\u4e5f\u6709 chinese-bert-wwm\u3001bert-base-chinese \u7b49\u4f18\u79c0\u57fa\u7840\u6a21\u578b\u3002\u5e93\u901a\u8fc7 PyTorch \u548c TensorFlow \u53cc\u540e\u7aef\u652f\u6301\uff0c\u5e76\u80fd\u65e0\u7f1d\u5bf9\u63a5 ONNX Runtime \u548c AWS SageMaker \u8fdb\u884c\u751f\u4ea7\u90e8\u7f72\u3002<\/p>\n

\u63a8\u7406\u4f18\u5316\u4e0e\u6a21\u578b\u538b\u7f29<\/h3>\n

\u5fae\u8c03\u540e\u7684\u6a21\u578b\u53ef\u901a\u8fc7 Optimum \u5e93\u8fdb\u884c INT8 \u91cf\u5316\u6216\u77e5\u8bc6\u84b8\u998f\uff0c\u5728\u4fdd\u6301 95% \u4ee5\u4e0a\u51c6\u786e\u7387\u7684\u540c\u65f6\u5c06\u63a8\u7406\u901f\u5ea6\u63d0\u5347 3-5 \u500d\uff0c\u7279\u522b\u9002\u5408\u79fb\u52a8\u7aef\u6216\u9ad8\u5e76\u53d1\u573a\u666f\u3002<\/p>\n

\u5178\u578b\u5e94\u7528\u573a\u666f<\/h2>\n
    \n
  • \u7535\u5546\u8bc4\u8bba\u5206\u6790<\/strong>\uff1a\u81ea\u52a8\u8bc6\u522b\u7528\u6237\u5bf9\u5546\u54c1\u7684\u60c5\u611f\u503e\u5411\uff08\u6b63\u9762\/\u8d1f\u9762\/\u4e2d\u6027\uff09\uff0c\u8f85\u52a9\u5ba2\u670d\u5206\u7ea7\u4e0e\u8206\u60c5\u76d1\u63a7\u3002<\/li>\n
  • \u793e\u4ea4\u5a92\u4f53\u8206\u60c5<\/strong>\uff1a\u5b9e\u65f6\u6293\u53d6\u5fae\u535a\u3001\u5c0f\u7ea2\u4e66\u5e16\u6587\uff0c\u5206\u6790\u54c1\u724c\u63d0\u53ca\u7684\u6b63\u5411\u6216\u8d1f\u5411\u60c5\u7eea\uff0c\u652f\u6301\u5371\u673a\u9884\u8b66\u3002<\/li>\n
  • \u91d1\u878d\u6587\u672c\u5206\u6790<\/strong>\uff1a\u5bf9\u7814\u62a5\u3001\u65b0\u95fb\u6807\u9898\u8fdb\u884c\u60c5\u611f\u8bc4\u5206\uff0c\u8f85\u52a9\u91cf\u5316\u4ea4\u6613\u7b56\u7565\u4e2d\u7684\u60c5\u7eea\u56e0\u5b50\u6784\u5efa\u3002<\/li>\n<\/ul>\n

    \u5feb\u901f\u4e0a\u624b\u6b65\u9aa4<\/h2>\n

    \u9996\u5148\u901a\u8fc7 pip install transformers[torch] datasets \u5b89\u88c5\u73af\u5883\u3002\u63a5\u4e0b\u6765\u7f16\u5199\u811a\u672c\u52a0\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b\u4e0e\u5206\u8bcd\u5668\u3002\u7136\u540e\u5b9a\u4e49\u8bad\u7ec3\u53c2\u6570\uff0c\u5982\u8f93\u51fa\u76ee\u5f55\u3001batch size \u4ee5\u53ca\u8bc4\u4f30\u7b56\u7565\u3002\u6700\u540e\u8c03\u7528 trainer.train() \u5e76\u4fdd\u5b58\u6a21\u578b\u3002\u5b98\u65b9\u6587\u6863\u63d0\u4f9b\u5b8c\u6574\u7684 Colab \u7b14\u8bb0\u672c\uff0c\u53ef\u5728\u6d4f\u89c8\u5668\u4e2d\u76f4\u63a5\u8fd0\u884c\u3002<\/p>\n

    \u5173\u952e\u8c03\u53c2\u5efa\u8bae<\/h3>\n

    \u5b66\u4e60\u7387\u5efa\u8bae\u8bbe\u4e3a 2e-5 \u81f3 5e-5\uff0c\u4f7f\u7528 AdamW \u4f18\u5316\u5668\uff1b\u82e5\u6570\u636e\u4e0d\u5e73\u8861\uff0c\u53ef\u8bbe\u7f6e class_weight \u6216\u91c7\u7528 Focal Loss \u81ea\u5b9a\u4e49\u635f\u5931\u51fd\u6570\u3002\u5bf9\u4e8e\u8f83\u5c0f\u6570\u636e\u96c6\uff08<500\u6761\uff09\uff0c\u5efa\u8bae\u4f7f\u7528\u51bb\u7ed3\u524d\u51e0\u5c42\u6216\u8fdb\u884c\u9010\u6b65\u89e3\u51bb\u7b56\u7565\u4ee5\u907f\u514d\u8fc7\u62df\u5408\u3002<\/p>\n

    \u603b\u4e4b\uff0cHugging Face Transformers \u7684\u5fae\u8c03\u5de5\u5177\u94fe\u5c06\u590d\u6742\u7684\u5927\u6a21\u578b\u9002\u914d\u8fc7\u7a0b\u7b80\u5316\u4e3a\u51e0\u884c\u4ee3\u7801\uff0c\u662f\u5f53\u524d\u60c5\u611f\u5206\u6790\u9879\u76ee\u843d\u5730\u7684\u6700\u4f73\u5b9e\u8df5\u4e4b\u4e00\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

    \u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\u9886\u57df\uff0c\u60c5\u611f\u5206\u6790\u662f\u6700\u5e38\u89c1\u4e14\u5546\u4e1a\u4ef7\u503c\u6781\u9ad8\u7684\u5e94\u7528\u4e4b\u4e00\u3002\u5229\u7528\u9884\u8bad\u7ec3\u6a21\u578b\u8fdb\u884c\u5fae\u8c03\uff08Fine-Tuning\uff09 […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[4],"tags":[13128,13130,13129,13131,13089],"class_list":["post-10572","post","type-post","status-publish","format-standard","hentry","category-4","tag-hugging-face-transformer","tag-nlp-","tag-13129","tag-13131","tag-13089"],"_links":{"self":[{"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/posts\/10572","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ab123.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=10572"}],"version-history":[{"count":1,"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/posts\/10572\/revisions"}],"predecessor-version":[{"id":10573,"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/posts\/10572\/revisions\/10573"}],"wp:attachment":[{"href":"https:\/\/ab123.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=10572"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ab123.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=10572"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ab123.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=10572"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}