\n1. \uc18c\uac1c<\/h2>\n
\n \ub525\ub7ec\ub2dd\uc758 \ubc1c\uc804\uc740 \uc790\uc5f0\uc5b4 \ucc98\ub9ac(NLP) \ubd84\uc57c\uc5d0\uc11c \ud2b9\ud788 \ub208\ubd80\uc2e0 \uc131\uacfc\ub97c \uc774\ub8e9\ud574\uc654\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ubc1c\uc804\uc758 \uc911\uc2ec\uc5d0\ub294
\n \uc0ac\uc804\ud559\uc2b5 \ubaa8\ub378\uc774 \uc788\uc2b5\ub2c8\ub2e4. \ud5c8\uae45\ud398\uc774\uc2a4(Hugging Face)\ub294 \uc774\ub7ec\ud55c \uc0ac\uc804\ud559\uc2b5 \ubaa8\ub378\uc744 \ubcf4\ub2e4 \uc27d\uac8c \uc0ac\uc6a9\ud560 \uc218 \uc788\ub3c4\ub85d
\n \ud574\uc8fc\ub294 \uac15\ub825\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\uc778 Transformers<\/strong>\ub97c \uc81c\uacf5\ud569\ub2c8\ub2e4. \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \ud5c8\uae45\ud398\uc774\uc2a4\uc758
\n Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc0ac\uc804\ud559\uc2b5 \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc790\uc138\ud788 \uc54c\uc544\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.\n <\/p>\n<\/section>\n\n2. \ud5c8\uae45\ud398\uc774\uc2a4 Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac\ub780?<\/h2>\n
\n \ud5c8\uae45\ud398\uc774\uc2a4 Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac\ub294 \ub2e4\uc591\ud55c \uc790\uc5f0\uc5b4 \ucc98\ub9ac(NLP) \ubaa8\ub378\uc744 \uc81c\uacf5\ud558\ub294 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub85c,
\n BERT<\/strong>, GPT<\/strong>, RoBERTa<\/strong>, T5<\/strong> \ub4f1
\n \uc5ec\ub7ec \ubaa8\ub378\uc744 \ud3ec\ud568\ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc774 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud1b5\ud574 \uac1c\ubc1c\uc790\ub294 \uc190\uc27d\uac8c \uc0ac\uc804\ud559\uc2b5\ub41c \uc5b8\uc5b4 \ubaa8\ub378\uc744 \ub85c\ub4dc\ud558\uace0
\n \uc774\ub97c \uae30\ubc18\uc73c\ub85c \ub2e4\uc591\ud55c NLP \uc791\uc5c5\uc744 \uc218\ud589\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n<\/section>\n\n3. \ud658\uacbd \uc124\uc815<\/h2>\n
\n \uc2dc\uc791\ud558\uae30 \uc804\uc5d0 \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud574\uc57c \ud569\ub2c8\ub2e4. \uc544\ub798\uc758 \uba85\ub839\uc5b4\ub97c \ud1b5\ud574 \uae30\ubcf8\uc801\uc778 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c
\n \uc124\uce58\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n
pip install transformers torch<\/code><\/pre>\n\n \uc5ec\uae30\uc11c, transformers<\/strong>\ub294 \ud5c8\uae45\ud398\uc774\uc2a4 \ub77c\uc774\ube0c\ub7ec\ub9ac\uc774\uba70, torch<\/strong>\ub294
\n PyTorch \ud504\ub808\uc784\uc6cc\ud06c\uc785\ub2c8\ub2e4. PyTorch \ub300\uc2e0 TensorFlow\ub97c \uc0ac\uc6a9\ud558\uace0 \uc2f6\ub2e4\uba74 TensorFlow\ub97c \uc124\uce58\ud558\uba74 \ub429\ub2c8\ub2e4.\n <\/p>\n<\/section>\n\n4. \uc0ac\uc804\ud559\uc2b5 \ubaa8\ub378 \ubd88\ub7ec\uc624\uae30<\/h2>\n
\n \uc774\uc81c \uc0ac\uc804\ud559\uc2b5\ub41c \ubaa8\ub378\uc744 \ubd88\ub7ec\uc640 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, BERT<\/strong> \ubaa8\ub378\uc744 \uc774\uc6a9\ud558\uc5ec
\n \ud14d\uc2a4\ud2b8\uc758 \uc758\ubbf8\ub97c \uc774\ud574\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc544\ub798\ub294 Python \ucf54\ub4dc\ub97c \ud1b5\ud574 BERT \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\ub294 \ubc29\ubc95\uc785\ub2c8\ub2e4.\n <\/p>\nfrom transformers import BertTokenizer, BertModel\n\n# BERT \ubaa8\ub378\uc758 \ud1a0\ud06c\ub098\uc774\uc800\uc640 \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\uae30\ntokenizer = BertTokenizer.from_pretrained('bert-base-uncased')\nmodel = BertModel.from_pretrained('bert-base-uncased')\n\n# \uc608\uc81c \ubb38\uc7a5\nsentence = \"Hugging Face is creating a tool that democratizes AI.\"\n\n# \ubb38\uc7a5\uc744 \ud1a0\ud070\ud654\ud558\uc5ec \uc785\ub825 \ubca1\ud130\ub85c \ubcc0\ud658\ninputs = tokenizer(sentence, return_tensors='pt')\noutputs = model(**inputs)\n\n# \ucd9c\ub825 \uacb0\uacfc \ud655\uc778\nprint(outputs)\n<\/code><\/pre>\n\n \uc704 \ucf54\ub4dc\uc5d0\uc11c BertTokenizer<\/strong> \ud074\ub798\uc2a4\ub294 \uc785\ub825\ub41c \ubb38\uc7a5\uc744 BERT \ubaa8\ub378\uc774 \uc774\ud574\ud560 \uc218 \uc788\ub294
\n \ud615\ud0dc\ub85c \ubcc0\ud658\ud569\ub2c8\ub2e4. BertModel<\/strong> \ud074\ub798\uc2a4\ub294 \uc2e4\uc81c \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\uace0, \ubcc0\ud658\ub41c
\n \uc785\ub825\uc744 \ubaa8\ub378\uc5d0 \ud1b5\uacfc\uc2dc\ucf1c \ucd9c\ub825\uc744 \uc0dd\uc131\ud569\ub2c8\ub2e4.\n <\/p>\n<\/section>\n\n5. \ucd9c\ub825 \uacb0\uacfc \ubd84\uc11d<\/h2>\n
\n \uc704\uc758 \ucf54\ub4dc\uc5d0\uc11c \ucd9c\ub825\ub418\ub294 outputs<\/strong> \ubcc0\uc218\ub294 \ub450 \uac00\uc9c0 \uc8fc\uc694 \uc815\ubcf4\ub97c \ud3ec\ud568\ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4:\n <\/p>\n\n- last_hidden_state<\/strong>: \ub9c8\uc9c0\ub9c9 \uc740\ub2c9 \uc0c1\ud0dc(hidden state)\ub85c, \uac01 \ud1a0\ud070\uc758 \ubca1\ud130 \ud45c\ud604\uc744 \ubcf4\uc5ec\uc90d\ub2c8\ub2e4.<\/li>\n
- pooler_output<\/strong>: \uc785\ub825 \uc2dc\ud000\uc2a4\uc758 \uc804\uccb4\ub97c \uc694\uc57d\ud55c \ubca1\ud130\ub85c, \uc8fc\ub85c \ubd84\ub958 \uc791\uc5c5\uc5d0 \uc0ac\uc6a9\ub429\ub2c8\ub2e4.<\/li>\n<\/ul>\n
\n \uac01 \ud1a0\ud070\uc758 \ubca1\ud130 \ud45c\ud604\uc740 \uc790\uc5f0\uc5b4 \ucc98\ub9ac\ub97c \uc704\ud55c \ub9e4\uc6b0 \uc720\uc6a9\ud55c \uc815\ubcf4\uc785\ub2c8\ub2e4. \uac01 \ud1a0\ud070\uc5d0 \ub300\ud574 \ucd9c\ub825\ub41c
\n \uc740\ub2c9 \uc0c1\ud0dc\ub294 \uc544\ub798\uc640 \uac19\uc774 \uc811\uadfc\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n
# \ub9c8\uc9c0\ub9c9 \uc740\ub2c9 \uc0c1\ud0dc \uc811\uadfc\nhidden_states = outputs.last_hidden_state\nprint(hidden_states.shape) # (\ubc30\uce58 \ud06c\uae30, \uc2dc\ud000\uc2a4 \uae38\uc774, \uc228\uaca8\uc9c4 \ucc28\uc6d0)\n<\/code><\/pre>\n<\/section>\n\n6. \ub2e4\uc591\ud55c \uc0ac\uc804\ud559\uc2b5 \ubaa8\ub378 \uc0ac\uc6a9\ud558\uae30<\/h2>\n
\n \ud5c8\uae45\ud398\uc774\uc2a4\ub294 BERT \uc678\uc5d0\ub3c4 \uc5ec\ub7ec \ub2e4\ub978 \ubaa8\ub378\uc744 \uc9c0\uc6d0\ud569\ub2c8\ub2e4. \uc0ac\uc6a9\uc790\uac00 \ub2e4\uc591\ud55c \uc791\uc5c5\uc5d0 \uc801\ud569\ud55c
\n \ubaa8\ub378\uc744 \uc120\ud0dd\ud560 \uc218 \uc788\ub3c4\ub85d \uc5ec\ub7ec \ubaa8\ub378\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4. GPT-2<\/strong>, RoBERTa<\/strong>,
\n T5<\/strong> \ub4f1 \uac01 \ubaa8\ub378\uc758 \uc0ac\uc6a9\ubc95\uc740 \ub300\ub3d9\uc18c\uc774\ud569\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, GPT-2<\/strong> \ubaa8\ub378\uc744
\n \uc0ac\uc6a9\ud558\uace0\uc790 \ud560 \uacbd\uc6b0, \uc544\ub798\uc640 \uac19\uc740 \ubc29\ubc95\uc73c\ub85c \ubd88\ub7ec\uc62c \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\nfrom transformers import GPT2Tokenizer, GPT2Model\n\n# GPT-2 \ubaa8\ub378\uc758 \ud1a0\ud06c\ub098\uc774\uc800\uc640 \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\uae30\ntokenizer = GPT2Tokenizer.from_pretrained('gpt2')\nmodel = GPT2Model.from_pretrained('gpt2')\n\n# \uc608\uc81c \ubb38\uc7a5\nsentence = \"Hugging Face has become a leader in NLP.\"\n\n# \ubb38\uc7a5\uc744 \ud1a0\ud070\ud654\ud558\uc5ec \uc785\ub825 \ubca1\ud130\ub85c \ubcc0\ud658\ninputs = tokenizer(sentence, return_tensors='pt')\noutputs = model(**inputs)\n\nprint(outputs)\n<\/code><\/pre>\n<\/section>\n\n7. \uc9c1\uc811 \ubaa8\ub378 \ud559\uc2b5\ud558\uae30<\/h2>\n
\n \uc0ac\uc804\ud559\uc2b5\ub41c \ubaa8\ub378\uc744 \uac00\uc838\uc624\ub294 \uac83\ubfd0\ub9cc \uc544\ub2c8\ub77c, \uc0ac\uc6a9\uc790\ub294 \uc790\uc2e0\uc758 \ub370\uc774\ud130\uc14b\uc5d0 \ub9de\ucdb0 \ubaa8\ub378\uc744 \ubbf8\uc138 \uc870\uc815\ud560
\n \uc218\ub3c4 \uc788\uc2b5\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc740 \uc544\ub798\uc758 \ub2e8\uacc4\ub97c \ud1b5\ud574 \uc9c4\ud589\ub429\ub2c8\ub2e4:\n <\/p>\n
\n- \ub370\uc774\ud130 \uc900\ube44 \ubc0f \uc804\ucc98\ub9ac<\/li>\n
- \uc0ac\uc804\ud559\uc2b5\ub41c \ubaa8\ub378 \ubd88\ub7ec\uc624\uae30<\/li>\n
- \ud559\uc2b5\uc744 \uc704\ud55c \uc190\uc2e4 \ud568\uc218 \ubc0f \uc635\ud2f0\ub9c8\uc774\uc800 \uc124\uc815<\/li>\n
- \ubaa8\ub378 \ud559\uc2b5<\/li>\n<\/ol>\n
7.1 \ub370\uc774\ud130 \uc900\ube44 \ubc0f \uc804\ucc98\ub9ac<\/h3>\n
\n \ub370\uc774\ud130\ub294 CSV \ud30c\uc77c \ub4f1\uc758 \ud615\uc2dd\uc73c\ub85c \uc900\ube44\ud560 \uc218 \uc788\uc73c\uba70, \uc774\ub97c \ubd88\ub7ec\uc640 \uc804\ucc98\ub9ac\ud558\ub294 \uc77c\ub828\uc758 \uacfc\uc815\uc774 \ud544\uc694\ud569\ub2c8\ub2e4.\n <\/p>\n
import pandas as pd\n\n# \ub370\uc774\ud130\uc14b \ubd88\ub7ec\uc624\uae30\ndata = pd.read_csv('data.csv')\nprint(data.head()) # \ub370\uc774\ud130\uc14b\uc758 \uccab 5\ud589 \ud655\uc778\n<\/code><\/pre>\n7.2 \uc0ac\uc804\ud559\uc2b5\ub41c \ubaa8\ub378 \ubd88\ub7ec\uc624\uae30<\/h3>\n
\n \uc704\uc5d0\uc11c \uc124\uba85\ud55c \ubc29\uc2dd\uc73c\ub85c \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\uba74 \ub429\ub2c8\ub2e4.\n <\/p>\n
7.3 \uc190\uc2e4 \ud568\uc218 \ubc0f \uc635\ud2f0\ub9c8\uc774\uc800 \uc124\uc815<\/h3>\n
\n \ubaa8\ub378 \ud559\uc2b5\uc744 \uc704\ud574\uc11c\ub294 \uc190\uc2e4 \ud568\uc218\uc640 \uc635\ud2f0\ub9c8\uc774\uc800\ub97c \uc124\uc815\ud574\uc57c \ud569\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uba74,
\n AdamW<\/strong> \uc635\ud2f0\ub9c8\uc774\uc800\uc640 CrossEntropyLoss<\/strong> \uc190\uc2e4 \ud568\uc218\ub97c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\nfrom transformers import AdamW\n\noptimizer = AdamW(model.parameters(), lr=5e-5) # \ud559\uc2b5\ub960 \uc124\uc815\nloss_fn = torch.nn.CrossEntropyLoss()\n<\/code><\/pre>\n7.4 \ubaa8\ub378 \ud559\uc2b5<\/h3>\n
\n \uc804\ucc98\ub9ac\ub41c \ub370\uc774\ud130\uc640 \uc124\uc815\ub41c \uc190\uc2e4 \ud568\uc218 \ubc0f \uc635\ud2f0\ub9c8\uc774\uc800\ub97c \uc774\uc6a9\ud558\uc5ec \ubaa8\ub378\uc744 \ud559\uc2b5\uc2dc\ud0ac \uc218 \uc788\uc2b5\ub2c8\ub2e4.
\n \uc77c\ubc18\uc801\uc73c\ub85c \uc5d0\ud3ed\uc744 \uc124\uc815\ud558\uace0 \uc774\ub97c \ubc18\ubcf5\ud558\uc5ec \ubaa8\ub378\uc744 \ucd5c\uc801\ud654\ud569\ub2c8\ub2e4.\n <\/p>\n
for epoch in range(num_epochs):\n model.train()\n outputs = model(**inputs) # \ubaa8\ub378\uc758 \ucd9c\ub825\n loss = loss_fn(outputs, labels) # \uc190\uc2e4 \uacc4\uc0b0\n loss.backward() # \uc5ed\uc804\ud30c\n optimizer.step() # \uac00\uc911\uce58 \uc5c5\ub370\uc774\ud2b8\n optimizer.zero_grad() # \uae30\uc6b8\uae30 \ucd08\uae30\ud654\n<\/code><\/pre>\n<\/section>\n\n8. \uacb0\ub860<\/h2>\n
\n \ubcf8 \uac15\uc88c\ub97c \ud1b5\ud574 \ud5c8\uae45\ud398\uc774\uc2a4\uc758 Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud65c\uc6a9\ud558\uc5ec \uc0ac\uc804\ud559\uc2b5 \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\uace0,
\n \ub610 \uc774\ub97c \uae30\ubc18\uc73c\ub85c \ub2e4\uc591\ud55c \uc791\uc5c5\uc744 \uc218\ud589\ud560 \uc218 \uc788\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc54c\uc544\ubcf4\uc558\uc2b5\ub2c8\ub2e4. \uc774 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub294
\n \uc790\uc5f0\uc5b4 \ucc98\ub9ac \ubd84\uc57c\uc5d0\uc11c \uac15\ub825\ud55c \ub3c4\uad6c\uac00 \ub418\uba70, \ud2b9\ud788 \uc77c\uc815\ud55c \ub370\uc774\ud130\uc14b\uacfc \ud3c9\uc18c\uc758 \uc190\uc26c\uc6b4 API\ub97c \ud1b5\ud574
\n \ubaa8\ub378\uc744 \ud65c\uc6a9\ud560 \uc218 \uc788\ub3c4\ub85d \ub3d5\uc2b5\ub2c8\ub2e4. \uc774\uc81c \uc5ec\ub7ec\ubd84\uc740 \ud5c8\uae45\ud398\uc774\uc2a4\uc758 \ud2b8\ub80c\uc2a4\ud3ec\uba38\ub97c \ud1b5\ud574
\n \uc790\uc2e0\uc758 \ud504\ub85c\uc81d\ud2b8\uc5d0 \ud65c\uc6a9\ud560 \uc218 \uc788\ub294 \ub2a5\ub825\uc744 \uac16\ucd94\uc5c8\uc2b5\ub2c8\ub2e4.\n <\/p>\n<\/section>\n