\n\ub525\ub7ec\ub2dd\uc744 \ud65c\uc6a9\ud55c \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc758 \uc0c8\ub85c\uc6b4 \uc7a5\uc744 \uc5f4\uae30 \uc704\ud55c \uc5ec\uc815\uc5d0 \uc5ec\ub7ec\ubd84\uc744 \ucd08\ub300\ud569\ub2c8\ub2e4. \uc774\ubc88 \uac15\uc88c\uc5d0\uc11c\ub294 \ud5c8\uae45\ud398\uc774\uc2a4(Hugging Face)\uc5d0\uc11c \uc81c\uacf5\ud558\ub294 \ud2b8\ub80c\uc2a4\ud3ec\uba38 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud65c\uc6a9\ud558\uc5ec DistilGPT2 \ubaa8\ub378\uc744 \uc774\uc6a9\ud55c \ubb38\uc7a5 \uc0dd\uc131\uc744 \ub2e4\ub8e8\uc5b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc6b4\uc601\uccb4\uc81c\uc640 \uad00\uacc4\uc5c6\uc774 \uc27d\uac8c \uc124\uce58\ud558\uace0 \uc0ac\uc6a9\ud560 \uc218 \uc788\ub294 \ubc29\ubc95\uc744 \ud3ec\ud568\ud558\uc5ec, \uc2e4\uc2b5\uc744 \ud1b5\ud574 \uc9c1\uc811 \ubb38\uc7a5\uc744 \uc0dd\uc131\ud574\ubcf4\ub294 \uc2dc\uac04\uc744 \uac00\uc9c8 \uac83\uc785\ub2c8\ub2e4.<\/p>\n<\/header>\n\n1. \ud5c8\uae45\ud398\uc774\uc2a4(Hugging Face)\ub780?<\/h2>\n
\ud5c8\uae45\ud398\uc774\uc2a4\ub294 \uc790\uc5f0\uc5b4 \ucc98\ub9ac(NLP) \ubc0f \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \uc27d\uac8c \uc0ac\uc6a9\ud560 \uc218 \uc788\ub3c4\ub85d \ub3c4\uc640\uc8fc\ub294 \ud50c\ub7ab\ud3fc\uc785\ub2c8\ub2e4. \ud2b9\ud788 \ud2b8\ub80c\uc2a4\ud3ec\uba38 \ubaa8\ub378\uc744 \ub9e4\uc6b0 \uc27d\uac8c \uc0ac\uc6a9\ud560 \uc218 \uc788\ub3c4\ub85d \ub2e4\uc591\ud55c API\uc640 \ub3c4\uad6c\ub4e4\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4. GPT-2\uc640 \uac19\uc740 \ubaa8\ub378\uc740 \uc790\uc5f0\uc5b4 \uc0dd\uc131\uc5d0 \ud5a5\uc0c1\ub41c \uc131\ub2a5\uc744 \ubcf4\uc5ec\uc8fc\uba70, \uc774 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud1b5\ud574 \uc774\ub7ec\ud55c \ubaa8\ub378\uc744 \uc190\uc27d\uac8c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n<\/section>\n\n2. DistilGPT2 \ubaa8\ub378 \uc18c\uac1c<\/h2>\n
DistilGPT2\ub294 OpenAI\uc5d0\uc11c \uac1c\ubc1c\ud55c GPT-2 \ubaa8\ub378\uc758 \uacbd\ub7c9\ud654 \ubc84\uc804\uc73c\ub85c, \ud30c\ub77c\ubbf8\ud130 \uc218\ub97c \uac10\uc18c\uc2dc\ucf1c \ub354 \ube60\ub978 \uc18d\ub3c4\ub85c \uc791\ub3d9\ud558\uba74\uc11c\ub3c4 \uc131\ub2a5\uc740 \ube44\uc2b7\ud55c \uc218\uc900\uc744 \uc720\uc9c0\ud569\ub2c8\ub2e4. \uc774\ub294 \uc11c\ubc84 \ub9ac\uc18c\uc2a4\ub97c \uc808\uc57d\ud560 \uc218 \uc788\uc73c\uba70, \uc77c\ubc18 \uc0ac\uc6a9\uc790\ub4e4\uc774 \uc0ac\uc6a9\ud558\uae30 \uc6a9\uc774\ud558\ub3c4\ub85d \ub9cc\ub4e4\uc5b4\uc84c\uc2b5\ub2c8\ub2e4.<\/p>\n
DistilGPT2\ub294 \uc8fc\uc5b4\uc9c4 \ud14d\uc2a4\ud2b8\uc758 \ub9e5\ub77d\uc744 \uc774\ud574\ud558\uace0, \uadf8\uc5d0 \ub9de\ub294 \ucd94\uac00 \ud14d\uc2a4\ud2b8\ub97c \uc0dd\uc131\ud558\ub294 \ub370 \ub6f0\uc5b4\ub09c \ub2a5\ub825\uc744 \uac00\uc9c0\uace0 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n<\/section>\n\n3. \uc2e4\uc2b5 \ud658\uacbd \uc124\uc815<\/h2>\n3.1. \uc124\uce58 \uc694\uad6c\uc0ac\ud56d<\/h3>\n
\uc774 \uac15\uc88c\uc5d0\uc11c\ub294 Python 3.6 \uc774\uc0c1\uacfc \ub2e4\uc74c \ud328\ud0a4\uc9c0\uac00 \ud544\uc694\ud569\ub2c8\ub2e4:<\/p>\n
\n- transformers<\/li>\n
- torch<\/li>\n
- numpy<\/li>\n<\/ul>\n
3.2. \ud328\ud0a4\uc9c0 \uc124\uce58<\/h3>\n
\ub2e4\uc74c \uba85\ub839\uc5b4\ub97c \uc2e4\ud589\ud558\uc5ec \ud544\uc694\ud55c \ud328\ud0a4\uc9c0\ub97c \uc124\uce58\ud569\ub2c8\ub2e4:<\/p>\n
pip install transformers torch numpy<\/code><\/pre>\n<\/section>\n\n4. DistilGPT2 \ubb38\uc7a5 \uc0dd\uc131<\/h2>\n
\uc774\uc81c DistilGPT2 \ubaa8\ub378\uc744 \ud65c\uc6a9\ud558\uc5ec \ubb38\uc7a5\uc744 \uc0dd\uc131\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uba3c\uc800 \uae30\ubcf8\uc801\uc778 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc784\ud3ec\ud2b8\ud558\uace0, \ubaa8\ub378\uacfc \ud1a0\ud06c\ub098\uc774\uc800\ub97c \uc124\uc815\ud569\ub2c8\ub2e4.<\/p>\n
4.1. \ubaa8\ub378\uacfc \ud1a0\ud06c\ub098\uc774\uc800 \ubd88\ub7ec\uc624\uae30<\/h3>\n
\uc544\ub798\uc758 \ucf54\ub4dc\ub97c \uc0ac\uc6a9\ud558\uc5ec \ud544\uc694 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ubd88\ub7ec\uc624\uace0, \ubaa8\ub378\uacfc \ud1a0\ud06c\ub098\uc774\uc800\ub97c \uc124\uc815\ud569\ub2c8\ub2e4.<\/p>\n
from transformers import GPT2LMHeadModel, GPT2Tokenizer\n\n# \ubaa8\ub378\uacfc \ud1a0\ud06c\ub098\uc774\uc800 \ubd88\ub7ec\uc624\uae30\ntokenizer = GPT2Tokenizer.from_pretrained(\"distilgpt2\")\nmodel = GPT2LMHeadModel.from_pretrained(\"distilgpt2\")<\/code><\/pre>\n4.2. \ud14d\uc2a4\ud2b8 \uc0dd\uc131 \ud568\uc218 \uc815\uc758<\/h3>\n
\ub2e4\uc74c\uc73c\ub85c, \uc8fc\uc5b4\uc9c4 \ud504\ub86c\ud504\ud2b8\ub97c \uae30\ubc18\uc73c\ub85c \ubb38\uc7a5\uc744 \uc0dd\uc131\ud558\ub294 \ud568\uc218\ub97c \uc815\uc758\ud558\uaca0\uc2b5\ub2c8\ub2e4. \uc774 \ud568\uc218\ub294 \ud504\ub86c\ud504\ud2b8\ub97c \ud1a0\ud070\ud654\ud558\uace0, \ubaa8\ub378\uc744 \ud1b5\ud574 \uc0c8\ub85c\uc6b4 \ud14d\uc2a4\ud2b8\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4.<\/p>\n
import torch\n\ndef generate_text(prompt, max_length=50):\n # \ud504\ub86c\ud504\ud2b8\ub97c \ud1a0\ud070\ud654\n inputs = tokenizer.encode(prompt, return_tensors=\"pt\")\n\n # \ubb38\uc7a5 \uc0dd\uc131\n outputs = model.generate(inputs, max_length=max_length, num_return_sequences=1, no_repeat_ngram_size=2, early_stopping=True)\n\n # \uacb0\uacfc \ub514\ucf54\ub529\n return tokenizer.decode(outputs[0], skip_special_tokens=True)<\/code><\/pre>\n4.3. \ubb38\uc7a5 \uc0dd\uc131\ud558\uae30<\/h3>\n
\uc774\uc81c \uc704\uc5d0\uc11c \uc815\uc758\ud55c \ud568\uc218\ub97c \uc0ac\uc6a9\ud558\uc5ec \ud2b9\uc815 \ud504\ub86c\ud504\ud2b8\ub97c \uc785\ub825\uc73c\ub85c \ubb38\uc7a5\uc744 \uc0dd\uc131\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
prompt = \"\ub525\ub7ec\ub2dd\uc758 \ubbf8\ub798\ub294\"\ngenerated_text = generate_text(prompt)\n\nprint(f\"Generated Text: {generated_text}\")<\/code><\/pre>\n\uc704 \ucf54\ub4dc\ub97c \uc2e4\ud589\ud558\uba74 “\ub525\ub7ec\ub2dd\uc758 \ubbf8\ub798\ub294”\uc774\ub77c\ub294 \ud504\ub86c\ud504\ud2b8\uc5d0 \uc774\uc5b4\uc9c0\ub294 \ubb38\uc7a5\uc774 \uc0dd\uc131\ub420 \uac83\uc785\ub2c8\ub2e4. \uc774\ub294 DistilGPT2\uac00 \uc5b8\uc5b4 \ubaa8\ub378\ub85c\uc11c \uc8fc\uc5b4\uc9c4 \ubb38\ub9e5\uc5d0 \ub9de\ucd94\uc5b4 \uc790\uc5f0\uc2a4\ub7fd\uace0 \ucc3d\uc758\uc801\uc778 \ub0b4\uc6a9\uc744 \uc0dd\uc131\ud588\uc74c\uc744 \ubcf4\uc5ec\uc90d\ub2c8\ub2e4.<\/p>\n<\/section>\n\n5. \ub2e4\uc591\ud55c \uc0dd\uc131 \uc635\uc158 \uc870\uc815\ud558\uae30<\/h2>\n
\ubb38\uc7a5 \uc0dd\uc131 \uc2dc \uc5ec\ub7ec \uac00\uc9c0 \uc635\uc158\uc744 \uc870\uc815\ud558\uc5ec \ub2e4\uc591\ud55c \uc2a4\ud0c0\uc77c\uacfc \ub0b4\uc6a9\uc744 \ub9cc\ub4e4\uc5b4 \ubcfc \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc5ec\uae30\uc5d0\uc11c\ub294 \uba87 \uac00\uc9c0 \uc8fc\uc694 \uc635\uc158\uc5d0 \ub300\ud574 \uc54c\uc544\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
5.1. max_length<\/h3>\n
\uc0dd\uc131\ud560 \ucd5c\ub300 \ubb38\uc7a5 \uae38\uc774\ub97c \uc124\uc815\ud569\ub2c8\ub2e4. \uc774 \uac12\uc744 \uc870\uc815\ud558\uc5ec \uae34 \ubb38\uc7a5 \ub610\ub294 \uc9e7\uc740 \ubb38\uc7a5\uc744 \uc0dd\uc131\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
5.2. temperature<\/h3>\n
\uc628\ub3c4 \ub9e4\uac1c\ubcc0\uc218\ub294 \uc0dd\uc131\ub418\ub294 \ud14d\uc2a4\ud2b8\uc758 \ucc3d\uc758\uc131\uc5d0 \uc601\ud5a5\uc744 \ubbf8\uce69\ub2c8\ub2e4. \uac12\uc774 \ub0ae\uc744\uc218\ub85d \ub354 \ubcf4\uc218\uc801\uc778 \ubb38\uc7a5\uc744 \uc0dd\uc131\ud558\uace0, \uac12\uc774 \ub192\uc744\uc218\ub85d \ub354 \ub2e4\uc591\ud558\uace0 \ucc3d\uc758\uc801\uc778 \ubb38\uc7a5\uc744 \uc0dd\uc131\ud569\ub2c8\ub2e4. \uc774\ub97c \ucee4\uc2a4\ud130\ub9c8\uc774\uc9d5\ud558\uae30 \uc704\ud574\uc11c\ub294 generate \ud568\uc218\uc758 input\uc5d0 temperature \ub9e4\uac1c\ubcc0\uc218\ub97c \ucd94\uac00\ud558\uba74 \ub429\ub2c8\ub2e4.<\/p>\n
outputs = model.generate(inputs, max_length=max_length, temperature=0.8)<\/code><\/pre>\n5.3. num_return_sequences<\/h3>\n
\uc774 \ub9e4\uac1c\ubcc0\uc218\ub294 \uc0dd\uc131\ud560 \ubb38\uc7a5\uc758 \uc218\ub97c \uacb0\uc815\ud569\ub2c8\ub2e4. \uc5ec\ub7ec \ubb38\uc7a5\uc744 \ub3d9\uc2dc\uc5d0 \uc0dd\uc131\ud558\uc5ec \ube44\uad50\ud574\ubcfc \uc218 \uc788\uac8c \ud569\ub2c8\ub2e4.<\/p>\n
outputs = model.generate(inputs, num_return_sequences=5)<\/code><\/pre>\n<\/section>\n\n6. \uc2e4\uc81c \uc751\uc6a9 \uc0ac\ub840<\/h2>\n
DistilGPT2\uc640 \uac19\uc740 \uc790\uc5f0\uc5b4 \uc0dd\uc131 \ubaa8\ub378\uc740 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0 \uc751\uc6a9\ub420 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4:<\/p>\n
\n- \ube14\ub85c\uadf8 \uae00\uc4f0\uae30: \ube14\ub85c\uadf8 \uae00\uc758 \ucd08\uc548\uc744 \uc791\uc131\ud558\ub294 \ub370 \ub3c4\uc6c0\uc774 \ub429\ub2c8\ub2e4.<\/li>\n
- \ucc57\ubd07 \uac1c\ubc1c: \uc790\uc5f0\uc2a4\ub7ec\uc6b4 \ub300\ud654\ub97c \uac00\ub2a5\ud558\uac8c \ud558\ub294 \ucc57\ubd07\uc744 \ub9cc\ub4e4 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/li>\n
- \uc2a4\ud1a0\ub9ac \uc791\uc131: \uc774\uc57c\uae30\ub098 \uc18c\uc124\uc758 \ud50c\ub86f\uc744 \uc0dd\uc131\ud558\uba70 \ucc3d\uc758\uc801 \uae00\uc4f0\uae30\ub97c \uc9c0\uc6d0\ud569\ub2c8\ub2e4.<\/li>\n<\/ul>\n<\/section>\n\n
7. \uacb0\ub860<\/h2>\n
\uc774\ubc88 \uac15\uc88c\ub97c \ud1b5\ud574 \ud5c8\uae45\ud398\uc774\uc2a4\uc758 DistilGPT2 \ubaa8\ub378\uc744 \ud65c\uc6a9\ud558\uc5ec \ubb38\uc7a5\uc744 \uc0dd\uc131\ud558\ub294 \ubc29\ubc95\uc744 \ubc30\uc6e0\uc2b5\ub2c8\ub2e4. \uc790\uc5f0\uc5b4 \ucc98\ub9ac \uae30\uc220\uc758 \ubc1c\uc804\uc73c\ub85c \ub354 \ub9ce\uc740 \uc0ac\ub78c\ub4e4\uc774 \ud14d\uc2a4\ud2b8 \uc0dd\uc131\uc758 \uc774\uc810\uc744 \ub204\ub9b4 \uc218 \uc788\uac8c \ub418\uc5c8\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \uae30\uc220\ub4e4\uc774 \uc55e\uc73c\ub85c\ub3c4 \ub354\uc6b1 \ub9ce\uc740 \ubd84\uc57c\uc5d0\uc11c \ud65c\uc6a9\ub418\uae30\ub97c \uae30\ub300\ud569\ub2c8\ub2e4.<\/p>\n<\/section>\n