{"id":29674,"date":"2024-10-28T01:56:02","date_gmt":"2024-10-28T01:56:02","guid":{"rendered":"http:\/\/atmokpo.com\/w\/?p=29674"},"modified":"2024-11-26T06:51:45","modified_gmt":"2024-11-26T06:51:45","slug":"%ed%97%88%ea%b9%85%ed%8e%98%ec%9d%b4%ec%8a%a4-%ed%8a%b8%eb%a0%8c%ec%8a%a4%ed%8f%ac%eb%a8%b8-%ed%99%9c%ec%9a%a9%ea%b0%95%ec%a2%8c-%ec%86%8c%ec%8a%a4-%ec%96%b8%ec%96%b4-m2m100-%ed%86%a0%ed%81%ac","status":"publish","type":"post","link":"https:\/\/atmokpo.com\/w\/29674\/","title":{"rendered":"\ud5c8\uae45\ud398\uc774\uc2a4 \ud2b8\ub80c\uc2a4\ud3ec\uba38 \ud65c\uc6a9\uac15\uc88c, \uc18c\uc2a4 \uc5b8\uc5b4 M2M100 \ud1a0\ud06c\ub098\uc774\uc9d5"},"content":{"rendered":"

<\/p>\n

\ub525\ub7ec\ub2dd\uc758 \ubc1c\uc804\uacfc \ud568\uaed8 \uc790\uc5f0\uc5b4 \ucc98\ub9ac(NLP)\ub294 \ud070 \ubcc0\ud654\ub97c \uacaa\uc5c8\uc2b5\ub2c8\ub2e4. \ud2b9\ud788, \ud5c8\uae45\ud398\uc774\uc2a4(Hugging Face)\uc758 \ud2b8\ub79c\uc2a4\ud3ec\uba38(transformers) \ub77c\uc774\ube0c\ub7ec\ub9ac\ub294 NLP \uc791\uc5c5\uc744 \uc704\ud55c \uac15\ub825\ud55c \ub3c4\uad6c\ub85c \uc790\ub9ac \uc7a1\uc558\uc2b5\ub2c8\ub2e4. \uc774 \uac15\uc88c\uc5d0\uc11c\ub294 M2M100 \ubaa8\ub378\uc744 \ud65c\uc6a9\ud55c \ub2e4\uad6d\uc5b4 \ubc88\uc5ed \ubc0f \ud1a0\ud06c\ub098\uc774\uc9d5 \uacfc\uc815\uc744 \uc18c\uac1c\ud569\ub2c8\ub2e4.<\/p>\n

M2M100 \ubaa8\ub378 \uac1c\uc694<\/h2>\n

M2M100(Multilingual To Multilingual Translation) \ubaa8\ub378\uc740 100\uac1c \uc774\uc0c1\uc758 \uc5b8\uc5b4 \uac04\uc758 \uc9c1\uc811 \ubc88\uc5ed\uc744 \uc9c0\uc6d0\ud558\ub294 \ub2e4\uad6d\uc5b4 \ubaa8\ub378\uc785\ub2c8\ub2e4. \uae30\uc874\uc758 \ubc88\uc5ed \ubaa8\ub378\uc740 \uc18c\uc2a4 \uc5b8\uc5b4\uc5d0\uc11c \uc911\uac04 \uc5b8\uc5b4(\uc608: \uc601\uc5b4)\ub85c \ubc88\uc5ed\ud55c \ud6c4 \ub300\uc0c1 \uc5b8\uc5b4\ub85c \ubcc0\ud658\ud558\ub294 \uac04\uc811 \ubc88\uc5ed \ubc29\uc2dd\uc744 \uc0ac\uc6a9\ud588\uc2b5\ub2c8\ub2e4. M2M100\uc740 \uc774\ub7ec\ud55c \ud55c\uacc4\ub97c \uadf9\ubcf5\ud558\uae30 \uc704\ud574 \ub2e4\uad6d\uc5b4 \uac04 \ubcc0\ud658\uc744 \uac00\ub2a5\ud558\uac8c \ud558\uc5ec \ub2e4\uc591\ud55c \uc5b8\uc5b4 \uc30d \uac04\uc758 \ubc88\uc5ed \ud6a8\uc728\uc131\uc744 \ud06c\uac8c \ud5a5\uc0c1\uc2dc\ud0b5\ub2c8\ub2e4.<\/p>\n

\ud1a0\ud06c\ub098\uc774\uc9d5\uc774\ub780?<\/h2>\n

\ud1a0\ud06c\ub098\uc774\uc9d5(tokenization)\uc740 \uc785\ub825\ub41c \ud14d\uc2a4\ud2b8\ub97c \ub354 \uc791\uc740 \ub2e8\uc704\uc778 \ud1a0\ud070(token)\uc73c\ub85c \ub098\ub204\ub294 \uacfc\uc815\uc785\ub2c8\ub2e4. \ub9ac\uc2a4\ud2b8 \ud615\uc2dd\uc73c\ub85c \ubcc0\ud658\ud55c \ud6c4, \uac01\uac01\uc758 \ud1a0\ud070\uc5d0 \ub300\ud574 \uace0\uc720\ud55c \uc778\ub371\uc2a4\ub97c \ubd80\uc5ec\ud558\uac8c \ub429\ub2c8\ub2e4. \ud1a0\ud06c\ub098\uc774\uc9d5\uc740 NLP\uc5d0\uc11c\ub294 \ud544\uc218\uc801\uc778 \uacfc\uc815\uc73c\ub85c, \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\ub97c \ubaa8\ub378\uc5d0 \uc785\ub825\ud558\uae30 \uc804\uc5d0 \ud544\uc694\ud569\ub2c8\ub2e4.<\/p>\n

\ud658\uacbd \uc124\uc815<\/h2>\n

\uac15\uc88c\ub97c \uc9c4\ud589\ud558\uae30 \uc804\uc5d0 \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud574\uc57c \ud569\ub2c8\ub2e4. \ud2b9\ud788 transformers<\/code>\uc640 torch<\/code>\ub97c \uc124\uce58\ud560 \uac83\uc785\ub2c8\ub2e4. \uc544\ub798\uc758 \uba85\ub839\uc5b4\ub85c \uc124\uce58\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n

\n        pip install transformers torch<\/code>\n    <\/pre>\n
\ud1a0\ud06c\ub098\uc774\uc800 \ubd88\ub7ec\uc624\uae30<\/h2>\n
M2M100 \ubaa8\ub378\uc5d0 \ub300\ud55c \ud1a0\ud06c\ub098\uc774\uc800\ub97c \ubd88\ub7ec\uc624\uae30 \uc704\ud574 transformers<\/code> \ub77c\uc774\ube0c\ub7ec\ub9ac\uc5d0\uc11c \uc81c\uacf5\ud558\ub294 M2M100Tokenizer<\/code> \ud074\ub798\uc2a4\ub97c \uc0ac\uc6a9\ud560 \uac83\uc785\ub2c8\ub2e4.<\/p>\n
\n        \nimport torch\nfrom transformers import M2M100Tokenizer\n\n# M2M100 \ubaa8\ub378\uc758 \ud1a0\ud06c\ub098\uc774\uc800 \ubd88\ub7ec\uc624\uae30\ntokenizer = M2M100Tokenizer.from_pretrained(\"facebook\/m2m100_418M\")\n        <\/code>\n    <\/pre>\n
\ud1a0\ud06c\ub098\uc774\uc9d5 \uacfc\uc815<\/h2>\n
\uc774\uc81c \uc6b0\ub9ac\ub294 \ud14d\uc2a4\ud2b8\ub97c \ud1a0\ud06c\ub098\uc774\uc9d5\ud560 \uc900\ube44\uac00 \ub418\uc5c8\uc2b5\ub2c8\ub2e4. \uc544\ub798\ub294 ‘\uc548\ub155\ud558\uc138\uc694, \uc5ec\ub7ec\ubd84!’\uc774\ub77c\ub294 \ubb38\uc7a5\uc744 \ud1a0\ud06c\ub098\uc774\uc9d5\ud558\ub294 \uc608\uc2dc\uc785\ub2c8\ub2e4.<\/p>\n
\n        \n# \uc785\ub825 \ud14d\uc2a4\ud2b8\ntext = \"\uc548\ub155\ud558\uc138\uc694, \uc5ec\ub7ec\ubd84!\"\n\n# \ud14d\uc2a4\ud2b8\ub97c \ud1a0\ud06c\ub098\uc774\uc9d5\nencoded_input = tokenizer(text, return_tensors=\"pt\")\n\n# \ud1a0\ud070\uacfc \uc778\ub371\uc2a4 \ucd9c\ub825\nprint(\"\ud1a0\ud06c\ub098\uc774\uc988\ub41c \ud1a0\ud070:\", tokenizer.convert_ids_to_tokens(encoded_input['input_ids'][0]))\nprint(\"\ud1a0\ud070 \uc778\ub371\uc2a4:\", encoded_input['input_ids'])\n        <\/code>\n    <\/pre>\n
\ud1a0\ud06c\ub098\uc774\uc9d5 \uacb0\uacfc<\/h3>\n
\uc704 \ucf54\ub4dc \uc2e4\ud589 \ud6c4 \ucd9c\ub825\ub418\ub294 \uacb0\uacfc\ub294 \uc785\ub825\ub41c \ud14d\uc2a4\ud2b8\uac00 \uc5b4\ub5bb\uac8c \ud1a0\ud06c\ub098\uc774\uc9d5\ub418\uace0 \uc778\ub371\uc2a4\uac00 \ub9e4\uaca8\uc84c\ub294\uc9c0 \ubcf4\uc5ec\uc90d\ub2c8\ub2e4. convert_ids_to_tokens<\/code> \uba54\uc11c\ub4dc\ub97c \ud1b5\ud574 \ud1a0\ud070\uc758 \uc2e4\uc81c \uac12\uc744 \ud655\uc778\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\ub2e4\uad6d\uc5b4 \ubc88\uc5ed<\/h2>\n
\ud1a0\ud06c\ub098\uc774\uc9d5\ub41c \ub370\uc774\ud130\ub97c \uc0ac\uc6a9\ud558\uc5ec \ub2e4\uad6d\uc5b4 \ubc88\uc5ed\uc744 \uc218\ud589\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. M2M100 \ubaa8\ub378\uc744 \ud65c\uc6a9\ud558\uc5ec \ud55c\uad6d\uc5b4\ub97c \uc601\uc5b4\ub85c \ubc88\uc5ed\ud558\ub294 \uc608\uc2dc\ub97c \ubcf4\uc5ec\ub4dc\ub9ac\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
\n        \nfrom transformers import M2M100ForConditionalGeneration\n\n# M2M100 \ubaa8\ub378 \ubd88\ub7ec\uc624\uae30\nmodel = M2M100ForConditionalGeneration.from_pretrained(\"facebook\/m2m100_418M\")\n\n# \ud55c\uad6d\uc5b4 \ud14d\uc2a4\ud2b8\ntext = \"\uc548\ub155\ud558\uc138\uc694, \uc5ec\ub7ec\ubd84!\"\nencoded_input = tokenizer(text, return_tensors=\"pt\")\n\n# \ubc88\uc5ed\ntranslated_tokens = model.generate(**encoded_input, forced_bos_token_id=tokenizer.get_lang_id(\"en\"))\n\n# \ubc88\uc5ed \uacb0\uacfc\ntranslated_text = tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)\nprint(\"\ubc88\uc5ed \uacb0\uacfc:\", translated_text[0])\n        <\/code>\n    <\/pre>\n
\ubc88\uc5ed \uacb0\uacfc \ud574\uc11d<\/h3>\n
\uc704 \ucf54\ub4dc\ub97c \ud1b5\ud574 \ud55c\uad6d\uc5b4 \ubb38\uc7a5\uc774 \uc601\uc5b4\ub85c \uc798 \ubc88\uc5ed\ub410\ub294\uc9c0 \ud655\uc778\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. generate<\/code> \uba54\uc11c\ub4dc\ub294 \uc785\ub825\ub41c \ud1a0\ud070 \ub370\uc774\ud130\ub97c \uae30\ubc18\uc73c\ub85c \ubc88\uc5ed\ub41c \uacb0\uacfc\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4.<\/p>\n
\uacb0\ub860<\/h2>\n
\uc774\ubc88 \uac15\uc88c\uc5d0\uc11c\ub294 \ud5c8\uae45\ud398\uc774\uc2a4\uc758 M2M100 \ubaa8\ub378\uc744 \ud65c\uc6a9\ud558\uc5ec \ub2e4\uad6d\uc5b4 \ud1a0\ud06c\ub098\uc774\uc9d5 \ubc0f \ubc88\uc5ed \uacfc\uc815\uc5d0 \ub300\ud574 \uc0b4\ud3b4\ubcf4\uc558\uc2b5\ub2c8\ub2e4. \uc790\uc5f0\uc5b4 \ucc98\ub9ac \ubd84\uc57c\uc758 \ubc1c\uc804\uc740 \uc55e\uc73c\ub85c\ub3c4 \uacc4\uc18d\ub420 \uac83\uc774\uba70, \uc774\ub7ec\ud55c \ub3c4\uad6c\ub97c \ud65c\uc6a9\ud558\uc5ec \ubcf4\ub2e4 \ub2e4\uc591\ud55c \uc5b8\uc5b4 \uac04\uc758 \uc18c\ud1b5\uc774 \uac00\ub2a5\ud574\uc9c8 \uac83\uc785\ub2c8\ub2e4. \uc55e\uc73c\ub85c\ub3c4 NLP\uc640 \ub525\ub7ec\ub2dd\uc5d0 \ub300\ud55c \ub9ce\uc740 \uad00\uc2ec\uacfc \uc5f0\uad6c\uac00 \uc9c0\uc18d\ub418\uae30\ub97c \ubc14\ub78d\ub2c8\ub2e4.<\/p>\n
\ucc38\uace0 \uc790\ub8cc<\/h2>\n