\n \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \ud5c8\uae45\ud398\uc774\uc2a4\uc758 \ud2b8\ub79c\uc2a4\ud3ec\uba38 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud65c\uc6a9\ud558\uc5ec M2M100 \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud574 \ud55c\uad6d\uc5b4\uc640 \uc601\uc5b4 \uac04\uc758 \uc790\ub3d9 \ubc88\uc5ed \uc791\uc5c5\uc744 \uc218\ud589\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc790\uc138\ud788 \uc124\uba85\ud569\ub2c8\ub2e4. M2M100\uc740 \ub2e4\uad6d\uc5b4 \ubc88\uc5ed\uc744 \uc9c0\uc6d0\ud558\ub294 \ubaa8\ub378\ub85c, 100\uac1c \uc774\uc0c1\uc758 \uc5b8\uc5b4 \uac04\uc758 \uc0c1\ud638 \ubc88\uc5ed\uc744 \uc9c0\uc6d0\ud569\ub2c8\ub2e4. \uc774 \uae00\uc5d0\uc11c\ub294 M2M100\uc758 \uac1c\uc694, \uc124\uce58 \ubc29\ubc95, \ub370\uc774\ud130 \uc900\ube44, \ubaa8\ub378 \ub85c\ub4dc \ubc0f \uc608\uce21\uc744 \uc218\ud589\ud558\ub294 \uacfc\uc815\uc744 \ub2e8\uacc4\ubcc4\ub85c \uc124\uba85\ud569\ub2c8\ub2e4.\n <\/p>\n
\n M2M100 (Multilingual to Multilingual) \ubaa8\ub378\uc740 Facebook AI Research\uc5d0\uc11c \uac1c\ubc1c\ud55c \ub2e4\uad6d\uc5b4 \uae30\uacc4 \ubc88\uc5ed \ubaa8\ub378\ub85c, 100\uac1c \uc774\uc0c1\uc758 \uc5b8\uc5b4\ub85c \uc11c\ub85c \ubc88\uc5ed\ud560 \uc218 \uc788\ub294 \uae30\ub2a5\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4. M2M100\uc758 \uac00\uc7a5 \ud070 \uc7a5\uc810\uc740 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4:\n <\/p>\n
\n M2M100 \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud558\uae30 \uc704\ud574 \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\uc640 \ud328\ud0a4\uc9c0\ub97c \uc124\uce58\ud558\ub294 \ubc29\ubc95\uc785\ub2c8\ub2e4. \ub2e4\uc74c \ub2e8\uacc4\ub97c \ub530\ub77c \ud658\uacbd\uc744 \uc124\uc815\ud569\ub2c8\ub2e4.\n <\/p>\n
pip install transformers torch<\/code><\/pre>\n\n \uc704 \uba85\ub839\uc5b4\ub97c \ud1b5\ud574 \ud5c8\uae45\ud398\uc774\uc2a4\uc758 \ud2b8\ub79c\uc2a4\ud3ec\uba38 \ub77c\uc774\ube0c\ub7ec\ub9ac\uc640 PyTorch\ub97c \uc124\uce58\ud569\ub2c8\ub2e4. \uc5ec\uae30\uc11c\ub294 PyTorch\ub97c \uae30\ubcf8\uc73c\ub85c \uc0ac\uc6a9\ud558\uaca0\uc2b5\ub2c8\ub2e4. \uc124\uce58\uac00 \uc644\ub8cc\ub418\uba74 \ub2e4\uc74c \ucf54\ub4dc\ub85c M2M100 \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud560 \uc900\ube44\uac00 \uc644\ub8cc\ub418\uc5c8\uc2b5\ub2c8\ub2e4.\n <\/p>\n
3. M2M100 \ubaa8\ub378 \ub85c\ub4dc<\/h2>\n
\n M2M100 \ubaa8\ub378\uc744 \ub85c\ub4dc\ud558\uae30 \uc704\ud574 \ub2e4\uc74c\uacfc \uac19\uc740 \ucf54\ub4dc snippets\ub97c \uc791\uc131\ud569\ub2c8\ub2e4.\n <\/p>\n
\nfrom transformers import M2M100ForConditionalGeneration, M2M100Tokenizer\n\n# \ubaa8\ub378\uacfc \ud1a0\ud06c\ub098\uc774\uc800 \ub85c\ub4dc\nmodel_name = \"facebook\/m2m100_418M\"\ntokenizer = M2M100Tokenizer.from_pretrained(model_name)\nmodel = M2M100ForConditionalGeneration.from_pretrained(model_name)\n <\/code><\/pre>\n\n \uc704 \ucf54\ub4dc\ub294 \ud544\uc694\ud55c \ubaa8\ub378\uacfc \ud1a0\ud06c\ub098\uc774\uc800\ub97c \ud5c8\uae45\ud398\uc774\uc2a4\uc758 \ud5c8\ube0c\uc5d0\uc11c \ub85c\ub4dc\ud569\ub2c8\ub2e4.\n <\/p>\n
4. \ub370\uc774\ud130 \uc900\ube44<\/h2>\n
\n \ubc88\uc5ed\ud560 \ubb38\uc7a5\uc744 \uc900\ube44\ud569\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \uc601\uc5b4 \ubb38\uc7a5\uc744 \ud55c\uad6d\uc5b4\ub85c \ubc88\uc5ed\ud558\uac70\ub098 \ud55c\uad6d\uc5b4 \ubb38\uc7a5\uc744 \uc601\uc5b4\ub85c \ubc88\uc5ed\ud558\ub294 \uacbd\uc6b0\ub97c \uace0\ub824\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc544\ub798 \ucf54\ub4dc\ub294 \ubc88\uc5ed\ud560 \ubb38\uc7a5\uc744 \uc900\ube44\ud558\ub294 \uc608\uc2dc\uc785\ub2c8\ub2e4.\n <\/p>\n
\n# \ubc88\uc5ed\ud560 \ubb38\uc7a5\ntext_to_translate = \"\ub525\ub7ec\ub2dd\uc740 \uc778\uacf5\uc9c0\ub2a5\uc758 \ud55c \ubd84\uc57c\ub85c, \ucef4\ud4e8\ud130\uac00 \ub370\uc774\ud130\ub97c \ud1b5\ud574 \ud559\uc2b5\ud558\ub3c4\ub85d \ud558\ub294 \uae30\uc220\uc785\ub2c8\ub2e4.\"\n <\/code><\/pre>\n\n \uc774\uc81c \uc774 \ubb38\uc7a5\uc744 M2M100 \ubaa8\ub378\uc744 \ud1b5\ud574 \ubc88\uc5ed\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.\n <\/p>\n
5. \ubc88\uc5ed \uc218\ud589<\/h2>\n
\n \ubaa8\ub378\uc544\uc5d0\uac8c \uc785\ub825\ub41c \ubb38\uc7a5\uc744 \uae30\ubc18\uc73c\ub85c \ubc88\uc5ed\uc744 \uc218\ud589\ud569\ub2c8\ub2e4. \ubaa8\ub378\uc758 \uc785\ub825\uc744 \uc900\ube44\ud55c \ud6c4, \ud1a0\ud070\ud654\ub97c \uc218\ud589\ud558\uace0 \ubaa8\ub378\uc744 \ud1b5\ud574 \uc608\uce21\uc744 \uc9c4\ud589\ud558\ub294 \uacfc\uc815\uc785\ub2c8\ub2e4.\n <\/p>\n
\n# \ud1a0\ud070\ud654\ntokenized_input = tokenizer(text_to_translate, return_tensors=\"pt\", padding=True)\n\n# \ubc88\uc5ed \uc218\ud589\nwith torch.no_grad():\n generated_ids = model.generate(\n tokenized_input['input_ids'], \n forced_bos_token_id=tokenizer.get_lang_id('ko') # \ud55c\uad6d\uc5b4\ub85c \ubc88\uc5ed\n )\n\n# \ub514\ucf54\ub529\ntranslated_text = tokenizer.decode(generated_ids[0], skip_special_tokens=True)\nprint(f\"\ubc88\uc5ed \uacb0\uacfc: {translated_text}\")\n <\/code><\/pre>\n\n \uc704 \ucf54\ub4dc\ub97c \uc2e4\ud589\ud558\uba74 \ub2e4\uc74c\uacfc \uac19\uc740 \uacb0\uacfc\ub97c \ud655\uc778\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n
6. \ubc88\uc5ed \uacb0\uacfc \ud655\uc778<\/h2>\n
\n \ubc88\uc5ed \uacb0\uacfc\ub294 \ubaa8\ub378\uc774 \uc0dd\uc131\ud55c \ud55c\uad6d\uc5b4 \ubb38\uc7a5\uc785\ub2c8\ub2e4. M2M100 \ubaa8\ub378\uc758 \uc131\ub2a5\uc740 \ubc88\uc5ed\ud560 \ubb38\uc7a5\uc5d0 \ub530\ub77c \ub2e4\uc18c \ucc28\uc774\uac00 \uc788\uc744 \uc218 \uc788\uc9c0\ub9cc, \uc804\ubc18\uc801\uc73c\ub85c \ub192\uc740 \ud488\uc9c8\uc758 \ubc88\uc5ed\uc744 \uc81c\uacf5\ud558\ub294 \ud3b8\uc785\ub2c8\ub2e4.\n <\/p>\n
7. \uc5ec\ub7ec \ubb38\uc7a5 \ubc88\uc5ed\ud558\uae30<\/h2>\n
\n M2M100 \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc5ec\ub7ec \ubb38\uc7a5\uc744 \ub3d9\uc2dc\uc5d0 \ubc88\uc5ed\ud558\ub294 \uac83\uc740 \ub9e4\uc6b0 \uac04\ud3b8\ud569\ub2c8\ub2e4. \ub2e4\uc74c\uacfc \uac19\uc774 \uc5ec\ub7ec \uac1c\uc758 \ubb38\uc7a5\uc744 \uc900\ube44\ud55c \ub2e4\uc74c \ubc18\ubcf5\ubb38\uc744 \ud1b5\ud574 \ubc88\uc5ed\ud558\uba74 \ub429\ub2c8\ub2e4.\n <\/p>\n
\n# \uc5ec\ub7ec \ubb38\uc7a5 \ubc88\uc5ed\nsentences_to_translate = [\n \"AI\ub294 21\uc138\uae30 \ucd5c\uace0\uc758 \uae30\uc220 \uc911 \ud558\ub098\ub85c \uc790\ub9ac\uc7a1\uace0 \uc788\uc2b5\ub2c8\ub2e4.\",\n \"\ud2b8\ub79c\uc2a4\ud3ec\uba38 \uc544\ud0a4\ud14d\ucc98\ub294 \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc758 \ud601\uc2e0\uc744 \uac00\uc838\uc654\uc2b5\ub2c8\ub2e4.\"\n]\n\nfor sentence in sentences_to_translate:\n input_ids = tokenizer(sentence, return_tensors=\"pt\", padding=True)['input_ids']\n with torch.no_grad():\n output_ids = model.generate(input_ids, forced_bos_token_id=tokenizer.get_lang_id('ko'))\n translated_sentence = tokenizer.decode(output_ids[0], skip_special_tokens=True)\n print(f\"{sentence} -> {translated_sentence}\")\n <\/code><\/pre>\n\n \uc774\ub7ec\ud55c \ubc29\uc2dd\uc73c\ub85c \uc5ec\ub7ec \ubb38\uc7a5\uc744 \uc190\uc27d\uac8c \ubc88\uc5ed\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n
8. \uc131\ub2a5 \ud3c9\uac00<\/h2>\n
\n \uc790\ub3d9 \ubc88\uc5ed\uc758 \uc131\ub2a5\uc744 \ud3c9\uac00\ud558\ub294 \ubc29\ubc95\uc5d0\ub294 BleU score, METEOR score \ub4f1\uc774 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ud3c9\uac00 \ubc29\ubc95\uc744 \uc0ac\uc6a9\ud558\uc5ec \ubaa8\ub378\uc758 \ubc88\uc5ed \ud488\uc9c8\uc744 \uc815\ub7c9\uc801\uc73c\ub85c \ud3c9\uac00\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ud604\uc7ac\uae4c\uc9c0\ub294 \uc790\uc5f0\uc5b4 \ucc98\ub9ac \ubd84\uc57c\uc5d0\uc11c\uc758 \uc77c\ubc18\uc801\uc778 \ud3c9\uac00 \ubc29\ubc95\uc5d0 \ub300\ud574 \ub17c\uc758\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4.\n <\/p>\n
9. \ub9c8\ubb34\ub9ac<\/h2>\n
\n \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \ud5c8\uae45\ud398\uc774\uc2a4\uc758 \ud2b8\ub79c\uc2a4\ud3ec\uba38 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud65c\uc6a9\ud558\uc5ec M2M100 \ubaa8\ub378\uc744 \ud1b5\ud55c \uc790\ub3d9 \ubc88\uc5ed\uc744 \uc218\ud589\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \ub2e4\ub918\uc2b5\ub2c8\ub2e4. \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \ud6a8\uc728\uc801\uc73c\ub85c \ud65c\uc6a9\ud558\uc5ec \ub2e4\uc591\ud55c \uc790\uc5f0\uc5b4 \ucc98\ub9ac \uc751\uc6a9 \ud504\ub85c\uadf8\ub7a8\uc744 \uad6c\ucd95\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc55e\uc73c\ub85c\ub3c4 \ub2e4\uc591\ud55c \ub525\ub7ec\ub2dd \ubaa8\ub378\uacfc \uae30\uc220\uc744 \ud1b5\ud574 \ub354\uc6b1 \ubc1c\uc804\ud560 \uc218 \uc788\uae30\ub97c \uae30\ub300\ud569\ub2c8\ub2e4.\n <\/p>\n
\ucc38\uace0 \uc790\ub8cc<\/h2>\n