{"id":29582,"date":"2024-10-28T01:55:34","date_gmt":"2024-10-28T01:55:34","guid":{"rendered":"http:\/\/atmokpo.com\/w\/?p=29582"},"modified":"2024-11-26T06:52:11","modified_gmt":"2024-11-26T06:52:11","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-dialogpt-%ec%9e%91%eb%ac%b8","status":"publish","type":"post","link":"https:\/\/atmokpo.com\/w\/29582\/","title":{"rendered":"\ud5c8\uae45\ud398\uc774\uc2a4 \ud2b8\ub80c\uc2a4\ud3ec\uba38 \ud65c\uc6a9\uac15\uc88c, DialoGPT \uc791\ubb38"},"content":{"rendered":"

<\/p>\n

\uc778\uacf5\uc9c0\ub2a5\uc758 \ubc1c\uc804\uacfc \ud568\uaed8, \uc790\uc5f0\uc5b4 \ucc98\ub9ac(NLP) \ubd84\uc57c\uc5d0\uc11c\ub3c4 \ud070 \ud601\uc2e0\uc774 \uc788\uc5c8\uc2b5\ub2c8\ub2e4. \ud2b9\ud788, \ub525\ub7ec\ub2dd \uae30\ubc18\uc758 \ub300\ud654\ud615 \ubaa8\ub378\ub4e4\uc774 \uc8fc\ubaa9\ubc1b\uace0 \uc788\ub294\ub370, \uadf8 \uc911\uc5d0\uc11c\ub3c4 DialoGPT<\/strong>\ub294 \ub9e4\uc6b0 \uc778\uae30 \uc788\ub294 \ubaa8\ub378\uc785\ub2c8\ub2e4. \uc774\ubc88 \uac15\uc88c\uc5d0\uc11c\ub294 DialoGPT\uc758 \uac1c\ub150, \ud65c\uc6a9 \ubc29\ubc95, \uadf8\ub9ac\uace0 \ud30c\uc774\uc36c\uc744 \ud1b5\ud55c \uad6c\ud604 \uc608\uc81c\ub97c \uc790\uc138\ud788 \ub2e4\ub904\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n

1. DialoGPT\ub780?<\/h2>\n

DialoGPT(Conversational Generative Pre-trained Transformer)\ub294 OpenAI\uac00 \ubc1c\ud45c\ud55c GPT-2 \ubaa8\ub378\uc744 \uae30\ubc18\uc73c\ub85c \ud55c \ub300\ud654\ud615 \ubaa8\ub378\uc785\ub2c8\ub2e4. DialoGPT\ub294 \uc0ac\ub78c\uacfc\uc758 \ub300\ud654\uc5d0 \uc801\ud569\ud558\ub3c4\ub85d \ud6c8\ub828\ub418\uc5c8\uc73c\uba70, \ub370\uc774\ud130\uc14b\uc5d0\ub294 \uc778\ud130\ub137\uc5d0\uc11c \uc218\uc9d1\ub41c \ub300\ud654 \ub85c\uadf8\uac00 \ud3ec\ud568\ub418\uc5b4 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub294 \ubaa8\ub378\uc774 \uc774\uc804 \ub300\ud654\uc758 \ub9e5\ub77d\uc744 \uace0\ub824\ud558\uc5ec \uc751\ub2f5\uc744 \uc0dd\uc131\ud560 \uc218 \uc788\ub3c4\ub85d \ud559\uc2b5\ud558\uac8c \ud569\ub2c8\ub2e4.<\/p>\n

2. \ud5c8\uae45\ud398\uc774\uc2a4(Hugging Face)\uc640 Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac<\/h2>\n

\ud5c8\uae45\ud398\uc774\uc2a4\ub294 \uc790\uc5f0\uc5b4 \ucc98\ub9ac \ubd84\uc57c\uc5d0\uc11c \uac00\uc7a5 \ub110\ub9ac \uc0ac\uc6a9\ub418\ub294 \ub77c\uc774\ube0c\ub7ec\ub9ac \uc911 \ud558\ub098\ub85c, \ub2e4\uc591\ud55c \uc0ac\uc804 \ud6c8\ub828\ub41c \uc5b8\uc5b4 \ubaa8\ub378\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4. Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac<\/strong>\ub294 \uc774\ub7ec\ud55c \ubaa8\ub378\uc744 mudah\ud558\uac8c \uc0ac\uc6a9\ud560 \uc218 \uc788\ub3c4\ub85d \ub3c4\uc640\uc8fc\ub294 Python \ub77c\uc774\ube0c\ub7ec\ub9ac\uc785\ub2c8\ub2e4. \uc124\uce58\ub294 \ub2e4\uc74c\uc758 pip \uba85\ub839\uc5b4\ub85c \uc9c4\ud589\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n

pip install transformers<\/code><\/pre>\n
3. DialoGPT \uc124\uce58<\/h2>\n
DialoGPT\ub97c \uc0ac\uc6a9\ud558\uae30 \uc704\ud574\uc11c\ub294 Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud55c \ud6c4, \uc801\uc808\ud55c \ubaa8\ub378\uc744 \ub2e4\uc6b4\ub85c\ub4dc\ud574\uc57c \ud569\ub2c8\ub2e4. DialoGPT\ub294 \ub2e4\uc591\ud55c \uc0ac\uc774\uc988\uc758 \ubaa8\ub378\ub85c \uc81c\uacf5\ub418\uba70, small<\/code>, medium<\/code>, large<\/code> \ub4f1\uc774 \uc788\uc2b5\ub2c8\ub2e4. \uc544\ub798\ub294 medium<\/code> \ubaa8\ub378\uc744 \ud65c\uc6a9\ud558\ub294 \uc608\uc81c \ucf54\ub4dc\uc785\ub2c8\ub2e4:<\/p>\n
from transformers import AutoModelForCausalLM, AutoTokenizer\n\n# \ubaa8\ub378\uacfc \ud1a0\ud06c\ub098\uc774\uc800 \ub2e4\uc6b4\ub85c\ub4dc\ntokenizer = AutoTokenizer.from_pretrained(\"microsoft\/DialoGPT-medium\")\nmodel = AutoModelForCausalLM.from_pretrained(\"microsoft\/DialoGPT-medium\")<\/code><\/pre>\n
4. \ub300\ud654 \uae30\ub2a5 \uad6c\ud604\ud558\uae30<\/h2>\n
\uc774\uc81c \ubaa8\ub378\uacfc \ud1a0\ud06c\ub098\uc774\uc800\ub97c \ub2e4\uc6b4\ub85c\ub4dc \ud588\uc73c\ubbc0\ub85c, \ub300\ud654 \uc0dd\uc131 \uae30\ub2a5\uc744 \uad6c\ud604\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc0ac\uc6a9\uc790\uc758 \uc785\ub825\uc744 \ubc1b\uc544 \uc774 \uc785\ub825\uc744 \uae30\ubc18\uc73c\ub85c \ubaa8\ub378\uc774 \uc751\ub2f5\uc744 \uc0dd\uc131\ud558\ub3c4\ub85d \ud558\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
4.1 \ub300\ud654 \uc0dd\uc131 \ucf54\ub4dc<\/h3>\n
import torch\n\n# \ub300\ud654 \uae30\ub85d\uc744 \ucd08\uae30\ud654\nchat_history_ids = None\n\nwhile True:\n    # \uc0ac\uc6a9\uc790 \uc785\ub825 \ubc1b\uae30\n    user_input = input(\"User: \")\n\n    # \uc785\ub825\uc744 \ud14d\uc2a4\ud2b8\uc5d0\uc11c \ud1a0\ud070\uc73c\ub85c \ubcc0\ud658\n    new_user_input_ids = tokenizer.encode(user_input + tokenizer.eos_token, return_tensors='pt')\n\n    # \uc774\uc804 \ub300\ud654\uc640 \uc0c8\ub85c\uc6b4 \uc785\ub825\uc744 \ud569\uce58\uae30\n    if chat_history_ids is not None:\n        bot_input_ids = torch.cat([chat_history_ids, new_user_input_ids], dim=-1)\n    else:\n        bot_input_ids = new_user_input_ids\n\n    # \ubaa8\ub378\uc744 \ud1b5\ud55c \uc751\ub2f5 \uc0dd\uc131\n    chat_history_ids = model.generate(bot_input_ids, max_length=1000, pad_token_id=tokenizer.eos_token_id)\n\n    # \ubaa8\ub378\uc758 \uc751\ub2f5\uc744 \ud14d\uc2a4\ud2b8\ub85c \ub514\ucf54\ub529\n    bot_response = tokenizer.decode(chat_history_ids[:, bot_input_ids.shape[-1]:][0], skip_special_tokens=True)\n\n    # \uc751\ub2f5 \ucd9c\ub825\n    print(\"Bot: \", bot_response)<\/code><\/pre>\n
4.2 \ucf54\ub4dc \uc124\uba85<\/h3>\n