{"id":2485,"date":"2026-03-23T05:00:00","date_gmt":"2026-03-22T20:00:00","guid":{"rendered":"https:\/\/team-y5.com\/fukugyo-sensei\/?p=2485"},"modified":"2026-03-13T15:49:35","modified_gmt":"2026-03-13T06:49:35","slug":"ai-hugging-face-intro","status":"publish","type":"post","link":"https:\/\/team-y5.com\/fukugyo-sensei\/ai-hugging-face-intro\/","title":{"rendered":"\u3010AI\u624b\u5e33 No.76\u3011Hugging Face\u2500\u2500AI\u306eGitHub\u3067\u526f\u696d\u30b3\u30b9\u30c8\u3092\u30bc\u30ed\u306b\u3059\u308b\u6700\u5f37\u30d7\u30e9\u30c3\u30c8\u30d5\u30a9\u30fc\u30e0"},"content":{"rendered":"
\n
AI HANDBOOK — AI\u624b\u5e33 — No.76<\/span><\/p>\n

Hugging Face<\/span><\/h1>\n

AI\u30e2\u30c7\u30eb\u30fb\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u30fbSpace\u3092\u7121\u6599\u3067\u5171\u6709\u3067\u304d\u308b\u300cAI\u306eGitHub\u300d<\/span><\/p>\n

ML\u30d7\u30e9\u30c3\u30c8\u30d5\u30a9\u30fc\u30e0<\/span>
\n2016\u5e74\u5275\u696d<\/span>
\n\u7121\u6599\u301c$9\/\u6708\u301c<\/span><\/div>\n<\/div>\n\n\n\n\n\n\n\n
DEVELOPER<\/th>\nHugging Face, Inc.\uff08\u7c73\u56fd\u30fb\u30cb\u30e5\u30fc\u30e8\u30fc\u30af\uff09<\/td>\n<\/tr>\n
CATEGORY<\/th>\nAI\u30e2\u30c7\u30eb\u30cf\u30d6\uff0fML\u30d7\u30e9\u30c3\u30c8\u30d5\u30a9\u30fc\u30e0\uff0f\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u30b3\u30df\u30e5\u30cb\u30c6\u30a3<\/td>\n<\/tr>\n
PRICING<\/th>\nFree\uff08\u7121\u5236\u9650\u306e\u30e2\u30c7\u30eb\u5229\u7528\u30fb\u516c\u958b\u30ea\u30dd\u30b8\u30c8\u30ea\uff09\uff0fPro $9\/\u6708\uff08\u30d7\u30e9\u30a4\u30d9\u30fc\u30c8\u30ea\u30dd\u30b8\u30c8\u30ea\u30fb\u512a\u5148\u30b5\u30dd\u30fc\u30c8\uff09\uff0fEnterprise $20\/\u30e6\u30fc\u30b6\u30fc\/\u6708\u301c\uff08SSO\u30fb\u76e3\u67fb\u30ed\u30b0\u30fb\u5c02\u7528\u30a4\u30f3\u30d5\u30e9\uff09\uff0fInference API\u30fbSpaces Hardware \u306f\u5f93\u91cf\u8ab2\u91d1<\/td>\n<\/tr>\n
PLATFORM<\/th>\nWeb\uff08huggingface.co\uff09\uff0fPython SDK\uff08transformers\u30fbdatasets\u30fbdiffusers\uff09\uff0fREST API\uff0fCLI<\/td>\n<\/tr>\n
BEST FOR<\/th>\nAI\u30a8\u30f3\u30b8\u30cb\u30a2\u30fb\u30c7\u30fc\u30bf\u30b5\u30a4\u30a8\u30f3\u30c6\u30a3\u30b9\u30c8\u30fb\u500b\u4eba\u958b\u767a\u8005\u30fbAI\u526f\u696d\u3092\u76ee\u6307\u3059\u6280\u8853\u8005<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
\ud83d\udce1<\/span>
\nHugging Face<\/span> \u3068\u306f\u4f55\u304b<\/span><\/div>\n

Hugging Face\u306f2016\u5e74\u306b\u30cb\u30e5\u30fc\u30e8\u30fc\u30af\u3067\u5275\u696d\u3055\u308c\u305fAI\u30b9\u30bf\u30fc\u30c8\u30a2\u30c3\u30d7\u3002
\n\u5f53\u521d\u306f\u30c1\u30e3\u30c3\u30c8\u30dc\u30c3\u30c8\u30a2\u30d7\u30ea\u3068\u3057\u3066\u51fa\u767a\u3057\u305f\u304c\u30012018\u5e74\u306bNLP\uff08\u81ea\u7136\u8a00\u8a9e\u51e6\u7406\uff09\u30e9\u30a4\u30d6\u30e9\u30ea\u300cTransformers\u300d\u3092\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u3067\u516c\u958b\u3057\u3066\u304b\u3089\u6025\u6210\u9577\u3002
\n\u73fe\u5728\u306f90\u4e07\u4ef6\u8d85\u306e\u30e2\u30c7\u30eb\u30fb25\u4e07\u4ef6\u8d85\u306e\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u30fb30\u4e07\u4ef6\u8d85\u306e\u30c7\u30e2\u30a2\u30d7\u30ea\uff08Spaces\uff09<\/strong>\u3092\u64c1\u3059\u308b\u4e16\u754c\u6700\u5927\u306eAI\u30b3\u30df\u30e5\u30cb\u30c6\u30a3\u30d7\u30e9\u30c3\u30c8\u30d5\u30a9\u30fc\u30e0\u3078\u3068\u9032\u5316\u3057\u305f\u3002<\/p>\n

GitHub\u304c\u30b3\u30fc\u30c9\u3092\u5171\u6709\u3059\u308b\u30d7\u30e9\u30c3\u30c8\u30d5\u30a9\u30fc\u30e0\u3067\u3042\u308b\u306a\u3089\u3001Hugging Face\u306fAI\u30e2\u30c7\u30eb\u305d\u306e\u3082\u306e\u3092\u5171\u6709\u3059\u308b\u5834\u6240\u3002
\nLlama 3\u30fbMistral\u30fbStable Diffusion\u30fbWhisper\u306a\u3069\u3001\u8457\u540d\u306a\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u30e2\u30c7\u30eb\u306e\u307b\u307c\u3059\u3079\u3066\u304c\u516c\u958b\u3055\u308c\u3066\u304a\u308a\u3001\u30b3\u30fc\u30c91\u884c\u3067\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u30fb\u5b9f\u884c\u3067\u304d\u308b<\/strong>\u3002
\n\u4f01\u696d\u8a55\u4fa1\u984d\u306f2023\u5e74\u6642\u70b9\u3067\u7d0445\u5104\u30c9\u30eb\u3002Google\u3084Amazon\u3001NVIDIA\u306a\u3069\u931a\u3005\u305f\u308bIT\u5927\u624b\u304c\u51fa\u8cc7\u3059\u308b\u30e6\u30cb\u30b3\u30fc\u30f3\u4f01\u696d\u3067\u3082\u3042\u308b\u3002<\/p>\n

\u526f\u696d\u76ee\u7dda\u3067\u306f\u300c\u7121\u6599\u3067\u30d7\u30ed\u54c1\u8cea\u306eAI\u30e2\u30c7\u30eb\u3092\u4f7f\u3044\u653e\u984c\u300d\u3068\u3044\u3046\u7834\u58ca\u529b\u304c\u969b\u7acb\u3064\u3002
\n\u6708\u984d\u6570\u4e07\u5186\u306eAPI\u30b3\u30b9\u30c8\u3092\u4e00\u5207\u304b\u3051\u305a\u306b\u3001\u81ea\u5206\u306e\u30b5\u30fc\u30d3\u30b9\u306bAI\u3092\u7d44\u307f\u8fbc\u3081\u308b\u3002
\n\u305d\u308c\u304cHugging Face\u306e\u6700\u5927\u306e\u4fa1\u5024\u3060\u3002<\/p>\n

\u26a1<\/span>
\n\u4e3b\u8981\u6a5f\u80fd 3\u3064<\/span><\/div>\n
FEATURE_01<\/span>
\nModel Hub \u2500\u2500 \u4e16\u754c\u6700\u5927\u306eAI\u30e2\u30c7\u30eb\u30ea\u30dd\u30b8\u30c8\u30ea<\/span>
\n\u30c6\u30ad\u30b9\u30c8\u751f\u6210\u30fb\u753b\u50cf\u751f\u6210\u30fb\u97f3\u58f0\u8a8d\u8b58\u30fb\u7ffb\u8a33\u30fb\u611f\u60c5\u5206\u6790\u306a\u3069\u3001\u3042\u3089\u3086\u308b\u30bf\u30b9\u30af\u306e\u30e2\u30c7\u30eb\u304c90\u4e07\u4ef6\u4ee5\u4e0a\u516c\u958b\u3055\u308c\u3066\u3044\u308b\u3002
\n\u691c\u7d22\u30fb\u30d5\u30a3\u30eb\u30bf\u30ea\u30f3\u30b0\u6a5f\u80fd\u304c\u5145\u5b9f\u3057\u3066\u304a\u308a\u3001\u300c\u65e5\u672c\u8a9e\u5bfe\u5fdc\u30fb\u5546\u7528\u5229\u7528\u53ef\u30fb\u30e9\u30a4\u30bb\u30f3\u30b9MIT\u300d\u3068\u3044\u3063\u305f\u6761\u4ef6\u3092\u7d5e\u308a\u8fbc\u3093\u3067\u5373\u5ea7\u306b\u76ee\u7684\u306e\u30e2\u30c7\u30eb\u3092\u898b\u3064\u3051\u3089\u308c\u308b\u3002
\n\u5404\u30e2\u30c7\u30eb\u30da\u30fc\u30b8\u306b\u306f\u30c7\u30e2\uff08Inference Widget\uff09\u304c\u57cb\u3081\u8fbc\u307e\u308c\u3066\u3044\u308b\u305f\u3081\u3001\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u524d\u306b\u30d6\u30e9\u30a6\u30b6\u4e0a\u3067\u52d5\u4f5c\u78ba\u8a8d\u3082\u53ef\u80fd\u3002\u500b\u4eba\u958b\u767a\u8005\u304cOpenAI API\u306e\u4ee3\u66ff\u3068\u3057\u3066\u81ea\u524d\u30b5\u30fc\u30d0\u30fc\u306b\u30c7\u30d7\u30ed\u30a4\u3059\u308b\u30b1\u30fc\u30b9\u304c\u6025\u5897\u3057\u3066\u3044\u308b\u3002<\/span><\/div>\n
FEATURE_02<\/span>
\nSpaces \u2500\u2500 AI\u30a2\u30d7\u30ea\u3092\u30ef\u30f3\u30af\u30ea\u30c3\u30af\u3067\u516c\u958b\u30fb\u5171\u6709<\/span>
\nGradio\u307e\u305f\u306fStreamlit\u3067\u66f8\u3044\u305fPython\u30a2\u30d7\u30ea\u3092\u3001\u30a4\u30f3\u30d5\u30e9\u8a2d\u5b9a\u306a\u3057\u3067\u7121\u6599\u516c\u958b\u3067\u304d\u308b\u30db\u30b9\u30c6\u30a3\u30f3\u30b0\u30b5\u30fc\u30d3\u30b9\u3002
\n\u81ea\u4f5c\u306eAI\u30c7\u30e2\u30fb\u30c4\u30fc\u30eb\u30fb\u30dd\u30fc\u30c8\u30d5\u30a9\u30ea\u30aa\u3092\u6570\u5206\u3067Web\u516c\u958b\u3067\u304d\u308b\u305f\u3081\u3001\u526f\u696d\u306e\u5b9f\u7e3e\u30fb\u30b9\u30ad\u30eb\u8a3c\u660e\u3068\u3057\u3066\u975e\u5e38\u306b\u5f37\u529b<\/strong>\u3002
\n\u7121\u6599\u67a0\u306fCPU\u74b0\u5883\uff082\u30b3\u30a2\u30fb16GB RAM\uff09\u3067\u52d5\u4f5c\u3057\u3001GPU\u74b0\u5883\uff08A10G\u30fbT4\u306a\u3069\uff09\u306f\u30aa\u30f3\u30c7\u30de\u30f3\u30c9\u3067\u8ab2\u91d1\u8ffd\u52a0\u3067\u304d\u308b\u3002\u6708\u984d$0\u304b\u3089\u59cb\u3081\u3089\u308c\u308b\u70b9\u304c\u500b\u4eba\u958b\u767a\u8005\u306b\u523a\u3055\u308b\u3002<\/span><\/div>\n
FEATURE_03<\/span>
\nTransformers & Inference API \u2500\u2500 \u30b3\u30fc\u30c91\u884c\u3067AI\u7d44\u307f\u8fbc\u307f<\/span>
\nPython\u30e9\u30a4\u30d6\u30e9\u30ea\u300ctransformers\u300d\u306fpip install\u3059\u308b\u3060\u3051\u3067\u3001\u30e2\u30c7\u30eb\u306e\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u304b\u3089\u63a8\u8ad6\u307e\u3067\u6700\u77ed3\u884c\u306e\u30b3\u30fc\u30c9\u3067\u52d5\u304f\u3002
\n\u3055\u3089\u306bHugging Face Inference API\uff08\u30db\u30b9\u30c8\u578b\uff09\u3092\u4f7f\u3048\u3070\u3001\u81ea\u524d\u30b5\u30fc\u30d0\u30fc\u4e0d\u8981\u3067REST API\u7d4c\u7531\u3067\u30e2\u30c7\u30eb\u3092\u547c\u3073\u51fa\u305b\u308b\u3002
\nServerless Inference API\uff08\u7121\u6599\u30fb\u30ec\u30fc\u30c8\u5236\u9650\u3042\u308a\uff09\u3068Dedicated Endpoints\uff08\u5c02\u7528\u30fb\u5f93\u91cf\u8ab2\u91d1\uff09\u306e2\u6bb5\u69cb\u3048\u3067\u3001\u5c0f\u898f\u6a21\u526f\u696d\u304b\u3089\u672c\u683cSaaS\u307e\u3067\u30b9\u30b1\u30fc\u30e9\u30d6\u30eb\u306b\u5bfe\u5fdc\u3067\u304d\u308b\u3002<\/span><\/div>\n
\u2696\ufe0f<\/span>
\n\u4f3c\u305fAI\u3068\u306e\u9055\u3044<\/span><\/div>\n\n\n\n\n\n\n\n
<\/th>\nHugging Face<\/th>\nReplicate<\/th>\nOpenAI API<\/th>\n<\/tr>\n
\u30e2\u30c7\u30eb\u6570<\/td>\n90\u4e07\u4ef6\u4ee5\u4e0a\uff08\u6700\u5927\u898f\u6a21\uff09<\/td>\n\u6570\u5343\u4ef6\uff08\u53b3\u9078\uff09<\/td>\n\u81ea\u793e\u30e2\u30c7\u30eb\u306e\u307f<\/td>\n<\/tr>\n
\u7121\u6599\u67a0<\/td>\nServerless API\u30fbSpaces \u7121\u6599\u3042\u308a<\/td>\n\u5f93\u91cf\u8ab2\u91d1\u306e\u307f<\/td>\n\u8a66\u7528\u30af\u30ec\u30b8\u30c3\u30c8\u306e\u307f<\/td>\n<\/tr>\n
\u30e2\u30c7\u30eb\u54c1\u8cea<\/td>\n\u7389\u77f3\u6df7\u4ea4\uff08\u8981\u7cbe\u67fb\uff09<\/td>\n\u53b3\u9078\u30fb\u5b89\u5b9a<\/td>\n\u6700\u9ad8\u6c34\u6e96\uff08GPT-4o\uff09<\/td>\n<\/tr>\n
\u5546\u7528\u30ab\u30b9\u30bf\u30de\u30a4\u30ba<\/td>\nFine-tuning\u30fb\u30ed\u30fc\u30ab\u30eb\u5b9f\u884c \u5b8c\u5168\u81ea\u7531<\/td>\nFine-tuning \u5bfe\u5fdc\uff08\u4e00\u90e8\uff09<\/td>\nFine-tuning \u5bfe\u5fdc\uff08GPT-4o\u306a\u3069\uff09<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
\ud83c\udfaf<\/span>
\n\u3053\u3093\u306a\u4f7f\u3044\u65b9\u304c\u5f37\u3044<\/span><\/div>\n
\n
\ud83e\udd16<\/span>
\nAPI\u4ee3\u66ff\u3067\u6708\u984d\u30b3\u30b9\u30c8\u6fc0\u6e1b<\/span>
\nOpenAI API\u306b\u6bce\u6708\u6570\u4e07\u5186\u6255\u3063\u3066\u3044\u308b\u306a\u3089\u3001Mistral\u30fbLlama 3\u306a\u3069\u306b\u5207\u308a\u66ff\u3048\u3066\u30b3\u30b9\u30c8\u3092\u307b\u307c\u30bc\u30ed\u306b\u3002\u7ffb\u8a33\u30fb\u8981\u7d04\u30fb\u5206\u985e\u30bf\u30b9\u30af\u306a\u3089\u54c1\u8cea\u3082\u5341\u5206\u5b9f\u7528\u30ec\u30d9\u30eb\u3002<\/span><\/div>\n
\ud83c\udfa8<\/span>
\n\u753b\u50cf\u751f\u6210AI\u3092\u81ea\u524d\u30b5\u30fc\u30d0\u30fc\u3067\u904b\u7528<\/span>
\nStable Diffusion\u30fbFLUX.1\u306a\u3069\u3092\u30ed\u30fc\u30ab\u30eb\u307e\u305f\u306f\u30af\u30e9\u30a6\u30c9GPU\u3067\u52d5\u304b\u3057\u3001\u753b\u50cf\u751f\u6210SaaS\u3084\u53d7\u8a17\u6848\u4ef6\u306b\u6d3b\u7528\u3002Midjourney\u4f9d\u5b58\u304b\u3089\u8131\u5374\u3067\u304d\u308b\u3002<\/span><\/div>\n
\ud83c\udf99\ufe0f<\/span>
\n\u97f3\u58f0\u6587\u5b57\u8d77\u3053\u3057\u30fb\u7ffb\u8a33\u30b5\u30fc\u30d3\u30b9\u69cb\u7bc9<\/span>
\nOpenAI Whisper\u306e\u91cd\u307f\u3092Hugging Face\u7d4c\u7531\u3067\u53d6\u5f97\u3057\u3001\u8b70\u4e8b\u9332\u81ea\u52d5\u5316SaaS\u3084\u30dd\u30c3\u30c9\u30ad\u30e3\u30b9\u30c8\u6587\u5b57\u8d77\u3053\u3057\u30c4\u30fc\u30eb\u3092\u526f\u696d\u3068\u3057\u3066\u8ca9\u58f2\u3002\u6708\u984d$2,000\u8d85\u306e\u53ce\u76ca\u4e8b\u4f8b\u3042\u308a\u3002<\/span><\/div>\n
\ud83d\udcca<\/span>
\nFine-tuning\u3067\u7279\u5316\u578b\u30e2\u30c7\u30eb\u3092\u8ca9\u58f2<\/span>
\n\u696d\u754c\u7279\u5316\uff08\u6cd5\u5f8b\u30fb\u533b\u7642\u30fbEC\u306a\u3069\uff09\u306e\u30c7\u30fc\u30bf\u3067\u30e2\u30c7\u30eb\u3092Fine-tuning\u3002\u72ec\u81ea\u30e2\u30c7\u30eb\u3092Hugging Face\u4e0a\u3067Private\u30db\u30b9\u30c8\u3057\u3001API\u3068\u3057\u3066\u4f01\u696d\u306b\u63d0\u4f9b\u3059\u308bBtoB\u30d3\u30b8\u30cd\u30b9\u304c\u53ef\u80fd\u3002<\/span><\/div>\n<\/div>\n
\ud83d\udcbb<\/span>
\n\u52b9\u679c\u7684\u306a\u30d7\u30ed\u30f3\u30d7\u30c8\u4f8b<\/span><\/div>\n
PROMPT_EXAMPLE<\/span>
\n
\n\u2500\u2500 \u30b3\u30fc\u30c9\u4f8b\u2460\uff1aTransformers\u3067\u65e5\u672c\u8a9e\u30c6\u30ad\u30b9\u30c8\u5206\u985e\uff08\u611f\u60c5\u5206\u6790\uff09
\nfrom transformers import pipeline
\nclassifier = pipeline(“sentiment-analysis”, model=”koheiduck\/bert-japanese-finetuned-sentiment”)
\nresult = classifier(“\u3053\u306e\u30b5\u30fc\u30d3\u30b9\u306f\u6700\u9ad8\u3067\u3059\uff01”)
\nprint(result) # [{‘label’: ‘POSITIVE’, ‘score’: 0.99}]<\/span><\/span>\u2500\u2500 \u30b3\u30fc\u30c9\u4f8b\u2461\uff1aInference API\uff08REST\uff09\u3067\u30c6\u30ad\u30b9\u30c8\u751f\u6210
\nimport requests
\nAPI_URL = “https:\/\/api-inference.huggingface.co\/models\/mistralai\/Mistral-7B-Instruct-v0.3”
\nheaders = {“Authorization”: “Bearer {\u3042\u306a\u305f\u306eHF\u30c8\u30fc\u30af\u30f3}”}
\npayload = {“inputs”: “\u526f\u696d\u3067\u670810\u4e07\u5186\u7a3c\u3050\u6700\u77ed\u30eb\u30fc\u30c8\u30923\u3064\u6559\u3048\u3066\u304f\u3060\u3055\u3044\u3002”}
\nresponse = requests.post(API_URL, headers=headers, json=payload)
\nprint(response.json())<\/p>\n

\u2500\u2500 \u30b3\u30fc\u30c9\u4f8b\u2462\uff1aSpaces\uff08Gradio\uff09\u3067AI\u30a2\u30d7\u30ea\u3092\u7121\u6599\u516c\u958b
\nimport gradio as gr
\nfrom transformers import pipeline
\npipe = pipeline(“summarization”, model=”facebook\/bart-large-cnn”)
\ndef summarize(text):
\nreturn pipe(text, max_length=130)[0][“summary_text”]
\ngr.Interface(fn=summarize, inputs=”textbox”, outputs=”textbox”).launch()<\/p>\n<\/div>\n

\ud83d\ude80<\/span>
\n\u526f\u696d\u30fb\u30d3\u30b8\u30cd\u30b9\u3078\u306e\u6d3b\u7528\u6cd5<\/span><\/div>\n
\u25b7 \u5b9f\u88c5\u30a2\u30a4\u30c7\u30a24\u9078<\/span><\/p>\n