{"id":3347,"date":"2025-01-17T08:00:00","date_gmt":"2025-01-16T23:00:00","guid":{"rendered":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/?p=3347"},"modified":"2025-01-16T11:06:54","modified_gmt":"2025-01-16T02:06:54","slug":"q8-chat-efficient-generative-ai-experience-xeon","status":"publish","type":"post","link":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/2025\/01\/17\/q8-chat-efficient-generative-ai-experience-xeon\/","title":{"rendered":"\u5c0f\u3055\u3044\u307b\u3069\u826f\u3044: Q8-Chat LLM \u306f\u30a4\u30f3\u30c6\u30eb\u00ae Xeon\u00ae \u30d7\u30ed\u30bb\u30c3\u30b5\u30fc\u4e0a\u3067\u52b9\u7387\u7684\u306a\u751f\u6210 AI \u30a8\u30af\u30b9\u30da\u30ea\u30a8\u30f3\u30b9\u3092\u5b9f\u73fe"},"content":{"rendered":"\n
\n\n\n\n

\u3053\u306e\u8a18\u4e8b\u306f\u3001\u30a4\u30f3\u30c6\u30eb\u306e\u30a6\u30a7\u30d6\u30b5\u30a4\u30c8\u3067\u516c\u958b\u3055\u308c\u3066\u3044\u308b\u300cSmaller is Better: Q8-Chat LLM is an Efficient Generative AI Experience on Intel\u00ae Xeon\u00ae Processors<\/a>\u300d\u306e\u65e5\u672c\u8a9e\u53c2\u8003\u8a33\u3067\u3059\u3002\u539f\u6587\u306f\u66f4\u65b0\u3055\u308c\u308b\u53ef\u80fd\u6027\u304c\u3042\u308a\u307e\u3059\u3002\u539f\u6587\u3068\u7ffb\u8a33\u6587\u306e\u5185\u5bb9\u304c\u7570\u306a\u308b\u5834\u5408\u306f\u539f\u6587\u3092\u512a\u5148\u3057\u3066\u304f\u3060\u3055\u3044\u3002<\/p>\n\n\n\n

\n\n\n\n

\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb (LLM) \u306f\u3001\u30de\u30b7\u30f3\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u4e16\u754c\u306b\u65cb\u98a8\u3092\u5dfb\u304d\u8d77\u3053\u3057\u3066\u3044\u307e\u3059\u3002\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30fb\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u30fc\u306e\u304a\u304b\u3052\u3067\u3001LLM \u306f\u30c6\u30ad\u30b9\u30c8\u3001\u753b\u50cf\u3001\u30d3\u30c7\u30aa\u3001\u97f3\u58f0\u306a\u3069\u306e\u81a8\u5927\u306a\u91cf\u306e\u975e\u69cb\u9020\u5316\u30c7\u30fc\u30bf\u304b\u3089\u5b66\u7fd2\u3059\u308b\u9a5a\u7570\u7684\u306a\u80fd\u529b\u3092\u5099\u3048\u3066\u3044\u307e\u3059\u3002\u30c6\u30ad\u30b9\u30c8\u5206\u985e\u306a\u3069\u306e\u62bd\u51fa\u578b\u30bf\u30b9\u30af\u3001\u30c6\u30ad\u30b9\u30c8\u8981\u7d04\u3084\u30c6\u30ad\u30b9\u30c8\u304b\u3089\u753b\u50cf\u751f\u6210\u306a\u3069\u306e\u751f\u6210\u578b\u30bf\u30b9\u30af\u306a\u3069\u3001\u591a\u304f\u306e\u30bf\u30b9\u30af\u30bf\u30a4\u30d7\u3067\u512a\u308c\u305f\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u767a\u63ee\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n

\u305d\u306e\u540d\u524d\u304c\u793a\u3059\u3088\u3046\u306b\u3001LLM \u306f\u5927\u898f\u6a21\u306a\u30e2\u30c7\u30eb\u3067\u3042\u308a\u3001\u901a\u5e38\u30d1\u30e9\u30e1\u30fc\u30bf\u30fc\u6570\u306f 100 \u5104\u3092\u8d85\u3048\u3001BLOOM \u306a\u3069\u4e00\u90e8\u306e\u30e2\u30c7\u30eb\u3067\u306f 1,000 \u5104\u3092\u8d85\u3048\u307e\u3059\u3002\u691c\u7d22\u3084\u4f1a\u8a71\u578b\u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3\u306a\u3069\u306e\u4f4e\u30ec\u30a4\u30c6\u30f3\u30b7\u30fc\u306e\u30e6\u30fc\u30b9\u30b1\u30fc\u30b9\u3067\u5341\u5206\u306a\u901f\u3055\u3067\u4e88\u6e2c\u3059\u308b\u306b\u306f\u3001\u901a\u5e38\u3001\u30cf\u30a4\u30a8\u30f3\u30c9 GPU \u306a\u3069\u306e\u5927\u91cf\u306e\u8a08\u7b97\u80fd\u529b\u3092\u5fc5\u8981\u3068\u3057\u307e\u3059\u3002\u6b8b\u5ff5\u306a\u304c\u3089\u3001\u591a\u304f\u306e\u7d44\u7e54\u306b\u3068\u3063\u3066\u3001\u95a2\u9023\u30b3\u30b9\u30c8\u306f\u975e\u5e38\u306b\u9ad8\u984d\u3067\u3042\u308a\u3001\u6700\u5148\u7aef\u306e LLM \u3092\u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3\u3067\u4f7f\u7528\u3059\u308b\u3053\u3068\u306f\u56f0\u96e3\u3067\u3059\u3002<\/p>\n

\u3053\u306e\u8a18\u4e8b\u3067\u306f\u3001LLM \u306e\u30b5\u30a4\u30ba\u3068\u63a8\u8ad6\u30ec\u30a4\u30c6\u30f3\u30b7\u30fc\u3092\u8efd\u6e1b\u3057\u3001\u30a4\u30f3\u30c6\u30eb\u306e CPU \u3067\u52b9\u7387\u826f\u304f\u5b9f\u884c\u3067\u304d\u308b\u3088\u3046\u306b\u3059\u308b\u6700\u9069\u5316\u624b\u6cd5\u306b\u3064\u3044\u3066\u8aac\u660e\u3057\u307e\u3059\u3002<\/p>\n

\u91cf\u5b50\u5316\u306b\u3064\u3044\u3066<\/h2>\n

LLM \u306f\u901a\u5e38\u300116 \u30d3\u30c3\u30c8\u6d6e\u52d5\u5c0f\u6570\u70b9\u30d1\u30e9\u30e1\u30fc\u30bf\u30fc (FP16\/BF16) \u3092\u4f7f\u7528\u3057\u3066\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3057\u307e\u3059\u3002\u3057\u305f\u304c\u3063\u3066\u3001\u5358\u4e00\u306e\u91cd\u307f\u307e\u305f\u306f\u6d3b\u6027\u5316\u306e\u5024\u306e\u4fdd\u5b58\u306b\u306f\u30012 \u30d0\u30a4\u30c8\u306e\u30e1\u30e2\u30ea\u30fc\u304c\u5fc5\u8981\u3067\u3059\u3002\u3055\u3089\u306b\u3001\u6d6e\u52d5\u5c0f\u6570\u70b9\u6f14\u7b97\u306f\u6574\u6570\u6f14\u7b97\u3088\u308a\u3082\u8907\u96d1\u3067\u6642\u9593\u304c\u304b\u304b\u308a\u3001\u8ffd\u52a0\u306e\u8a08\u7b97\u80fd\u529b\u3092\u5fc5\u8981\u3068\u3057\u307e\u3059\u3002<\/p>\n

\u91cf\u5b50\u5316\u306f\u3001\u30e2\u30c7\u30eb\u30fb\u30d1\u30e9\u30e1\u30fc\u30bf\u30fc\u306e\u4e00\u610f\u306e\u5024\u306e\u7bc4\u56f2\u3092\u7e2e\u5c0f\u3059\u308b\u3053\u3068\u3067\u3001\u4e21\u65b9\u306e\u554f\u984c\u3092\u89e3\u6c7a\u3059\u308b\u3053\u3068\u3092\u76ee\u7684\u3068\u3057\u305f\u30e2\u30c7\u30eb\u5727\u7e2e\u624b\u6cd5\u3067\u3059\u3002\u4f8b\u3048\u3070\u3001\u30e2\u30c7\u30eb\u3092 8 \u30d3\u30c3\u30c8\u6574\u6570 (INT8) \u306a\u3069\u306e\u4f4e\u3044\u7cbe\u5ea6\u306b\u91cf\u5b50\u5316\u3057\u3066\u7e2e\u5c0f\u3057\u3001\u8907\u96d1\u306a\u6d6e\u52d5\u5c0f\u6570\u70b9\u6f14\u7b97\u3092\u3088\u308a\u5358\u7d14\u3067\u9ad8\u901f\u306a\u6574\u6570\u6f14\u7b97\u306b\u7f6e\u304d\u63db\u3048\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002<\/p>\n

\u7c21\u5358\u306b\u8a00\u3046\u3068\u3001\u91cf\u5b50\u5316\u306f\u30e2\u30c7\u30eb\u30fb\u30d1\u30e9\u30e1\u30fc\u30bf\u30fc\u3092\u3088\u308a\u5c0f\u3055\u306a\u5024\u306e\u7bc4\u56f2\u306b\u518d\u30b9\u30b1\u30fc\u30ea\u30f3\u30b0\u3057\u307e\u3059\u3002\u30d9\u30b9\u30c8\u30b1\u30fc\u30b9\u3067\u306f\u3001\u30e2\u30c7\u30eb\u306e\u7cbe\u5ea6\u306b\u5f71\u97ff\u3092\u4e0e\u3048\u308b\u3053\u3068\u306a\u304f\u3001\u30e2\u30c7\u30eb\u30b5\u30a4\u30ba\u3092\u5c11\u306a\u304f\u3068\u3082\u534a\u5206\u306b\u7e2e\u5c0f\u3057\u307e\u3059\u3002<\/p>\n

\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u4e2d\u306b\u91cf\u5b50\u5316\u3092\u9069\u7528\u3067\u304d\u308b\u91cf\u5b50\u5316\u5bfe\u5fdc\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0 (QAT<\/a> (\u82f1\u8a9e)) \u306f\u901a\u5e38\u3001\u6700\u826f\u306e\u7d50\u679c\u3092\u3082\u305f\u3089\u3057\u307e\u3059\u3002\u65e2\u5b58\u306e\u30e2\u30c7\u30eb\u3092\u91cf\u5b50\u5316\u3059\u308b\u5834\u5408\u306f\u3001\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u5f8c\u306e\u91cf\u5b50\u5316 (PTQ<\/a> (\u82f1\u8a9e)) \u3092\u9069\u7528\u3067\u304d\u307e\u3059\u3002\u3053\u308c\u306f\u3001\u308f\u305a\u304b\u306a\u8a08\u7b97\u80fd\u529b\u3057\u304b\u5fc5\u8981\u3068\u305b\u305a\u3001\u306f\u308b\u304b\u306b\u9ad8\u901f\u306a\u624b\u6cd5\u3067\u3059\u3002<\/p>\n

\u3055\u307e\u3056\u307e\u306a\u91cf\u5b50\u5316\u30c4\u30fc\u30eb\u3042\u308a\u307e\u3059\u3002\u4f8b\u3048\u3070\u3001PyTorch \u306b\u306f\u91cf\u5b50\u5316<\/a> (\u82f1\u8a9e) \u30b5\u30dd\u30fc\u30c8\u304c\u7d44\u307f\u8fbc\u307e\u308c\u3066\u3044\u307e\u3059\u3002\u307e\u305f\u3001\u958b\u767a\u8005\u306b\u3068\u3063\u3066\u4f7f\u3044\u3084\u3059\u3044 QAT \u304a\u3088\u3073 PTQ \u7528\u306e API \u3092\u542b\u3080 Hugging Face Optimum Intel<\/a> (\u82f1\u8a9e) \u30e9\u30a4\u30d6\u30e9\u30ea\u30fc\u3092\u4f7f\u7528\u3059\u308b\u3053\u3068\u3082\u3067\u304d\u307e\u3059\u3002<\/p>\n

LLM \u306e\u91cf\u5b50\u5316<\/h2>\n

\u6700\u8fd1\u306e\u7814\u7a76[1]<\/sup><\/a> [2]<\/sup><\/a> \u306b\u3088\u308b\u3068\u3001\u73fe\u5728\u306e\u91cf\u5b50\u5316\u624b\u6cd5\u306f LLM \u3067\u306f\u3046\u307e\u304f\u6a5f\u80fd\u3057\u307e\u305b\u3093\u3002\u7279\u306b\u3001LLM \u306f\u3059\u3079\u3066\u306e\u30ec\u30a4\u30e4\u30fc\u3068\u30c8\u30fc\u30af\u30f3\u306b\u304a\u3044\u3066\u7279\u5b9a\u306e\u6d3b\u6027\u5316\u30c1\u30e3\u30cd\u30eb\u3067\u5927\u304d\u306a\u5916\u308c\u5024\u3092\u793a\u3057\u307e\u3059\u3002<\/p>\n

OPT-13B \u30e2\u30c7\u30eb\u306e\u4f8b\u3092\u6b21\u306b\u793a\u3057\u307e\u3059\u3002\u6d3b\u6027\u5316\u30c1\u30e3\u30cd\u30eb\u306e 1 \u3064\u304c\u3001\u3059\u3079\u3066\u306e\u30c8\u30fc\u30af\u30f3\u3067\u307b\u304b\u306e\u3059\u3079\u3066\u306e\u30c1\u30e3\u30cd\u30eb\u3088\u308a\u3082\u306f\u308b\u304b\u306b\u5927\u304d\u306a\u5024\u3092\u6301\u3063\u3066\u3044\u308b\u3053\u3068\u304c\u5206\u304b\u308a\u307e\u3059\u3002\u3053\u306e\u73fe\u8c61\u306f\u3001\u30e2\u30c7\u30eb\u306e\u3059\u3079\u3066\u306e Transformer \u30ec\u30a4\u30e4\u30fc\u3067\u78ba\u8a8d\u3067\u304d\u307e\u3059\u3002<\/p>\n

\"\"<\/p>\n

\u51fa\u5178: SmoothQuant<\/strong><\/p>\n

\u3053\u308c\u307e\u3067\u306e\u6700\u826f\u306e\u91cf\u5b50\u5316\u624b\u6cd5\u306f\u3001\u6d3b\u6027\u5316\u3092\u30c8\u30fc\u30af\u30f3\u5358\u4f4d\u3067\u91cf\u5b50\u5316\u3059\u308b\u305f\u3081\u3001\u5916\u308c\u5024\u304c\u5207\u308a\u6368\u3066\u3089\u308c\u308b\u304b\u3001\u5c0f\u3055\u306a\u6d3b\u6027\u5316\u306e\u5024\u304c\u30a2\u30f3\u30c0\u30fc\u30d5\u30ed\u30fc\u3068\u306a\u308a\u3001\u30e2\u30c7\u30eb\u306e\u54c1\u8cea\u304c\u5927\u5e45\u306b\u4f4e\u4e0b\u3057\u307e\u3059\u3002\u3055\u3089\u306b\u3001\u91cf\u5b50\u5316\u5bfe\u5fdc\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u306b\u306f\u8ffd\u52a0\u306e\u30e2\u30c7\u30eb\u30fb\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u304c\u5fc5\u8981\u3067\u3059\u304c\u3001\u3053\u308c\u306f\u307b\u3068\u3093\u3069\u306e\u5834\u5408\u3001\u8a08\u7b97\u30ea\u30bd\u30fc\u30b9\u3068\u30c7\u30fc\u30bf\u306e\u4e0d\u8db3\u306b\u3088\u308a\u73fe\u5b9f\u7684\u3067\u306f\u3042\u308a\u307e\u305b\u3093\u3002<\/p>\n

SmoothQuant<\/strong>[3]<\/sup><\/a> [4]<\/sup><\/a> \u306f\u3001\u3053\u306e\u554f\u984c\u3092\u89e3\u6c7a\u3059\u308b\u65b0\u3057\u3044\u91cf\u5b50\u5316\u624b\u6cd5\u3067\u3059<\/strong>\u3002\u3053\u308c\u306f\u3001\u91cd\u307f\u3068\u6d3b\u6027\u5316\u306b\u6570\u5b66\u7684\u5909\u63db\u3092\u9069\u7528\u3057\u3001\u91cd\u307f\u306e\u6bd4\u7387\u3092\u5897\u3084\u3059\u4ee3\u308f\u308a\u306b\u3001\u6d3b\u6027\u5316\u306e\u5916\u308c\u5024\u3068\u975e\u5916\u308c\u5024\u306e\u6bd4\u7387\u3092\u6e1b\u3089\u3057\u307e\u3059\u3002\u3053\u306e\u5909\u63db\u306b\u3088\u308a\u3001Transformer \u30ec\u30a4\u30e4\u30fc\u304c\u300c\u91cf\u5b50\u5316\u306b\u9069\u3057\u305f\u300d\u3082\u306e\u306b\u306a\u308a\u3001\u30e2\u30c7\u30eb\u306e\u54c1\u8cea\u3092\u640d\u306a\u3046\u3053\u3068\u306a\u304f 8 \u30d3\u30c3\u30c8\u306e\u91cf\u5b50\u5316\u304c\u53ef\u80fd\u306b\u306a\u308a\u307e\u3059\u3002\u305d\u306e\u7d50\u679c\u3001SmoothQuant \u306f\u3001\u30a4\u30f3\u30c6\u30eb\u306e CPU \u30d7\u30e9\u30c3\u30c8\u30d5\u30a9\u30fc\u30e0\u3067\u9069\u5207\u306b\u52d5\u4f5c\u3059\u308b\u3001\u3088\u308a\u5c0f\u3055\u304f\u3001\u3088\u308a\u9ad8\u901f\u306a\u30e2\u30c7\u30eb\u3092\u751f\u6210\u3057\u307e\u3059\u3002<\/p>\n

\"\"<\/p>\n

\u305d\u308c\u3067\u306f\u3001SmoothQuant \u3092\u4e00\u822c\u7684\u306a LLM \u306b\u9069\u7528\u3057\u305f\u5834\u5408\u306b\u3069\u306e\u3088\u3046\u306b\u52d5\u4f5c\u3059\u308b\u304b\u3092\u898b\u3066\u307f\u307e\u3057\u3087\u3046\u3002<\/p>\n

SmoothQuant \u306b\u3088\u308b LLM \u306e\u91cf\u5b50\u5316<\/h2>\n

\u30a4\u30f3\u30c6\u30eb\u306e\u30a8\u30f3\u30b8\u30cb\u30a2\u306f\u3001SmoothQuant-O3 \u3092\u4f7f\u7528\u3057\u3066\u3001OPT 2.7B<\/a> \u3068 6.7B<\/a>[5]<\/sup><\/a>\u3001LLaMA 7B[6]<\/sup><\/a>\u3001Alpaca7B<\/a> [7]<\/sup><\/a>\u3001Vicuna 7B<\/a>[8]<\/sup><\/a>\u3001BloomZ 7.1B<\/a>[9]<\/sup><\/a>\u3001MPT-7B-chat[10]<\/sup><\/a> \u3092\u91cf\u5b50\u5316\u3057\u3001Language Model Evaluation Harness<\/a> (\u82f1\u8a9e) \u3092\u4f7f\u7528\u3057\u3066\u3001\u91cf\u5b50\u5316\u3057\u305f\u30e2\u30c7\u30eb\u306e\u7cbe\u5ea6\u3092\u8a55\u4fa1\u3057\u307e\u3057\u305f\u3002<\/p>\n

\u4ee5\u4e0b\u306e\u8868\u306b\u7d50\u679c\u306e\u6982\u8981\u3092\u793a\u3057\u307e\u3059\u30022 \u756a\u76ee\u306e\u5217\u306f\u3001\u91cf\u5b50\u5316\u5f8c\u306b\u6539\u5584\u3055\u308c\u305f\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306e\u6bd4\u7387\u3092\u793a\u3057\u307e\u3059\u30023 \u756a\u76ee\u306e\u5217\u306f\u3001\u5e73\u5747\u7684\u306a\u6027\u80fd\u4f4e\u4e0b\u3092\u793a\u3057\u307e\u3059 (*\u8ca0\u306e\u5024\u306f\u3001\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u304c\u6539\u5584\u3055\u308c\u305f\u3053\u3068\u3092\u793a\u3057\u307e\u3059)\u3002\u8a73\u7d30\u306a\u7d50\u679c\u306f\u3001\u3053\u306e\u8a18\u4e8b\u306e\u6700\u5f8c\u306b\u3042\u308a\u307e\u3059\u3002<\/p>\n

\"\"<\/p>\n

OPT \u30e2\u30c7\u30eb\u306f SmoothQuant \u91cf\u5b50\u5316\u306b\u6700\u9069\u306a\u5019\u88dc\u3067\u3042\u308b\u3053\u3068\u304c\u5206\u304b\u308a\u307e\u3059\u3002\u30e2\u30c7\u30eb\u30b5\u30a4\u30ba\u306f\u3001\u4e8b\u524d\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u6e08\u307f\u306e 16 \u30d3\u30c3\u30c8\u30fb\u30e2\u30c7\u30eb\u3068\u6bd4\u8f03\u3057\u3066\u7d04\u534a\u5206\u306b\u306a\u3063\u3066\u3044\u307e\u3059\u3002\u307b\u3068\u3093\u3069\u306e\u30e1\u30c8\u30ea\u30c3\u30af\u304c\u5411\u4e0a\u3057\u3001\u5411\u4e0a\u3057\u306a\u3044\u3082\u306e\u3082\u30da\u30ca\u30eb\u30c6\u30a3\u30fc\u306f\u308f\u305a\u304b\u3067\u3059\u3002<\/p>\n

LLaMA 7B \u3068 BloomZ 7.1B \u306e\u7d50\u679c\u306f\u5bfe\u7167\u7684\u3067\u3059\u3002\u30e2\u30c7\u30eb\u30b5\u30a4\u30ba\u306f\u7d04\u534a\u5206\u306b\u5727\u7e2e\u3055\u308c\u3001\u7d04\u534a\u5206\u306e\u30bf\u30b9\u30af\u3067\u306f\u30e1\u30c8\u30ea\u30c3\u30af\u304c\u5411\u4e0a\u3057\u3066\u3044\u307e\u3059\u3002\u6b8b\u308a\u306e\u534a\u5206\u3067\u306f\u30011 \u3064\u306e\u30bf\u30b9\u30af\u3067 3% \u3092\u8d85\u3048\u308b\u76f8\u5bfe\u7684\u306a\u4f4e\u4e0b\u304c\u898b\u3089\u308c\u307e\u3059\u304c\u3001\u305d\u306e\u307b\u304b\u306e\u30bf\u30b9\u30af\u3078\u306e\u5f71\u97ff\u306f\u308f\u305a\u304b\u3067\u3059\u3002<\/p>\n

\u3088\u308a\u5c0f\u3055\u306a\u30e2\u30c7\u30eb\u3067\u4f5c\u696d\u3059\u308b\u3053\u3068\u306e\u660e\u3089\u304b\u306a\u5229\u70b9\u306f\u3001\u63a8\u8ad6\u30ec\u30a4\u30c6\u30f3\u30b7\u30fc\u304c\u5927\u5e45\u306b\u77ed\u7e2e\u3055\u308c\u308b\u3053\u3068\u3067\u3059\u3002\u30b7\u30f3\u30b0\u30eb\u30bd\u30b1\u30c3\u30c8\u306e\u30a4\u30f3\u30c6\u30eb\u00ae Xeon\u00ae \u30b9\u30b1\u30fc\u30e9\u30d6\u30eb\u30fb\u30d7\u30ed\u30bb\u30c3\u30b5\u30fc (\u958b\u767a\u30b3\u30fc\u30c9\u540d Sapphire Rapids) \u4e0a\u3067 32 \u30b3\u30a2\u3001\u30d0\u30c3\u30c1\u30b5\u30a4\u30ba 1 \u3067 MPT-7B-chat \u30e2\u30c7\u30eb\u3092\u4f7f\u7528\u3057\u3066\u30ea\u30a2\u30eb\u30bf\u30a4\u30e0\u306e\u30c6\u30ad\u30b9\u30c8\u751f\u6210\u3092\u5b9f\u6f14\u3059\u308b\u30d3\u30c7\u30aa\u3092\u3053\u3061\u3089\u304b\u3089<\/a> (\u82f1\u8a9e) \u8996\u8074\u3067\u304d\u307e\u3059\u3002<\/p>\n

\u3053\u306e\u4f8b\u3067\u306f\u3001\u30e2\u30c7\u30eb\u306b\u300cWhat is the role of Hugging Face in democratizing NLP? (NLP \u306e\u5927\u8846\u5316\u306b\u304a\u3051\u308b Hugging Face \u306e\u5f79\u5272\u306f\u4f55\u3067\u3059\u304b?)\u300d\u3068\u8cea\u554f\u3057\u3066\u3044\u307e\u3059\u3002\u3053\u308c\u306b\u3088\u308a\u3001\u30e2\u30c7\u30eb\u306b\u6b21\u306e\u30d7\u30ed\u30f3\u30d7\u30c8\u304c\u9001\u4fe1\u3055\u308c\u307e\u3059\u3002<\/p>\n

A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user’s questions.
(\u597d\u5947\u5fc3\u65fa\u76db\u306a\u30e6\u30fc\u30b6\u30fc\u3068\u4eba\u5de5\u77e5\u80fd\u30a2\u30b7\u30b9\u30bf\u30f3\u30c8\u306e\u9593\u306e\u30c1\u30e3\u30c3\u30c8\u3002\u30a2\u30b7\u30b9\u30bf\u30f3\u30c8\u306f\u3001\u30e6\u30fc\u30b6\u30fc\u306e\u8cea\u554f\u306b\u5bfe\u3057\u3066\u3001\u5f79\u306b\u7acb\u3064\u3001\u8a73\u7d30\u3067\u4e01\u5be7\u306a\u56de\u7b54\u3092\u63d0\u4f9b\u3057\u307e\u3059\u3002)<\/p>\n

USER: What is the role of Hugging Face in democratizing NLP?
(\u30e6\u30fc\u30b6\u30fc: NLP \u306e\u6c11\u4e3b\u5316\u306b\u304a\u3051\u308b Hugging Face \u306e\u5f79\u5272\u306f\u4f55\u3067\u3059\u304b?)<\/p>\n

ASSISTANT:
(\u30a2\u30b7\u30b9\u30bf\u30f3\u30c8:)<\/p>\n<\/blockquote>\n

\u6b21\u306e\u30b9\u30c6\u30c3\u30d7<\/h2>\n

\u73fe\u5728\u3001\u3053\u308c\u3089\u306e\u65b0\u3057\u3044\u91cf\u5b50\u5316\u624b\u6cd5\u3092\u30a4\u30f3\u30c6\u30eb\u00ae \u30cb\u30e5\u30fc\u30e9\u30eb\u30fb\u30b3\u30f3\u30d7\u30ec\u30c3\u30b5\u30fc<\/a> (\u82f1\u8a9e) \u3092\u901a\u3058\u3066 Hugging Face Optimum Intel<\/a> (\u82f1\u8a9e) \u30e9\u30a4\u30d6\u30e9\u30ea\u30fc\u306b\u7d71\u5408\u3059\u308b\u4f5c\u696d\u304c\u9032\u884c\u4e2d\u3067\u3059\u3002\u3053\u306e\u4f5c\u696d\u304c\u5b8c\u4e86\u3059\u308b\u3068\u3001\u6570\u884c\u306e\u30b3\u30fc\u30c9\u3067\u3053\u308c\u3089\u306e\u30c7\u30e2\u3092\u518d\u73fe\u3067\u304d\u308b\u3088\u3046\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n

\u3054\u671f\u5f85\u304f\u3060\u3055\u3044\u3002\u5c06\u6765\u306f 8 \u30d3\u30c3\u30c8\u3067\u3059!<\/p>\n

\u3053\u306e\u8a18\u4e8b\u306f\u3001100% ChatGPT \u30d5\u30ea\u30fc\u3067\u3059\u3002<\/p>\n

\u8b1d\u8f9e<\/h3>\n

\u3053\u306e\u8a18\u4e8b\u306f\u3001\u30a4\u30f3\u30c6\u30eb\u30fb\u30e9\u30dc\u306e Ofir Zafrir\u3001Igor Margulis\u3001Guy Boudoukh\u3001Moshe Wasserblat \u3068\u5171\u540c\u3067\u4f5c\u6210\u3057\u307e\u3057\u305f\u3002\u7d20\u6674\u3089\u3057\u3044\u30b3\u30e1\u30f3\u30c8\u3068\u5354\u529b\u3092\u3044\u305f\u3060\u3044\u305f\u7686\u69d8\u306b\u611f\u8b1d\u3057\u307e\u3059\u3002<\/p>\n

\u4ed8\u9332: \u8a73\u7d30\u306a\u7d50\u679c<\/h3>\n

\u8ca0\u306e\u5024\u306f\u3001\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u304c\u6539\u5584\u3055\u308c\u305f\u3053\u3068\u3092\u793a\u3057\u307e\u3059\u3002<\/p>\n

\"\"<\/p>\n

\"\"<\/p>\n

\"\"<\/p>\n

\"\"<\/p>\n

\u95a2\u9023\u60c5\u5831:<\/p>\n

\u30a4\u30f3\u30c6\u30eb\u306e AI\/ML \u30a8\u30b3\u30b7\u30b9\u30c6\u30e0 – HuggingFace<\/a> (\u82f1\u8a9e)
\nHugging \u306e\u30a4\u30f3\u30c6\u30eb\u30fb\u30da\u30fc\u30b8<\/a> (\u82f1\u8a9e)<\/p>\n

\u53c2\u8003\u8cc7\u6599:<\/p>\n

<\/a>[1]<\/sup> https:\/\/arxiv.org\/abs\/2206.01861<\/a> (\u82f1\u8a9e)
\n<\/a>[2]<\/sup> https:\/\/arxiv.org\/abs\/2211.10438<\/a> (\u82f1\u8a9e)
\n<\/a>[3]<\/sup> https:\/\/arxiv.org\/abs\/2211.10438<\/a> (\u82f1\u8a9e)
\n<\/a>[4]<\/sup> https:\/\/github.com\/mit-han-lab\/smoothquant<\/a> (\u82f1\u8a9e)
\n<\/a>[5]<\/sup> https:\/\/arxiv.org\/pdf\/2205.01068.pdf<\/a> (\u82f1\u8a9e)
\n<\/a>[6]<\/sup> https:\/\/ai.meta.com\/blog\/large-language-model-llama-meta-ai\/<\/a> (\u82f1\u8a9e)
\n<\/a>[7]<\/sup> https:\/\/crfm.stanford.edu\/2023\/03\/13\/alpaca.html<\/a> (\u82f1\u8a9e)
\n<\/a>[8]<\/sup> https:\/\/vicuna.lmsys.org\/<\/a> (\u82f1\u8a9e)
\n<\/a>[9]<\/sup> https:\/\/huggingface.co\/bigscience\/bloomz<\/a> (\u82f1\u8a9e)
\n<\/a>[10]<\/sup> https:\/\/www.databricks.com\/blog\/mpt-7b<\/a> (\u82f1\u8a9e)<\/p>\n

\n

\u88fd\u54c1\u304a\u3088\u3073\u6027\u80fd\u306b\u95a2\u3059\u308b\u60c5\u5831<\/h4>\n

1<\/sup> \u6027\u80fd\u306f\u3001\u4f7f\u7528\u72b6\u6cc1\u3001\u69cb\u6210\u3001\u305d\u306e\u4ed6\u306e\u8981\u56e0\u306b\u3088\u3063\u3066\u7570\u306a\u308a\u307e\u3059\u3002\u8a73\u7d30\u306b\u3064\u3044\u3066\u306f\u3001http:\/\/www.intel.com\/PerformanceIndex\/<\/a> (\u82f1\u8a9e) \u3092\u53c2\u7167\u3057\u3066\u304f\u3060\u3055\u3044\u3002<\/p><\/p>\n\n\n\n

\n

\u6700\u65b0\u306e\u30a4\u30f3\u30c6\u30eb\u00ae Xeon\u00ae \u30b9\u30b1\u30fc\u30e9\u30d6\u30eb\u30fb\u30d7\u30ed\u30bb\u30c3\u30b5\u30fc\u5411\u3051\u306e\u6700\u9069\u5316\u3092\u652f\u63f4<\/h3>\n\n\n\n

\u30a8\u30af\u30bb\u30eb\u30bd\u30d5\u30c8\u306f\u3001\u6700\u65b0\u306e\u30a4\u30f3\u30c6\u30eb\u00ae Xeon\u00ae \u30b9\u30b1\u30fc\u30e9\u30d6\u30eb\u30fb\u30d7\u30ed\u30bb\u30c3\u30b5\u30fc\u3092\u30bf\u30fc\u30b2\u30c3\u30c8\u3068\u3059\u308b\u4f01\u696d\u3084\u958b\u767a\u30c1\u30fc\u30e0\u5411\u3051\u306b\u3001\u30a4\u30f3\u30c6\u30eb\u00ae \u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u958b\u767a\u30c4\u30fc\u30eb\u3092\u5229\u7528\u3057\u305f\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u306e\u6700\u9069\u5316\u652f\u63f4\u3092\u63d0\u4f9b\u3057\u3066\u3044\u307e\u3059\u3002\u30b3\u30f3\u30d1\u30a4\u30e9\u30fc\u30fb\u30aa\u30d7\u30b7\u30e7\u30f3\u306e\u63d0\u6848\u3001\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u5206\u6790\u30c4\u30fc\u30eb\u306b\u3088\u308b\u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3\u306e\u89e3\u6790\u3084\u30c1\u30e5\u30fc\u30cb\u30f3\u30b0\u306a\u3069\u3092\u901a\u3057\u3066\u3001\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3068\u958b\u767a\u52b9\u7387\u306e\u5411\u4e0a\u306b\u3088\u308b\u30b3\u30b9\u30c8\u524a\u6e1b\u3092\u652f\u63f4\u3057\u307e\u3059\u3002<\/p>\n\n\n\n

\u6700\u65b0\u306e\u30a4\u30f3\u30c6\u30eb\u00ae Xeon\u00ae \u30b9\u30b1\u30fc\u30e9\u30d6\u30eb\u30fb\u30d7\u30ed\u30bb\u30c3\u30b5\u30fc\u3092\u642d\u8f09\u3057\u305f\u30af\u30e9\u30a6\u30c9\u30b5\u30fc\u30d3\u30b9\u300c\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> Tiber™<\/sup> AI \u30af\u30e9\u30a6\u30c9\u300d\u306e\u5c0e\u5165\u30b5\u30dd\u30fc\u30c8\u3067\u306f\u3001LLM \u3084\u751f\u6210 AI \u3092\u6d3b\u7528\u3057\u305f AI \u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3\u3084\u30b5\u30fc\u30d3\u30b9\u306e\u958b\u767a\u304a\u3088\u3073\u904b\u7528\u306b\u643a\u308f\u308b\u4f01\u696d\u5411\u3051\u306b\u3001POC \u69cb\u7bc9\u3084\u6700\u9069\u306a\u4fa1\u683c\u30d7\u30e9\u30f3\u306e\u3054\u63d0\u6848\u304b\u3089\u56fd\u5185\u3067\u306e\u8cfc\u5165\u30b5\u30dd\u30fc\u30c8\u307e\u3067\u3001\u4f01\u696d\u306e\u30cb\u30fc\u30ba\u306b\u5408\u308f\u305b\u305f\u3092\u652f\u63f4\u3057\u307e\u3059\u3002\u307e\u305f\u3001AI \u30b9\u30bf\u30fc\u30c8\u30a2\u30c3\u30d7\u4f01\u696d\u306e\u80b2\u6210\u306b\u7279\u5316\u3057\u305f\u30a4\u30f3\u30c6\u30eb\u306e\u7121\u6599\u30d7\u30ed\u30b0\u30e9\u30e0\u306e\u63a8\u85a6\u306a\u3069\u3082\u884c\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n

\n
\n
\n
\u30af\u30e9\u30a6\u30c9\u30b5\u30fc\u30d3\u30b9\u5c0e\u5165\u652f\u63f4<\/strong><\/strong><\/a><\/div>\n<\/div>\n<\/div>\n\n\n\n
\n
\n
\u958b\u767a\u30c4\u30fc\u30eb\u5229\u7528\u652f\u63f4<\/strong><\/strong><\/a><\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div><\/div>\n\n\n\n
\n

\u30a4\u30f3\u30c6\u30eb\u00ae \u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u958b\u767a\u30c4\u30fc\u30eb2025 \u30ea\u30ea\u30fc\u30b9\u8a18\u5ff5\u30ad\u30e3\u30f3\u30da\u30fc\u30f3<\/h2>\n\n\n\n

\u30a4\u30f3\u30c6\u30eb\u00ae \u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u958b\u767a\u30c4\u30fc\u30eb\u306e\u6700\u65b0\u30d0\u30fc\u30b8\u30e7\u30f3 2025 \u306e\u30ea\u30ea\u30fc\u30b9\u3092\u8a18\u5ff5\u3057\u3066\u3001\u5bfe\u8c61\u306e\u30aa\u30f3\u30e9\u30a4\u30f3\u30a4\u30d9\u30f3\u30c8\u306b\u53c2\u52a0\u3044\u305f\u3060\u304d\u3001\u4e8b\u5f8c\u30a2\u30f3\u30b1\u30fc\u30c8\u306b\u3054\u56de\u7b54\u3044\u305f\u3060\u3044\u305f\u65b9\u306e\u4e2d\u304b\u3089\u3001\u62bd\u9078\u3067\u5408\u8a08 18 \u540d\u69d8\u306b\u30a4\u30f3\u30c6\u30eb Arc B580 \u30b0\u30e9\u30d5\u30a3\u30c3\u30af\u30b9\u306a\u3069\u304c\u5f53\u305f\u308b\u30d7\u30ec\u30bc\u30f3\u30c8\u30fb\u30ad\u30e3\u30f3\u30da\u30fc\u30f3\u3092\u5b9f\u65bd\u4e2d\u3067\u3059\u3002\u3010\u7de0\u5207 : 2025\u5e742\u670820\u65e5 (\u6728)\u3011<\/p>\n\n\n\n

\n
\n
\n
\u30ad\u30e3\u30f3\u30da\u30fc\u30f3\u306e\u8a73\u7d30\u306f\u3053\u3061\u3089<\/strong><\/a><\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div><\/div>\n\n\n\n

<\/p>\n","protected":false},"excerpt":{"rendered":"

\u3053\u306e\u8a18\u4e8b\u306f\u3001\u30a4\u30f3\u30c6\u30eb\u306e\u30a6\u30a7\u30d6\u30b5\u30a4\u30c8\u3067\u516c\u958b\u3055\u308c\u3066\u3044\u308b […]<\/a><\/p>\n<\/div>","protected":false},"author":1,"featured_media":3361,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[12,128,121],"class_list":["post-3347","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tech-tips","tag-ai","tag-llm","tag-xeon"],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/posts\/3347","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/comments?post=3347"}],"version-history":[{"count":13,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/posts\/3347\/revisions"}],"predecessor-version":[{"id":3376,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/posts\/3347\/revisions\/3376"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/media\/3361"}],"wp:attachment":[{"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/media?parent=3347"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/categories?post=3347"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/tags?post=3347"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}