{"id":307,"date":"2026-07-03T19:01:27","date_gmt":"2026-07-03T10:01:27","guid":{"rendered":"https:\/\/www.wetch.co.jp\/ai\/?p=307"},"modified":"2026-07-03T19:01:27","modified_gmt":"2026-07-03T10:01:27","slug":"%e3%83%ad%e3%83%bc%e3%82%ab%e3%83%abllm%e6%9c%80%e5%89%8d%e7%b7%9a%ef%bc%9aollama-v0-30-8%e3%81%aenvfp4%e9%87%8f%e5%ad%90%e5%8c%96%e3%80%81lm-studio%e3%81%aeiphone%e9%80%a3%e6%90%ba%e3%80%8clm-link","status":"publish","type":"post","link":"https:\/\/www.wetch.co.jp\/ai\/2026\/07\/03\/%e3%83%ad%e3%83%bc%e3%82%ab%e3%83%abllm%e6%9c%80%e5%89%8d%e7%b7%9a%ef%bc%9aollama-v0-30-8%e3%81%aenvfp4%e9%87%8f%e5%ad%90%e5%8c%96%e3%80%81lm-studio%e3%81%aeiphone%e9%80%a3%e6%90%ba%e3%80%8clm-link\/","title":{"rendered":"\u30ed\u30fc\u30ab\u30ebLLM\u6700\u524d\u7dda\uff1aOllama v0.30.8\u306eNVFP4\u91cf\u5b50\u5316\u3001LM Studio\u306eiPhone\u9023\u643a\u300cLM Link\u300d\u3001Liquid AI\u306e\u8d85\u5c0f\u578b\u30e2\u30c7\u30ebLFM2.5-230M"},"content":{"rendered":"<p><em>\u4eca\u9031\u306e\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u30fb\u30ed\u30fc\u30ab\u30ebLLM\u754c\u9688\u306f\u3001\u65b0\u3057\u3044\u5de8\u5927\u30e2\u30c7\u30eb\u306e\u767a\u8868\u3088\u308a\u3082\u300c\u4eca\u3042\u308b\u30e2\u30c7\u30eb\u3092\u3069\u3046\u901f\u304f\u30fb\u8efd\u304f\u30fb\u4fbf\u5229\u306b\u52d5\u304b\u3059\u304b\u300d\u3068\u3044\u3046\u63a8\u8ad6\u57fa\u76e4\u5074\u306e\u9032\u5316\u304c\u76ee\u7acb\u3063\u305f\u3002Ollama\u306f\u91cf\u5b50\u5316\u3068\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u5411\u3051\u30ad\u30e3\u30c3\u30b7\u30e5\u3092\u5f37\u5316\u3057\u3001LM Studio\u306f\u30b9\u30de\u30db\u304b\u3089PC\u306e\u30e2\u30c7\u30eb\u3092\u547c\u3073\u51fa\u305b\u308b\u3088\u3046\u306b\u3057\u3001Liquid AI\u306fRaspberry Pi\u3067\u3082\u52d5\u304f230M\u30d1\u30e9\u30e1\u30fc\u30bf\u30e2\u30c7\u30eb\u3067AI\u5bb6\u96fb\u30fb\u30ed\u30dc\u30c3\u30c8\u5411\u3051\u306e\u73fe\u5b9f\u7684\u306a\u9078\u629e\u80a2\u3092\u793a\u3057\u305f\u3002\u3042\u308f\u305b\u3066\u3001\u30ed\u30fc\u30ab\u30eb\u63a8\u8ad6\u306e\u5b9a\u756a\u30a8\u30f3\u30b8\u30f3\u3067\u3042\u308b<code>llama.cpp<\/code>\u3068<code>vLLM<\/code>\u306e\u4f7f\u3044\u5206\u3051\u3082\u6539\u3081\u3066\u6574\u7406\u3059\u308b\u3002<\/em><\/p>\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/www.wetch.co.jp\/ai\/wp-content\/uploads\/2026\/06\/ill-oss.png\" alt=\"\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u30fb\u30ed\u30fc\u30ab\u30ebLLM\u306e\u30a4\u30e1\u30fc\u30b8\u30a4\u30e9\u30b9\u30c8\"\/><\/figure>\n<h2>Ollama v0.30.8\uff1aMLX\u30a8\u30f3\u30b8\u30f3\u304cNVFP4\u91cf\u5b50\u5316\u3068\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u5411\u3051\u30b9\u30ca\u30c3\u30d7\u30b7\u30e7\u30c3\u30c8\u306b\u5bfe\u5fdc<\/h2>\n<p><a href=\"https:\/\/ollama.com\/\" target=\"_blank\" rel=\"noopener\">Ollama<\/a>\u306f<a href=\"https:\/\/ollama.com\/blog\/mlx-performance\" target=\"_blank\" rel=\"noopener\">2026\u5e746\u670811\u65e5\u306e\u30d6\u30ed\u30b0<\/a>\u3067\u3001Apple Silicon\u5411\u3051<a href=\"https:\/\/github.com\/ml-explore\/mlx\" target=\"_blank\" rel=\"noopener\">MLX<\/a>\u30a8\u30f3\u30b8\u30f3\u306e\u5927\u5e45\u5f37\u5316\u3092\u767a\u8868\u3057\u305f\u3002\u76ee\u7389\u306f<a href=\"https:\/\/developer.nvidia.com\/blog\/introducing-nvfp4-for-efficient-and-accurate-low-precision-inference\/\" target=\"_blank\" rel=\"noopener\">NVFP4<\/a>\u3068\u3044\u3046\u65b0\u3057\u30444bit\u91cf\u5b50\u5316\u30d5\u30a9\u30fc\u30de\u30c3\u30c8\u3078\u306e\u5bfe\u5fdc\u3067\u3001Google\u306e<a href=\"https:\/\/ai.google.dev\/gemma\" target=\"_blank\" rel=\"noopener\">Gemma<\/a> 4 12B\u30e2\u30c7\u30eb\u3067\u691c\u8a3c\u3057\u305f\u3068\u3053\u308d\u3001\u5f93\u6765\u306e<code>q4_K_M<\/code>\u5f62\u5f0f\u306b\u6bd4\u3079\u3066\u91cf\u5b50\u5316\u306b\u3088\u308b\u54c1\u8cea\u52a3\u5316\u304c\u304a\u3088\u305d\u534a\u5206\u306b\u6291\u3048\u3089\u308c\u305f\u3068\u3044\u3046\u3002\u30c7\u30fc\u30bf\u30bb\u30f3\u30bf\u30fc\u5411\u3051\u306b\u6700\u9069\u5316\u3055\u308c\u305fNVFP4\u30e2\u30c7\u30eb\u3092\u305d\u306e\u307e\u307e\u30c7\u30b9\u30af\u30c8\u30c3\u30d7\u306b\u6301\u3061\u8fbc\u3081\u308b\u305f\u3081\u3001\u30af\u30e9\u30a6\u30c9\u3068\u30ed\u30fc\u30ab\u30eb\u306e\u9593\u3067\u30e2\u30c7\u30eb\u3092\u884c\u304d\u6765\u3055\u305b\u3084\u3059\u304f\u306a\u308b\u3002<\/p>\n<p>\u52a0\u3048\u3066\u3001Metal\u30ab\u30fc\u30cd\u30eb\u306e\u878d\u5408\u3084GPU\u30b5\u30f3\u30d7\u30ea\u30f3\u30b0\u306e\u898b\u76f4\u3057\u306b\u3088\u308a\u51fa\u529b\u901f\u5ea6\u304c\u6700\u592720%\u5411\u4e0a\u3002\u3055\u3089\u306b\u6ce8\u76ee\u3059\u3079\u304d\u306f\u300c\u30b9\u30ca\u30c3\u30d7\u30b7\u30e7\u30c3\u30c8\u300d\u6a5f\u80fd\u3067\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u304c\u30b5\u30d6\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306b\u51e6\u7406\u3092\u59d4\u8b72\u3057\u305f\u308a\u3001\u8907\u6570\u30bb\u30c3\u30b7\u30e7\u30f3\u304c\u540c\u6642\u306b\u8d70\u3063\u305f\u308a\u3059\u308b\u5834\u9762\u3067\u3001\u5171\u901a\u3059\u308b\u30b7\u30b9\u30c6\u30e0\u30d7\u30ed\u30f3\u30d7\u30c8\u3084\u30c4\u30fc\u30eb\u5b9a\u7fa9\u3001\u8aad\u307f\u8fbc\u3093\u3060\u30d5\u30a1\u30a4\u30eb\u306e\u518d\u8a08\u7b97\u3092\u7701\u7565\u3067\u304d\u308b\u3088\u3046\u306b\u3057\u305f\u3002\u9577\u3044\u4f1a\u8a71\u306e\u9014\u4e2d\u3067\u5206\u5c90\u30fb\u518d\u751f\u6210\u304c\u8d77\u304d\u3066\u3082\u3001\u5206\u5c90\u70b9\u304b\u3089\u51e6\u7406\u3092\u518d\u958b\u3067\u304d\u308b\u70b9\u304c\u7279\u5fb4\u3060\u3002<\/p>\n<p>\u51fa\u5178: <a href=\"https:\/\/ollama.com\/blog\/mlx-performance\" target=\"_blank\" rel=\"noopener\">Ollama Blog<\/a><\/p>\n<h2>LM Studio\u300cLM Link\u300d\uff1a\u81ea\u5b85PC\u306e\u5927\u578b\u30ed\u30fc\u30ab\u30eb\u30e2\u30c7\u30eb\u3092iPhone\u304b\u3089\u4f7f\u3046<\/h2>\n<p><a href=\"https:\/\/lmstudio.ai\/\" target=\"_blank\" rel=\"noopener\">LM Studio<\/a>\u306f<a href=\"https:\/\/lmstudio.ai\/blog\/locally-lm-link\" target=\"_blank\" rel=\"noopener\">6\u67084\u65e5<\/a>\u3001\u30e2\u30d0\u30a4\u30eb\u30a2\u30d7\u30ea\u300c<a href=\"https:\/\/apps.apple.com\/us\/app\/locally-ai-local-ai-chat\/id6741426692\" target=\"_blank\" rel=\"noopener\">Locally<\/a>\u300d\u306b\u65b0\u6a5f\u80fd\u300cLM Link\u300d\u3092\u642d\u8f09\u3057\u305f\u3068\u767a\u8868\u3057\u305f\u3002\u81ea\u5b85\u3084\u30aa\u30d5\u30a3\u30b9\u306ePC\u3067\u52d5\u304b\u3057\u3066\u3044\u308b\u5927\u304d\u306a\u30ed\u30fc\u30ab\u30eb\u30e2\u30c7\u30eb\u306b\u3001\u5916\u51fa\u5148\u306eiPhone\u30fbiPad\u304b\u3089\u5b89\u5168\u306b\u63a5\u7d9a\u3057\u3066\u4f7f\u3048\u308b\u4ed5\u7d44\u307f\u3067\u3001\u901a\u4fe1\u306f\u7aef\u672b\u9593\u3067\u30a8\u30f3\u30c9\u30c4\u30fc\u30a8\u30f3\u30c9\u6697\u53f7\u5316\u3055\u308c\u3001\u30c1\u30e3\u30c3\u30c8\u5c65\u6b74\u3082\u5404\u7aef\u672b\u306b\u30ed\u30fc\u30ab\u30eb\u4fdd\u5b58\u3055\u308c\u308b\u3002<\/p>\n<p>\u300c\u30d5\u30ed\u30f3\u30c6\u30a3\u30a2\u7d1a\u306e\u63a8\u8ad6\u6027\u80fd\u3092\u81ea\u5206\u306e\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u3067\u4f7f\u3046\u300d\u3053\u3068\u81ea\u4f53\u306f\u73fe\u5b9f\u7684\u306b\u306a\u3063\u3066\u304d\u305f\u304c\u3001\u5e38\u306b\u305d\u306e\u5927\u578b\u30de\u30b7\u30f3\u306e\u305d\u3070\u306b\u3044\u3089\u308c\u308b\u308f\u3051\u3067\u306f\u306a\u3044\u3001\u3068\u3044\u3046\u8ab2\u984c\u306b\u5bfe\u3059\u308b\u56de\u7b54\u3060\u3002\u8a2d\u5b9a\u306fLM Studio\u5074\u3067LM Link\u3092\u6709\u52b9\u5316\u3057\u3001Locally\u30a2\u30d7\u30ea\u3067\u30da\u30a2\u30ea\u30f3\u30b0\u3059\u308b\u3060\u3051\u3068\u30b7\u30f3\u30d7\u30eb\u3002\u30af\u30e9\u30a6\u30c9API\u3092\u4ecb\u3055\u305a\u306b\u3001\u81ea\u5b85\u306e\u9ad8\u6027\u80fdGPU\u3092\u30b9\u30de\u30db\u304b\u3089\u9060\u9694\u5229\u7528\u3059\u308b\u767a\u60f3\u306f\u3001\u30d7\u30e9\u30a4\u30d0\u30b7\u30fc\u91cd\u8996\u306e\u30ed\u30fc\u30ab\u30ebLLM\u904b\u7528\u306b\u304a\u3051\u308b\u65b0\u3057\u3044\u5f62\u3068\u8a00\u3048\u308b\u3002<\/p>\n<p>\u51fa\u5178: <a href=\"https:\/\/lmstudio.ai\/blog\/locally-lm-link\" target=\"_blank\" rel=\"noopener\">LM Studio Blog<\/a><\/p>\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/www.wetch.co.jp\/ai\/wp-content\/uploads\/2026\/07\/raspberry-pi.jpg\" alt=\"Raspberry Pi\u306e\u57fa\u677f\"\/><figcaption><small>Photo: <a href=\"https:\/\/www.flickr.com\/photos\/45703688@N07\" target=\"_blank\" rel=\"noopener\">lespounder<\/a> \/ CC BY-SA 2.0\uff08Openverse\u7d4c\u7531\uff09<\/small><\/figcaption><\/figure>\n<h2>Liquid AI\u3001\u8d85\u5c0f\u578b\u30aa\u30f3\u30c7\u30d0\u30a4\u30b9\u30e2\u30c7\u30eb\u300cLFM2.5-230M\u300d\u3092\u516c\u958b<\/h2>\n<p><a href=\"https:\/\/www.liquid.ai\/\" target=\"_blank\" rel=\"noopener\">Liquid AI<\/a>\u306f\u81ea\u793e\u6700\u5c0f\u3068\u306a\u308b<a href=\"https:\/\/www.liquid.ai\/blog\/lfm2-5-230m\" target=\"_blank\" rel=\"noopener\">LFM2.5-230M<\/a>\u3092\u767a\u8868\u3057\u305f\u3002\u30d1\u30e9\u30e1\u30fc\u30bf\u6570\u306f\u308f\u305a\u304b2.3\u5104\u3060\u304c\u300119\u5146\u30c8\u30fc\u30af\u30f3\u3067\u4e8b\u524d\u5b66\u7fd2\u3057\u3001\u84b8\u7559\u30fbDPO\u30fb\u591a\u9818\u57df\u5f37\u5316\u5b66\u7fd2\u306e3\u6bb5\u968epost-training\u3092\u7d4c\u3066\u3044\u308b\u3002Raspberry Pi 5\u4e0a\u306742 tok\/s\u3001Samsung Galaxy S25 Ultra\uff08Snapdragon Gen4\uff09\u4e0a\u3067213 tok\/s\u306e\u30c7\u30b3\u30fc\u30c9\u901f\u5ea6\u3092\u5b9f\u73fe\u3057\u3001\u30e1\u30e2\u30ea\u4f7f\u7528\u91cf\u3082\u540c\u30af\u30e9\u30b9\u6700\u5c0f\u3068\u3044\u3046\u3002\u30d9\u30fc\u30b9\u7248\u30fbpost-trained\u7248\u3068\u3082<a href=\"https:\/\/huggingface.co\/LiquidAI\/LFM2.5-230M\" target=\"_blank\" rel=\"noopener\">Hugging Face<\/a>\u3067\u516c\u958b\u6e08\u307f\u3060\u3002<\/p>\n<p>\u30c4\u30fc\u30eb\u547c\u3073\u51fa\u3057(BFCL)\u3084\u30c7\u30fc\u30bf\u62bd\u51fa(CaseReportBench)\u306e\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3067\u306f\u3001<a href=\"https:\/\/ai.google.dev\/gemma\" target=\"_blank\" rel=\"noopener\">Gemma<\/a> 3 1B IT\u3084<a href=\"https:\/\/huggingface.co\/Qwen\" target=\"_blank\" rel=\"noopener\">Qwen<\/a>3.5-0.8B\u306a\u30692\u500d\u4ee5\u4e0a\u306e\u30b5\u30a4\u30ba\u306e\u30e2\u30c7\u30eb\u3092\u4e0a\u56de\u308b\u5834\u9762\u3082\u3042\u3063\u305f\u3002Liquid AI\u306fNVIDIA Jetson Orin\u642d\u8f09\u306eUnitree G1\u30d2\u30e5\u30fc\u30de\u30ce\u30a4\u30c9\u30ed\u30dc\u30c3\u30c8\u306b\u672c\u30e2\u30c7\u30eb\u3092\u7d44\u307f\u8fbc\u307f\u3001\u81ea\u7136\u8a00\u8a9e\u306e\u6307\u793a\u3092\u6b69\u884c\u30fb\u7247\u819d\u7acb\u3061\u306a\u3069\u306e\u30b9\u30ad\u30eb\u547c\u3073\u51fa\u3057\u306b\u5909\u63db\u3059\u308b\u30c7\u30e2\u3082\u516c\u958b\u3057\u3066\u3044\u308b\u3002\u6570\u5b66\u3084\u30b3\u30fc\u30c9\u751f\u6210\u306a\u3069\u306e\u9ad8\u5ea6\u306a\u63a8\u8ad6\u306b\u306f\u5411\u304b\u306a\u3044\u3068\u3057\u3064\u3064\u3001\u30a8\u30c3\u30b8\u6a5f\u5668\u3084\u30ed\u30dc\u30c3\u30c8\u306e\u300c\u8a00\u8a9e\u7406\u89e3\u30a4\u30f3\u30bf\u30fc\u30d5\u30a7\u30fc\u30b9\u5c64\u300d\u3068\u3057\u3066\u306e\u7528\u9014\u3092\u660e\u78ba\u306b\u6253\u3061\u51fa\u3057\u305f\u683c\u597d\u3060\u3002<\/p>\n<p>\u51fa\u5178: <a href=\"https:\/\/www.liquid.ai\/blog\/lfm2-5-230m\" target=\"_blank\" rel=\"noopener\">Liquid AI Blog<\/a><\/p>\n<h2>\u63a8\u8ad6\u30a8\u30f3\u30b8\u30f3\u306e\u9078\u3073\u65b9\uff1allama.cpp \u3068 vLLM \u306f\u3069\u3046\u9055\u3046\u306e\u304b<\/h2>\n<p>\u30ed\u30fc\u30ab\u30ebLLM\u306e\u571f\u53f0\u3068\u306a\u308b\u63a8\u8ad6\u30a8\u30f3\u30b8\u30f3\u306b\u3064\u3044\u3066\u3001<a href=\"https:\/\/developers.redhat.com\/articles\/2026\/06\/15\/llamacpp-vs-vllm-choosing-right-local-llm-inference-engine\" target=\"_blank\" rel=\"noopener\">Red Hat Developers\u306e\u89e3\u8aac\u8a18\u4e8b<\/a>\u304c\u6539\u3081\u3066\u4e21\u8005\u306e\u9055\u3044\u3092\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u4ed8\u304d\u3067\u6574\u7406\u3057\u3066\u3044\u308b\u3002<a href=\"https:\/\/github.com\/ggml-org\/llama.cpp\" target=\"_blank\" rel=\"noopener\">llama.cpp<\/a>\u306fCPU\u30d5\u30a1\u30fc\u30b9\u30c8\u306e\u8a2d\u8a08\u3067\u3001GGUF\u5f62\u5f0f\u306b\u3088\u308b4bit\u91cf\u5b50\u5316\u3092\u6b66\u5668\u306b\u3001GPU\u306a\u3057\u306e\u4e00\u822c\u7684\u306aPC\u3067\u3082\u30e2\u30c7\u30eb\u3092\u52d5\u304b\u305b\u308b\u3088\u3046\u306b\u3057\u305f\u7acb\u5f79\u8005\u3002Ollama\u3084LM Studio\u3082\u5185\u90e8\u30a8\u30f3\u30b8\u30f3\u3068\u3057\u3066llama.cpp\u3092\u4f7f\u3063\u3066\u3044\u308b\u3002\u4e00\u65b9<a href=\"https:\/\/github.com\/vllm-project\/vllm\" target=\"_blank\" rel=\"noopener\">vLLM<\/a>\u306f\u3001PagedAttention\u306b\u3088\u308bKV\u30ad\u30e3\u30c3\u30b7\u30e5\u7ba1\u7406\u3068\u9023\u7d9a\u30d0\u30c3\u30c1\u51e6\u7406\u3067\u3001\u30c7\u30fc\u30bf\u30bb\u30f3\u30bf\u30fcGPU\u4e0a\u3067\u306e\u591a\u6570\u540c\u6642\u30e6\u30fc\u30b6\u30fc\u5bfe\u5fdc\u306b\u5f37\u307f\u3092\u6301\u3064\u3002<\/p>\n<table>\n<thead>\n<tr>\n<th>\u6bd4\u8f03\u8ef8<\/th>\n<th>llama.cpp\uff08Ollama\/LM Studio\u7b49\uff09<\/th>\n<th>vLLM<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u60f3\u5b9a\u74b0\u5883<\/td>\n<td>\u30ce\u30fc\u30c8PC\u30fbGPU\u306a\u3057\u3067\u3082OK<\/td>\n<td>\u30c7\u30fc\u30bf\u30bb\u30f3\u30bf\u30fcGPU\uff08A100\/H100\u7b49\uff09<\/td>\n<\/tr>\n<tr>\n<td>\u30e2\u30c7\u30eb\u5f62\u5f0f<\/td>\n<td>GGUF\uff08\u5358\u4e00\u30d5\u30a1\u30a4\u30eb\u3001\u91cf\u5b50\u5316\u524d\u63d0\uff09<\/td>\n<td>safetensors\uff0bFP8\/INT8\/INT4\u91cf\u5b50\u5316<\/td>\n<\/tr>\n<tr>\n<td>\u5f97\u610f\u306a\u7528\u9014<\/td>\n<td>\u500b\u4eba\u5229\u7528\u30fb\u30d7\u30ed\u30c8\u30bf\u30a4\u30d4\u30f3\u30b0\u30fb\u30aa\u30d5\u30e9\u30a4\u30f3<\/td>\n<td>\u591a\u6570\u540c\u6642\u30e6\u30fc\u30b6\u30fc\u3078\u306e\u672c\u756a\u63d0\u4f9b<\/td>\n<\/tr>\n<tr>\n<td>\u540c\u664264\u30e6\u30fc\u30b6\u30fc\u6642\u306e\u30b9\u30eb\u30fc\u30d7\u30c3\u30c8<\/td>\n<td>\u57fa\u6e96<\/td>\n<td>\u7d0444\u500d\uff08Red Hat\u8a08\u6e2c\uff09<\/td>\n<\/tr>\n<tr>\n<td>API\u4e92\u63db\u6027<\/td>\n<td>OpenAI\u4e92\u63db\u30a8\u30f3\u30c9\u30dd\u30a4\u30f3\u30c8<\/td>\n<td>OpenAI\u4e92\u63db\u30a8\u30f3\u30c9\u30dd\u30a4\u30f3\u30c8<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u8a18\u4e8b\u306f\u300c\u307e\u305a\u30af\u30e9\u30a6\u30c9API\u3067\u8a66\u4f5c\u3057\u3001\u30b3\u30b9\u30c8\u304c\u6c17\u306b\u306a\u3063\u305f\u3089llama.cpp\u3067\u30ed\u30fc\u30ab\u30eb\u958b\u767a\u3001\u672c\u756a\u5c55\u958b\u3067vLLM\u306b\u79fb\u884c\u3059\u308b\u300d\u3068\u3044\u3046\u5178\u578b\u7684\u306a\u79fb\u884c\u30d1\u30b9\u3092\u7d39\u4ecb\u3057\u3066\u3044\u308b\u3002\u3069\u3061\u3089\u3082OpenAI\u4e92\u63dbAPI\u3092\u63d0\u4f9b\u3059\u308b\u305f\u3081\u3001\u30a8\u30f3\u30c9\u30dd\u30a4\u30f3\u30c8\u306eURL\u3092\u5dee\u3057\u66ff\u3048\u308b\u3060\u3051\u3067\u79fb\u884c\u3067\u304d\u308b\u70b9\u3082\u5b9f\u52d9\u4e0a\u306e\u30dd\u30a4\u30f3\u30c8\u3060\u3002<\/p>\n<p>\u51fa\u5178: <a href=\"https:\/\/developers.redhat.com\/articles\/2026\/06\/15\/llamacpp-vs-vllm-choosing-right-local-llm-inference-engine\" target=\"_blank\" rel=\"noopener\">Red Hat Developers<\/a><\/p>\n<hr \/>\n<h2>\u4e3b\u306a\u60c5\u5831\u6e90<\/h2>\n<ul>\n<li><a href=\"https:\/\/ollama.com\/blog\/mlx-performance\" target=\"_blank\" rel=\"noopener\">Ollama Blog: Ollama&#8217;s highest performance on Apple Silicon yet with MLX<\/a><\/li>\n<li><a href=\"https:\/\/lmstudio.ai\/blog\/locally-lm-link\" target=\"_blank\" rel=\"noopener\">LM Studio Blog: Run (your largest) local models from your iPhone<\/a><\/li>\n<li><a href=\"https:\/\/www.liquid.ai\/blog\/lfm2-5-230m\" target=\"_blank\" rel=\"noopener\">Liquid AI Blog: LFM2.5-230M: Built to Run Anywhere<\/a><\/li>\n<li><a href=\"https:\/\/developers.redhat.com\/articles\/2026\/06\/15\/llamacpp-vs-vllm-choosing-right-local-llm-inference-engine\" target=\"_blank\" rel=\"noopener\">Red Hat Developers: llama.cpp vs. vLLM<\/a><\/li>\n<\/ul>\n<p><small>\u5404\u30b5\u30fc\u30d3\u30b9\u306f\u63d0\u4f9b\u5143\u306e\u898f\u7d04\u306b\u5f93\u3063\u3066\u3054\u5229\u7528\u304f\u3060\u3055\u3044\u3002\u60c5\u5831\u306f\u5404\u60c5\u5831\u6e90\u306e\u516c\u958b\u6642\u70b9\u306e\u3082\u306e\u3067\u3059\u3002\u53ce\u96c6\u30fb\u8981\u7d04\u306fAI\u3001\u516c\u958b\u5224\u65ad\u306f\u4eba\u3002<\/small><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Ollama\u306eMLX\u30a8\u30f3\u30b8\u30f3\u304cNVFP4\u91cf\u5b50\u5316\u3068\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u5411\u3051\u30b9\u30ca\u30c3\u30d7\u30b7\u30e7\u30c3\u30c8\u306b\u5bfe\u5fdc\u3001LM Studio\u306f\u81ea\u5b85PC\u306e\u30e2\u30c7\u30eb\u3092iPhone\u304b\u3089\u4f7f\u3048\u308b\u300cLM Link\u300d\u3092\u516c\u958b\u3002Liquid AI\u306fRaspberry Pi\u3067\u3082\u52d5\u304f230M\u30d1\u30e9\u30e1\u30fc\u30bf\u306eLFM2.5\u3092\u767a\u8868\u3057\u3001llama.cpp\u3068vLLM\u306e\u4f7f\u3044\u5206\u3051\u3082\u6574\u7406\u3059\u308b\u3002<\/p>\n","protected":false},"author":1,"featured_media":58,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"page_builder":"","footnotes":""},"categories":[46],"tags":[],"class_list":["post-307","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-aillm"],"_links":{"self":[{"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/posts\/307","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/comments?post=307"}],"version-history":[{"count":1,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/posts\/307\/revisions"}],"predecessor-version":[{"id":316,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/posts\/307\/revisions\/316"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/media\/58"}],"wp:attachment":[{"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/media?parent=307"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/categories?post=307"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/tags?post=307"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}