{"published":"2025-08-15 06:01:52","width":"100%","title":"\u4eca\u65e5\u306e\u5de1\u56de","provider_name":"Hatena Blog","image_url":null,"categories":[],"html":"<iframe src=\"https://hatenablog-parts.com/embed?url=https%3A%2F%2Fginzanomama.hatenablog.com%2Fentry%2F2025%2F08%2F15%2F060152\" title=\"\u4eca\u65e5\u306e\u5de1\u56de - \u307f\u3093\u306a\u306e\u3061\u304b\u3089\u306b\u306a\u308a\u305f\u3044\" class=\"embed-card embed-blogcard\" scrolling=\"no\" frameborder=\"0\" style=\"display: block; width: 100%; height: 190px; max-width: 500px; margin: 10px 0px;\"></iframe>","version":"1.0","author_name":"ginzanomama","author_url":"https://blog.hatena.ne.jp/ginzanomama/","blog_title":"\u307f\u3093\u306a\u306e\u3061\u304b\u3089\u306b\u306a\u308a\u305f\u3044","description":"1. GPU\u30e1\u30e2\u30ea4GB\u3042\u308c\u3070GPT-oss 20B\u304c14tok/sec\u3067\u52d5\u304f LM Studio\u306e\u65b0\u6a5f\u80fd\u306b\u3088\u308a\u3001MoE\u30e2\u30c7\u30eb\u306eGPT-oss\u304c\u5c11\u306a\u3044GPU\u30e1\u30e2\u30ea\u3067\u3082\u52d5\u4f5c\u53ef\u80fd\u306b\u3002CPU/GPU\u306e\u632f\u308a\u5206\u3051\u6700\u9069\u5316\u30674GB\u306eVRAM\u3067\u308214tok/sec\u3092\u5b9f\u73fe\u3002 \u4e3b\u306a\u30c8\u30d4\u30c3\u30af: MoE\uff08Mixture of Experts\uff09: Expert\u30a6\u30a7\u30a4\u30c8\u3092CPU\u306b\u914d\u7f6e\u3057\u3066VRAM\u4f7f\u7528\u91cf\u3092\u524a\u6e1b LM Studio v0.3.23: Force Model Expert weight onto CPU\u30b9\u30a4\u30c3\u30c1\u306e\u8ffd\u52a0 llama.cpp: --n-cpu-moe\u30aa\u30d7\u30b7\u30e7\u30f3\u306b\u3088\u308bCPU/GPU\u632f\u308a\u5206\u3051\u6700\u9069\u5316 \u30a2\u30c6\u2026","type":"rich","url":"https://ginzanomama.hatenablog.com/entry/2025/08/15/060152","blog_url":"https://ginzanomama.hatenablog.com/","provider_url":"https://hatena.blog","height":"190"}