{"categories":[],"author_name":"Henry_Lee","height":"190","blog_url":"https://henry-lee-genai-fullstack.hatenablog.com/","author_url":"https://blog.hatena.ne.jp/Henry_Lee/","published":"2026-03-05 12:57:29","description":"\u8fd9\u4efd\u7b14\u8bb0\u6765\u81ea\u4e00\u6b21\u771f\u5b9e\u7684\u8bad\u7ec3\u5b9e\u8df5\uff1a\u5728 RTX 3070 (8GB VRAM) + 16GB RAM \u4e0a\uff0c\u4ece\u96f6\u5b9e\u73b0\u5e76\u8bad\u7ec3\u4e86\u4e00\u4e2a ~10M \u53c2\u6570\u7684 GPT \u8bed\u8a00\u6a21\u578b\u3002\u5305\u542b\u5b8c\u6574\u4ee3\u7801\u3001\u8e29\u8fc7\u7684\u5751\u3001\u8bad\u7ec3\u6570\u636e\uff0c\u9002\u5408\u4f5c\u4e3a\u6df1\u5ea6\u5b66\u4e60\u5bf9\u8bdd\u7684\u4e0a\u4e0b\u6587\u3002 \u4e00\u3001GPT \u662f\u4ec0\u4e48\uff1f\u4e00\u53e5\u8bdd\u7248\u672c GPT = \u53ea\u6709 Decoder \u7684 Transformer\uff0c\u4efb\u52a1\u662f\"\u7ed9\u5b9a\u524d\u9762\u7684\u6587\u5b57\uff0c\u9884\u6d4b\u4e0b\u4e00\u4e2a\u5b57\"\uff08\u81ea\u56de\u5f52\u8bed\u8a00\u6a21\u578b\uff09\u3002 \u8bad\u7ec3\u65f6\uff1a\u5582\u4e00\u6bb5\u6587\u672c\uff0c\u6a21\u578b\u5c1d\u8bd5\u9884\u6d4b\u6bcf\u4e2a\u4f4d\u7f6e\u7684\u4e0b\u4e00\u4e2a token\uff0c\u7528 cross-entropy loss \u8861\u91cf\u9884\u6d4b\u597d\u574f\u3002 \u751f\u6210\u65f6\uff1a\u7ed9\u4e00\u4e2a\u5f00\u5934\uff0c\u6a21\u578b\u4e00\u4e2a\u5b57\u4e00\u4e2a\u5b57\u5f80\u540e\"\u7eed\u5199\"\u3002 \u4e8c\u3001\u6838\u5fc3\u67b6\u6784\uff08\u9010\u5c42\u62c6\u89e3\uff09 2.1 \u6574\u4f53\u7ed3\u6784 \u8f93\u5165\u2026","title":"\u4ece\u96f6\u8bad\u7ec3\u4e00\u4e2a Mini GPT \u2014 \u5b9e\u6218\u5b66\u4e60\u7b14\u8bb0","width":"100%","url":"https://henry-lee-genai-fullstack.hatenablog.com/entry/2026/03/05/125729","html":"<iframe src=\"https://hatenablog-parts.com/embed?url=https%3A%2F%2Fhenry-lee-genai-fullstack.hatenablog.com%2Fentry%2F2026%2F03%2F05%2F125729\" title=\"\u4ece\u96f6\u8bad\u7ec3\u4e00\u4e2a Mini GPT \u2014 \u5b9e\u6218\u5b66\u4e60\u7b14\u8bb0 - Levis&#39;s GenAI Fullstack Engineer Blog\" 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","blog_title":"Levis's GenAI Fullstack Engineer Blog","provider_url":"https://hatena.blog","image_url":null,"provider_name":"Hatena Blog","type":"rich"}