{"id":2893,"date":"2026-03-08T14:13:31","date_gmt":"2026-03-08T06:13:31","guid":{"rendered":"https:\/\/www.starverse-ai.com\/guide\/archives\/2893"},"modified":"2026-03-08T14:13:31","modified_gmt":"2026-03-08T06:13:31","slug":"%e4%bb%8ellama-3%e5%88%b0codellama%ef%bc%8c2026%e5%bc%80%e6%ba%90%e5%a4%a7%e6%a8%a1%e5%9e%8b%e6%9c%88%e6%9b%b4%e6%97%b6%e4%bb%a3%ef%bc%8c%e6%98%9f%e5%ae%87%e6%99%ba%e7%ae%97%e5%b8%ae%e4%bd%a0","status":"publish","type":"post","link":"https:\/\/www.starverse-ai.com\/guide\/archives\/2893","title":{"rendered":"\u4eceLlama-3\u5230CodeLlama+\uff0c2026\u5f00\u6e90\u5927\u6a21\u578b\u6708\u66f4\u65f6\u4ee3\uff0c\u661f\u5b87\u667a\u7b97\u5e2e\u4f60\u300c\u4e00\u952e\u4e0d\u6389\u961f\u300d"},"content":{"rendered":"<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/www.starverse-ai.com\/guide\/wp-content\/uploads\/2026\/03\/1772950411_37f308.png\" alt=\"\u4eceLlama-3\u5230CodeLlama+\uff0c2026\u5f00\u6e90\u5927\u6a21\u578b\u6708\u66f4\u65f6\u4ee3\uff0c\u661f\u5b87\u667a\u7b97\u5e2e\u4f60\u300c\u4e00\u952e\u4e0d\u6389\u961f\u300d\" style=\"display:block; margin:10px auto; max-width:100%; height:auto;\" \/><\/figure>\n<blockquote>\n<p>2026 \u5e74 4 \u6708\uff0cHuggingFace Trending \u699c\u518d\u6b21\u88ab\u5237\u65b0\uff1aLlama-3-70B-Instruct \u4e0a\u7ebf\u4ec5 18 \u5c0f\u65f6\u6536\u83b7 8.3k Star\uff0c\u7d27\u63a5\u7740 CodeLlama+ 34B Python \u4e13\u9879\u7248\u8fde\u591c\u8ddf\u8fdb\u3002\u5f00\u6e90\u793e\u533a\u7528\u201c\u6708\u66f4\u201d\u5f62\u5bb9\u5df2\u663e\u4fdd\u5b88\u2014\u2014\u8fc7\u53bb 30 \u5929\uff0c\u4ec5 10B \u4ee5\u4e0a\u53c2\u6570\u91cf\u7684\u65b0\u6a21\u578b\u5c31\u91ca\u51fa 7 \u4e2a\uff0c\u5e73\u5747\u6bcf\u5468 1.6 \u4e2a\u3002\u5f53\u7b97\u6cd5\u8fed\u4ee3\u901f\u5ea6\u4ee5\u201c\u5929\u201d\u4e3a\u5355\u4f4d\uff0c\u5f00\u53d1\u8005\u5374\u4ecd\u5728\u4e3a CUDA \u9a71\u52a8\u51b2\u7a81\u3001Transformers \u7248\u672c\u6f02\u79fb\u3001PCIe \u5e26\u5bbd\u74f6\u9888\u8fd9\u4e9b\u201c\u8001\u6bdb\u75c5\u201d\u53cd\u590d\u52a0\u73ed\u3002\u7b97\u529b\u4e0e\u6a21\u578b\u4e4b\u95f4\u7684\u9e3f\u6c9f\uff0c\u6b63\u5728\u6210\u4e3a AI \u521b\u65b0\u7684\u6700\u5927\u51cf\u901f\u5e26\u3002<\/p>\n<\/blockquote>\n<h2>\u5f00\u6e90\u72c2\u6b22\u80cc\u540e\u7684\u9690\u5f62\u52a0\u73ed<\/h2>\n<p>\u201c\u53c8\u8981\u91cd\u65b0\u642d\u73af\u5883\uff1f\u201d\u2014\u2014\u8fd9\u662f\u5b9e\u9a8c\u5ba4\u91cc\u6700\u5e38\u542c\u5230\u7684\u4e00\u53e5\u5410\u69fd\u3002<br \/>\n\u672c\u5730\u5de5\u4f5c\u7ad9\u597d\u4e0d\u5bb9\u6613\u628a torch 2.3 \u4e0e CUDA 12.2 \u5bf9\u9f50\uff0c\u7b2c\u4e8c\u5929\u65b0\u6a21\u578b\u5374\u8981\u6c42 flash-attention 2.5+\uff1b\u60f3\u56de\u9000\u9a71\u52a8\uff0c\u65e7\u9879\u76ee\u53c8\u62a5\u9519\u3002\u66f4\u5c34\u5c2c\u7684\u662f\uff0c34B \u6a21\u578b\u6743\u91cd\u521a\u62c9\u5230\u4e00\u534a\uff0cNVMe \u76d8\u544a\u6025\uff0c\u4e0b\u8f7d\u6d41\u91cf\u8fd8\u8d70\u7684\u662f 1 \u5143\/GB \u7684\u516c\u7f51\u5e26\u5bbd\u3002\u5982\u6b64\u53cd\u590d\uff0c\u7b97\u6cd5\u5de5\u7a0b\u5e08\u7684\u4e00\u534a\u7cbe\u529b\u88ab\u201c\u8fd0\u7ef4\u201d\u541e\u566c\uff0c\u8bba\u6587 deadline \u5374\u88ab\u8001\u677f\u7262\u7262\u9489\u6b7b\u3002<\/p>\n<h2>\u628a\u201c\u6708\u66f4\u201d\u538b\u529b\u5916\u5305\u7ed9\u4e91<\/h2>\n<p>\u6709\u6ca1\u6709\u4e00\u79cd\u65b9\u5f0f\uff0c\u8ba9\u6a21\u578b\u53d1\u5e03\u4e0e\u672c\u5730\u5f00\u53d1\u201c\u96f6\u65f6\u5dee\u201d\uff1f<br \/>\n\u661f\u5b87\u667a\u7b97\u7ed9\u51fa\u7684\u7b54\u6848\u662f\uff1a\u628a GPU\u670d\u52a1\u5668\u79df\u7528 \u505a\u6210\u201c\u70ed\u66f4\u65b0\u201d\u6a21\u5f0f\u3002\u5e73\u53f0\u5728 HuggingFace \u5b98\u65b9 Repo \u6253\u4e0a Tag \u7684 5 \u5206\u949f\u5185\uff0c\u5373\u5b8c\u6210\u5bb9\u5668\u955c\u50cf\u5c01\u88c5\u5e76\u63a8\u9001\u81f3\u300cAI\u5e94\u7528\u300d\u9891\u9053\u3002\u7528\u6237\u53ea\u9700\u70b9\u51fb\u201c\u7acb\u5373\u4f53\u9a8c\u201d\uff0c\u7cfb\u7edf\u4fbf\u81ea\u52a8\u62c9\u8d77\u4e00\u53f0\u9884\u88c5\u6700\u65b0\u7248\u6a21\u578b\u3001\u4f9d\u8d56\u3001\u52a0\u901f\u5e93\u7684 GPU\u4e91\u4e3b\u673a\uff0c\u65e0\u9700\u5173\u5fc3\u9a71\u52a8\u3001CUDA\u3001Python \u7248\u672c\u3002\u7b2c\u4e00\u6b21\u4f7f\u7528\u8fd8\u80fd\u9886\u53d6 10 \u5143\u4f53\u9a8c\u91d1\uff0c0 \u6210\u672c\u8dd1\u901a 70B \u5927\u6a21\u578b\u4e0d\u518d\u662f\u53e3\u53f7\u3002<\/p>\n<h2>15 \u5206\u949f\u5b9e\u6d4b\uff1aCodeLlama+ 34B \u5fae\u8c03\u5168\u7eaa\u5f55<\/h2>\n<p>\u4e3a\u4e86\u9a8c\u8bc1\u201c\u70ed\u66f4\u65b0\u201d\u662f\u5426\u5671\u5934\uff0c\u6211\u4eec\u505a\u4e00\u6b21\u5168\u6d41\u7a0b\u8ba1\u65f6\u3002<br \/>\n\u2460 00:00\u2014\u6ce8\u518c\u5e76\u5b9e\u540d\u8ba4\u8bc1\uff0c\u65b0\u7528\u6237\u5230\u8d26 10 \u5143\u4f53\u9a8c\u91d1\uff1b<br \/>\n\u2461 01:20\u2014\u8fdb\u5165\u300cAI\u5e94\u7528\u300d\u9891\u9053\uff0c\u9009\u62e9\u201cCodeLlama+ 34B Python\u201d\u955c\u50cf\uff0c\u5b9e\u4f8b\u89c4\u683c 8\u00d7RTX 4090\uff0c\u6309\u91cf\u8ba1\u8d39 1.9 \u5143\/\u5c0f\u65f6\uff1b<br \/>\n\u2462 02:50\u2014\u5b9e\u4f8b\u542f\u52a8\u5b8c\u6210\uff0c\u6d4f\u89c8\u5668\u81ea\u52a8\u5f39\u51fa JupyterLab\uff0c\u7cfb\u7edf\u76d8\u5df2\u5185\u7f6e\u6a21\u578b\u6743\u91cd\u3001DeepSpeed\u3001LoRA \u811a\u672c\uff1b<br \/>\n\u2463 05:10\u2014\u6302\u8f7d\u5e73\u53f0\u8d60\u9001\u7684 20G \u9ad8\u8d28\u91cf\u4e2d\u6587\u6307\u4ee4\u6570\u636e\u96c6\uff0c\u8def\u5f84 <code>\/datasets\/code_zh_120k<\/code> \u4e00\u952e\u53ef\u8bfb\u5199\uff1b<br \/>\n\u2464 08:40\u2014\u8fd0\u884c <code>train_lora.py<\/code>\uff0cbatch_size=4\uff0cgradient_accumulation_steps=8\uff0c\u663e\u5b58\u5360\u7528 4\u00d722GB\uff1b<br \/>\n\u2465 14:30\u2014\u9a8c\u8bc1\u96c6 Rouge-L \u4ece 20.1 \u63d0\u5347\u81f3 28.7\uff0c\u8bad\u7ec3\u7ed3\u675f\uff0c\u603b\u8017\u65f6 15 \u5206 10 \u79d2\u3002  <\/p>\n<p>\u6574\u4e2a\u8fc7\u7a0b\u6ca1\u6709\u4e00\u6b21 SSH \u9ed1\u7a97\uff0c\u4e5f\u65e0\u9700\u624b\u52a8\u8c03\u6574 NCCL_P2P\uff0c\u771f\u6b63\u505a\u5230\u4e86\u201c\u5f00\u7bb1\u5373\u8bad\u201d\u3002<\/p>\n<h2>\u4e3a\u4ec0\u4e48\u80fd\u8fd9\u4e48\u5feb\uff1f<\/h2>\n<ol>\n<li>\u9884\u88c5\u9a71\u52a8\u4e0e CUDA \u9759\u6001\u7ed1\u5b9a\uff1a\u955c\u50cf\u6784\u5efa\u9636\u6bb5\u5c31\u628a 535.54.03 \u9a71\u52a8\u4e0e CUDA 12.2.x \u5c01\u88c5\u5728\u4e00\u8d77\uff0c\u907f\u514d\u201c\u9a71\u52a8\u5347\u7ea7\u2014\u5bb9\u5668\u91cd\u542f\u201d\u6b7b\u5faa\u73af\u3002  <\/li>\n<li>\u6743\u91cd\u9884\u70ed\u673a\u5236\uff1a\u70ed\u95e8\u6a21\u578b\u63d0\u524d\u7f13\u5b58\u5230\u5185\u7f51 NAS\uff0c\u7528\u6237\u70b9\u51fb\u521b\u5efa\u65f6\u8d70 10 Gb\/s \u79c1\u7f51\u62c9\u53d6\uff0c\u5e26\u5bbd\u6210\u672c\u5e73\u53f0\u627f\u62c5\u3002  <\/li>\n<li>\u6301\u4e45\u5316\u4e91\u5b58\u50a8\uff1a\u8bad\u7ec3\u5f97\u5230\u7684 LoRA \u6743\u91cd\u5b9e\u65f6\u5199\u5165 <code>\/my-workspace<\/code>\uff0c\u5b9e\u4f8b\u91ca\u653e\u540e\u6570\u636e\u4ecd\u4fdd\u7559\uff0c\u4e0b\u6b21\u5f00\u65b0 GPU\u4e91\u4e3b\u673a \u79d2\u7ea7\u6302\u8f7d\u3002  <\/li>\n<li>\u5f39\u6027\u8ba1\u8d39\uff1a\u8dd1\u4efb\u52a1\u65f6\u9009\u201c\u6309\u91cf\u201d\uff0c\u8c03\u8bd5\u9636\u6bb5\u7528\u201c\u65e0 GPU \u542f\u52a8\u201d\u6a21\u5f0f\uff0c\u8d39\u7528\u76f4\u63a5\u964d\u5230 0.1 \u5143\/\u5c0f\u65f6\uff0c\u6210\u672c\u53ef\u63a7\u3002  <\/li>\n<\/ol>\n<h2>\u4ece\u201c\u80fd\u7528\u201d\u5230\u201c\u597d\u7528\u201d\u7684\u751f\u6001\u52a0\u6cd5<\/h2>\n<p>\u661f\u5b87\u667a\u7b97\u4e0d\u6b62\u4e8e GPU\u670d\u52a1\u5668\u79df\u7528\u3002<br \/>\n\u5728\u5e73\u53f0\u8d44\u6e90\u6c60\u91cc\uff0c\u4f60\u80fd\u627e\u5230 1600+ \u516c\u5171\u6a21\u578b\u3001300+ \u9ad8\u8d28\u91cf\u6570\u636e\u96c6\uff0c\u4ee5\u53ca Stable Diffusion\u3001ComfyUI\u3001Text-Generation-WebUI \u7b49 40 \u4f59\u6b3e\u4e00\u952e\u5373\u73a9\u7684 AI\u5e94\u7528\u3002\u9ad8\u6821\u56e2\u961f\u53ef\u901a\u8fc7\u201c\u5171\u4eab\u5b58\u50a8\u201d\u628a\u6570\u636e\u96c6\u6302\u5728\u591a\u53f0\u5b9e\u4f8b\u95f4\u5faa\u73af\u590d\u7528\uff1b\u521d\u521b\u516c\u53f8\u5219\u80fd\u628a\u8bad\u7ec3\u597d\u7684\u6a21\u578b\u76f4\u63a5\u53d1\u5e03\u81f3\u201c\u6a21\u578b\u5e02\u573a\u201d\uff0c\u6309\u8c03\u7528\u91cf\u6536\u8d39\uff0c\u5b9e\u73b0\u201c\u7b97\u6cd5\u53d8\u73b0\u201d\u3002\u5e73\u53f0\u63d0\u4f9b CI\/CD \u6d41\u6c34\u7ebf\uff0c\u4ee3\u7801\u63a8\u9001\u540e\u81ea\u52a8\u6253\u5305\u955c\u50cf\uff0c\u771f\u6b63\u8ba9\u7b97\u6cd5\u5de5\u7a0b\u5e08\u56de\u5f52\u7b97\u6cd5\u672c\u8eab\u3002<\/p>\n<h2>\u5199\u5728\u6700\u540e\uff1a\u8ba9\u521b\u65b0\u53ea\u5173\u5fc3\u521b\u65b0<\/h2>\n<p>\u5f53\u5f00\u6e90\u5927\u6a21\u578b\u8fdb\u5165\u201c\u6708\u66f4\u201d\u751a\u81f3\u201c\u5468\u66f4\u201d\u65f6\u4ee3\uff0c\u73af\u5883\u642d\u5efa\u3001\u9a71\u52a8\u51b2\u7a81\u3001\u6743\u91cd\u4e0b\u8f7d\u8fd9\u4e9b\u4f4e\u6548\u52b3\u52a8\u672c\u5c31\u4e0d\u8be5\u7531\u4eba\u7c7b\u91cd\u590d\u3002\u661f\u5b87\u667a\u7b97\u901a\u8fc7\u201c\u70ed\u66f4\u65b0\u201d\u6280\u672f\u628a GPU\u670d\u52a1\u5668\u79df\u7528 \u505a\u6210\u50cf\u6253\u5f00\u81ea\u6765\u6c34\u4e00\u6837\u7b80\u5355\uff0c\u8ba9 CUDA \u7248\u672c\u3001\u4f9d\u8d56\u5e93\u3001\u6570\u636e\u96c6\u5168\u90e8\u4e0b\u6c89\u5230 PaaS \u5c42\u3002\u4f60\u53ea\u9700\u4e13\u6ce8 loss \u66f2\u7ebf\u4e0e\u4e1a\u52a1\u6307\u6807\uff0c\u5269\u4e0b\u7684\u4ea4\u7ed9\u4e91\u3002<br \/>\n\u73b0\u5728\u6ce8\u518c\u5373\u53ef\u9886\u53d6 10 \u5143\u4f53\u9a8c\u91d1\uff0c<a href=\"https:\/\/www.starverse-ai.com\">\u70b9\u6b64\u7acb\u523b\u4f53\u9a8c GPU\u670d\u52a1\u5668\u79df\u7528<\/a>\uff0c\u4e00\u8d77\u628a 2026 \u5e74\u7684\u6a21\u578b\u66f4\u65b0\u901f\u5ea6\uff0c\u53d8\u6210\u4f60\u7684\u7ade\u4e89\u4f18\u52bf\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>2026 \u5e74 4 \u6708\uff0cHuggingFace Trendin&hellip;<\/p>\n","protected":false},"author":2,"featured_media":2892,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-2893","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-zixun"],"views":70,"_links":{"self":[{"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/posts\/2893","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/comments?post=2893"}],"version-history":[{"count":0,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/posts\/2893\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/media\/2892"}],"wp:attachment":[{"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/media?parent=2893"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/categories?post=2893"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/tags?post=2893"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}