<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[盛世芳华]]></title><description><![CDATA[人工智能、机器学习爱好者]]></description><link>https://blog.csdn.net/lanyan90</link><language>zh-cn</language><generator>https://blog.csdn.net/</generator><copyright><![CDATA[Copyright &copy; lanyan90]]></copyright><item><title><![CDATA[mmpretrain自定义评估指标]]></title><link>https://blog.csdn.net/lanyan90/article/details/146939930</link><guid>https://blog.csdn.net/lanyan90/article/details/146939930</guid><author>lanyan90</author><pubDate>Wed, 02 Apr 2025 11:40:09 +0800</pubDate><description><![CDATA[在验证和测试期间计算 top-1 准确率、top-5 准确度、精确度和召回率。多标签分类指的是一个图片可能属于多个分类；单标签分类指的是一个图片只属于一个分类；]]></description><category></category></item><item><title><![CDATA[vllm运行过程中出现cuda out of memory]]></title><link>https://blog.csdn.net/lanyan90/article/details/146566610</link><guid>https://blog.csdn.net/lanyan90/article/details/146566610</guid><author>lanyan90</author><pubDate>Thu, 27 Mar 2025 17:26:12 +0800</pubDate><description><![CDATA[将长序列分解为多个小块，每个块分别处理。这样可以减少单次 Prefill 的计算量和显存占用，从而优化 TTFT 和显存使用效率。考虑使用 CPU 卸载（如 --cpu-offload-gb），将部分缓存卸载到 CPU 内存中，扩展 GPU 的虚拟内存。在大语言模型中，当输入序列很长（例如超过 1024 个 Token）时，Prefill 阶段的计算量会非常大。最后增加了enable-chunked-prefill参数和max_num_batched_tokens解决了问题。参数来进一步优化性能。]]></description><category></category></item><item><title><![CDATA[prisma创建新表，添加数据时找不到对象]]></title><link>https://blog.csdn.net/lanyan90/article/details/146542068</link><guid>https://blog.csdn.net/lanyan90/article/details/146542068</guid><author>lanyan90</author><pubDate>Wed, 26 Mar 2025 21:53:22 +0800</pubDate><description><![CDATA[使用vscode时，await prisma.order.create却找不到对象，重启vscode即可。]]></description><category></category></item><item><title><![CDATA[vercel部署项目报错Error [PrismaClientInitializationError]: Prisma has detected that this project was built]]></title><link>https://blog.csdn.net/lanyan90/article/details/146525405</link><guid>https://blog.csdn.net/lanyan90/article/details/146525405</guid><author>lanyan90</author><pubDate>Wed, 26 Mar 2025 12:00:10 +0800</pubDate><description><![CDATA[【代码】vercel部署项目报错Error [PrismaClientInitializationError]: Prisma has detected that this project was built。]]></description><category></category></item><item><title><![CDATA[shadcn如何给dialog增加关闭按钮和隐藏右上角关闭按钮]]></title><link>https://blog.csdn.net/lanyan90/article/details/146488735</link><guid>https://blog.csdn.net/lanyan90/article/details/146488735</guid><author>lanyan90</author><pubDate>Mon, 24 Mar 2025 22:59:01 +0800</pubDate><description><![CDATA[【代码】shadcn如何给dialog增加关闭按钮和隐藏右上角关闭按钮。]]></description><category></category></item><item><title><![CDATA[prisma新增字段后，查询不到字段的原因]]></title><link>https://blog.csdn.net/lanyan90/article/details/146462715</link><guid>https://blog.csdn.net/lanyan90/article/details/146462715</guid><author>lanyan90</author><pubDate>Sun, 23 Mar 2025 21:46:53 +0800</pubDate><description><![CDATA[在schema.prisma中新定义了一个字段，但使用prisma却查不到该字段，原来需要重新生成数据模型的client。2.执行npx prisma generate。3.成功后再启动服务即可。1.必须先停用服务。]]></description><category></category></item><item><title><![CDATA[nextjs使用next-intl要注意]]></title><link>https://blog.csdn.net/lanyan90/article/details/146448246</link><guid>https://blog.csdn.net/lanyan90/article/details/146448246</guid><author>lanyan90</author><pubDate>Sat, 22 Mar 2025 22:56:18 +0800</pubDate><description><![CDATA[网页一直没出错，就没注意，但如果page设置成了async，就会出现[Error: Expected a suspended thenable. This is a bug in React. Please file an issue.]错误。]]></description><category></category></item><item><title><![CDATA[Next.js集成prisma保存NextAuth第三方用户登录信息]]></title><link>https://blog.csdn.net/lanyan90/article/details/146340077</link><guid>https://blog.csdn.net/lanyan90/article/details/146340077</guid><author>lanyan90</author><pubDate>Tue, 18 Mar 2025 14:12:07 +0800</pubDate><description><![CDATA[修改prisma/schema.prisma文件，定义数据表字段@db.Text@db.Text@db.Text@unique。]]></description><category></category></item><item><title><![CDATA[DeepSeek-VL2使用swift微调的环境搭建和微调过程]]></title><link>https://blog.csdn.net/lanyan90/article/details/146032374</link><guid>https://blog.csdn.net/lanyan90/article/details/146032374</guid><author>lanyan90</author><pubDate>Wed, 05 Mar 2025 08:46:34 +0800</pubDate><description><![CDATA[最大模型用8张卡跑起来很难，需要降低--lora_rank、--lora_alpha、--max_length、--deepspeed、--per_device_train_batch_size、--gradient_accumulation_steps、--dataloader_num_workers才能刚好跑起来，虽然跑起来了，但训练速度却非常的慢。模型合并后，将模型复制到deepseek-vl2/deepseek-ai/目录下，使用deepseek-vl2的web.demo可以跑一下看看运行效果。]]></description><category></category></item><item><title><![CDATA[DeepSeek-VL2多显卡运行Gradio Demo]]></title><link>https://blog.csdn.net/lanyan90/article/details/145993439</link><guid>https://blog.csdn.net/lanyan90/article/details/145993439</guid><author>lanyan90</author><pubDate>Mon, 03 Mar 2025 17:08:43 +0800</pubDate><description><![CDATA[关键位置就是：'deepseek-ai/deepseek-vl2': [4, 4, 4, 4, 4, 4, 3, 3]，为不同的卡分配不同层。虽然指定了8张卡，但实际只用了一张卡，Gradio Demo程序并不支持多卡推理。把原来的load_model注释掉。]]></description><category></category></item><item><title><![CDATA[新装openwebui网页等好久才能打开？]]></title><link>https://blog.csdn.net/lanyan90/article/details/145908366</link><guid>https://blog.csdn.net/lanyan90/article/details/145908366</guid><author>lanyan90</author><pubDate>Thu, 27 Feb 2025 16:52:29 +0800</pubDate><description><![CDATA[因为模型设置里，默认连接openai的api，耐心等等，进来修改一下openai的api连接为你的地址即可。]]></description><category></category></item><item><title><![CDATA[小程序读取 wxfile://开头的临时文件]]></title><link>https://blog.csdn.net/lanyan90/article/details/145147467</link><guid>https://blog.csdn.net/lanyan90/article/details/145147467</guid><author>lanyan90</author><pubDate>Tue, 14 Jan 2025 20:57:15 +0800</pubDate><description><![CDATA[【代码】小程序读取 wxfile://开头的临时文件。]]></description><category></category></item><item><title><![CDATA[显卡Unable to determine the device handle for GPU等错误解决办法]]></title><link>https://blog.csdn.net/lanyan90/article/details/144665422</link><guid>https://blog.csdn.net/lanyan90/article/details/144665422</guid><author>lanyan90</author><pubDate>Mon, 23 Dec 2024 14:15:36 +0800</pubDate><description><![CDATA[最近公司多台AI服务器在训练过程中出现各自显卡报错，找不到原因，最后猜测是电源或功率太大造成的，用的是rtx3090，最后锁住功率，正常了！]]></description><category></category></item><item><title><![CDATA[openwebui连接oneapi]]></title><link>https://blog.csdn.net/lanyan90/article/details/143774242</link><guid>https://blog.csdn.net/lanyan90/article/details/143774242</guid><author>lanyan90</author><pubDate>Thu, 14 Nov 2024 16:15:47 +0800</pubDate><description><![CDATA[参考：【免费外语陪练】open-webui+oneapi连接在线大模型（通义等） - 哔哩哔哩]]></description><category></category></item><item><title><![CDATA[openwebui二改界面环境搭建]]></title><link>https://blog.csdn.net/lanyan90/article/details/143762776</link><guid>https://blog.csdn.net/lanyan90/article/details/143762776</guid><author>lanyan90</author><pubDate>Thu, 14 Nov 2024 11:19:55 +0800</pubDate><description><![CDATA[后端首次执行时，会去huggingface下载模型，网络原因，需要梯子。]]></description><category></category></item><item><title><![CDATA[openwebui通过Docker运行报错OSError: We couldn‘t connect to ‘https://huggingface.co‘ to load this file]]></title><link>https://blog.csdn.net/lanyan90/article/details/143731623</link><guid>https://blog.csdn.net/lanyan90/article/details/143731623</guid><author>lanyan90</author><pubDate>Wed, 13 Nov 2024 10:12:28 +0800</pubDate><description><![CDATA[运行后就会发现在data目录下有“models--sentence-transformers--all-MiniLM-L6-v2”，这就是联网下载下来的模型，下次可以整体拷贝整个data目录到docker-compose.yml放在一起，不挂代理也能正常访问了。docker运行openwebui报错OSError: We couldn't connect to ']]></description><category></category></item><item><title><![CDATA[Python保存CSV文件，Excel打开后中文乱码]]></title><link>https://blog.csdn.net/lanyan90/article/details/142880208</link><guid>https://blog.csdn.net/lanyan90/article/details/142880208</guid><author>lanyan90</author><pubDate>Sat, 12 Oct 2024 15:59:19 +0800</pubDate><description><![CDATA[在做多语言文件处理时， 使用 pandas， 并且指定了encoding为 UTF-8， 在 IDE， Sublime等编辑器上查看都显示正常，使用Excel打开非英文字符， 例如汉字， 阿拉伯文， 希伯来文等显示乱码。Excel 对于非BOM (字节顺序标记)1 保存的CSV文件默认采取ASNI编码 23, pandas 默认的保存方式并没有在文件头加上 (0xEF, 0xBB, 0xBF) or \ufeff。]]></description><category></category></item><item><title><![CDATA[vue3集成google第三方登陆]]></title><link>https://blog.csdn.net/lanyan90/article/details/142378358</link><guid>https://blog.csdn.net/lanyan90/article/details/142378358</guid><author>lanyan90</author><pubDate>Fri, 20 Sep 2024 10:42:42 +0800</pubDate><description><![CDATA[网上资源很多，但乱七八糟，踩坑几小时后，发现下面的方式没问题。]]></description><category></category></item><item><title><![CDATA[多网卡服务器，互换网卡名]]></title><link>https://blog.csdn.net/lanyan90/article/details/142359436</link><guid>https://blog.csdn.net/lanyan90/article/details/142359436</guid><author>lanyan90</author><pubDate>Thu, 19 Sep 2024 14:26:53 +0800</pubDate><description><![CDATA[服务器两张网卡，eth0和eth1，网卡1出现了问题，在不影响服务器内部各种桥接网络的情况下，直接将网卡2改为eth0即可。]]></description><category></category></item><item><title><![CDATA[list is not in GROUP BY clause and contains nonaggregated column ‘*‘ which is not functionally depen]]></title><link>https://blog.csdn.net/lanyan90/article/details/142177339</link><guid>https://blog.csdn.net/lanyan90/article/details/142177339</guid><author>lanyan90</author><pubDate>Thu, 12 Sep 2024 15:28:22 +0800</pubDate><description><![CDATA[因为当前SQL 运行模式为 ONLY_FULL_GROUP_BY，该模式要求 SELECT 列表中的所有非聚合字段必须出现在 GROUP BY 子句中，或者与 GROUP BY 列中的字段存在函数依赖关系。最简单的办法就是禁用 ONLY_FULL_GROUP_BY。]]></description><category></category></item></channel></rss>