Broccoli's Dev

参考视频：【DeepSeek+LoRA+FastAPI】开发人员如何微调大模型并暴露接口给后端调用 按照博主的视频安装之后遇到一些问题，可能是博主使用的lunix ， 而我使用的是windows又或是博主租用的服务器会有一些基础配置，所以按照步骤安装会报一些错误。 1. pip install -e “.[torch,metrics]”报错 查看.condarc配置文件，将default删除，替换成国内的镜像源 123456789show_channel_urls: truechannels: - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/r - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/msys2envs_dirs: - D:\Software\LLaMA-Factory - e:\llama-factory-domain 2. 运行...

Reference Video:【DeepSeek+LoRA+FastAPI】How Developers Fine-tune Large Models and Expose Interfaces for Backend Calls After following the blogger’s video installation, I encountered some issues. It might be because the blogger was using Linux, while I am using Windows, or the server rented by the blogger had some basic configurations pre-installed. Therefore, following the steps exactly resulted in some errors. 1. Error: pip install -e ".[torch,metrics]" Check the .condarc configurat...

由于VS新版本1.102.*无法访问Centos的服务器，所以采用了降级处理，但是降级之后重新安装remote点击openFolder一直显示There is no data provider registered that can provide view data. 并且整个控制台卡死放大界面也会出现无法适配的卡顿情况。 尝试更新新版Git,失败没有作用 CentOS 7 升级 git 版本到 2.x 删除%LocalRomingData%\Code下面的数据将所有插件的缓存信息删除，重新打开Vs Code然后登录便能够正确访问，如果显示文件被占用，可以控制台关闭所有的Python程序，怀疑是因为启动了太多的环境导致互相访问同一个remote资源，加上版本不一致造成卡顿

Since the new VS Code version 1.102.* cannot access CentOS servers, I adopted a downgrade approach. However, after downgrading and reinstalling Remote, clicking Open Folder constantly displayed There is no data provider registered that can provide view data. Additionally, the entire console froze, and scaling the interface also resulted in unadapted lag. Attempted to update to a newer version of Git, but it failed and had no effect. CentOS 7 Upgrade git version to 2.x Deleted the data und...

Linux Command Command Content mv(cp) /root/path/* /etc/ Move/ Copy i Modify ? ** Find wq Exit firewall-cmd --zone=public --list-all Find all firewall open ports ll All file information cd .. Return to the previous level firewall-cmd --zone=public --add-port=11206/tcp --permanent Open port firewall-cmd --reload Configuration takes effect :q! Exit without saving

Linux命令 命令 内容 mv(cp) /root/path/* /etc/ 移动/复制 i 修改 ? ** 查找 wq 退出 firewall-cmd --zone=public --list-all 查找所有的防火墙开放端口 ll 所有的文件信息 cd .. 返回上一级 firewall-cmd --zone=public --add-port=11206/tcp --permanent 开放端口 firewall-cmd --reload 配置生效 :q! 不保存退出

Lora-Script摘要主要是为了记录自己再模型微调的过程中遇到的一些问题，和一些参数配置 ，显卡为云端租赁使用 多卡模型训练主要是使用koyha_ss的框架修改，并使用deepspeed3进行多卡训练 单卡则是使用的aki的工具包进行lora-script的训练 bash config1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465[model]v2 = falsev_parameterization = falsepretrained_model_name_or_path = "./sd-models/realismEngineSDXL_v30VAE.safetensors"vae = "./sd-models/sdxl_vae.safetensors"[dataset]train_data_dir = "./train/001&...

Lora-ScriptAbstractThis post primarily records issues encountered during model fine-tuning and some parameter configurations. The graphics cards used were rented from a cloud provider. Multi-card model training mainly uses a modified kohya_ss framework and utilizes deepspeed 3 for multi-GPU training. Single-card training uses the aki toolkit for lora-script training. Bash Config123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616...

Proxifier连接Cursor与Antigravity 下载proxfier ， 可以从官网下载 ： https://www.proxifier.com/ 激活31天试用版，可以从网上找一个激活码使用 下载Antigravity ， 下载 ：https://antigravity.google/?hl=zh-cn 增加Proxy Server,可以从VPN中看到相关的端口之类数据填入进去 增加新的Proxy Rule , 我这里将所有的Antigravity的相关exe进行代理，因为我同时使用cursor与antigravity所以我便将所有的相关都填入进去 打开Antigravity ， 点击Sign In WitH Google , 这个时候会有两种情况 如果显示区域不符，则取像谷歌发送更改邮箱地址的请求，更改为美国，我是更改为美国使用新加坡的节点也没有任何问题 如果跳转后显示账户无法登录，但是同时收到了欢迎加入Antigravity的邮件，那么就是本次登录存在我呢提，可以重新关闭这个网站 再次点击Sign In ， 并将跳转网址粘贴进无痕浏览器中，...

Connect Cursor and Antigravity with Proxifier Download Proxifier. You can download it from the official website: https://www.proxifier.com/ Activate the 31-day trial, or find an activation code online. Download Antigravity. Download: https://antigravity.google/?hl=en Add a Proxy Server. You can find the relevant port and other data from your VPN and enter it. Add a new Proxy Rule. I proxy all relevant executables for Antigravity here. Since I use both Cursor and Antigravity, I entered all...

SSH Connection Failure TroubleshootingInitial State Client: Local Windows (OpenSSH client) Server: Cloud Platform Linux VM (Azure) Symptoms: Local SSH client connection failed. Known: Cloud platform’s built-in Bastion (web terminal) can login normally. Investigation Module 1: Server Log Analysis Issue: sshd service logs (journalctl -u sshd) show multiple warnings and errors. Deprecated option RhostsRSAAuthentication (and other Deprecated warnings) error: Unable to load host key: /etc/ssh/ss...

SSH 连接失败排查记录初始状态 客户端： 本地 Windows (OpenSSH 客户端) 服务器： 云平台 Linux 虚拟机 (Azure) 症状： 本地 SSH 客户端连接失败。 已知条件： 云平台自带的 Bastion（网页终端）可以正常登录。 排查模块一：服务器日志分析 问题： sshd 服务日志 (journalctl -u sshd) 显示多条警告和错误。 Deprecated option RhostsRSAAuthentication (及其他 Deprecated 警告) error: Unable to load host key: /etc/ssh/ssh_host_dsa_key 排查过程： 分析 Deprecated 警告：确认这些是 sshd_config 配置文件中的旧配置项，它们是“警告”而非“错误”，服务会忽略它们并正常运行。 分析 Unable to load host key：确认 dsa_key 是一种过时的密钥类型，新系统默认不生成。服务会跳过此密钥，正常加载其他（如 RSA, ED25519）密钥。 结论： 服务器日志中...

由于VS新版本1.102.*无法访问Centos的服务器，所以采用了降级处理，但是降级之后重新安装remote点击openFolder一直显示There is no data provider registered that can provide view data. 并且整个控制台卡死放大界面也会出现无法适配的卡顿情况。 尝试更新新版Git,失败没有作用 CentOS 7 升级 git 版本到 2.x 删除%LocalRomingData%\Code下面的数据将所有插件的缓存信息删除，重新打开Vs Code然后登录便能够正确访问，如果显示文件被占用，可以控制台关闭所有的Python程序，怀疑是因为启动了太多的环境导致互相访问同一个remote资源，加上版本不一致造成卡顿

Due to new version 1.102.* of VS cannot access Centos server, so adopted downgrade processing. But after downgrade reinstalled remote clicking openFolder always displays There is no data provider registered that can provide view data. and entire console freezes, enlarging interface also appears stuttering unable to adapt. Tried to update new version Git, failed no effect CentOS 7 Upgrade git version to 2.x Delete data under %LocalRomingData%\Code delete all plugin cache info, reopen Vs Co...

Githubhttps://github.com/imkcrevit/RevitMCP_Blog IntroductionCurrently, the internet is flooded with various AI-related information. As recent development work involves AI technology, this article introduces the integration method between MCP and Revit to achieve the “one-click modeling” feature in public accounts. This is just one of the possible implementation approaches, and other technical paths exist. However, this method allows for rapid deployment, helping teams quickly explore relevan...

Githubhttps://github.com/imkcrevit/RevitMCP_Blog 摘要当前，各类AI相关资讯充斥着网络。由于近期开发工作涉及AI技术，因此撰写本文介绍MCP与Revit的联动方式，实现公众号中的”一句话建模”功能。这只是其中一种实现方案，还存在其他技术路径。不过，这种方法能够快速落地，有助于大家迅速探索相关应用方向。 Function Call在讲MCP之前需要先讲一下Function Call , 这是一个LLM大模型API提供的方式，通过写入Function Call 在LLM交互的时候可以自动选用相应的Function执行。**图片来源：ailydoseofds **以DeepSeek为例，在API中就介绍了Function Call的使用方法 12345678910111213141516171819tools = [ { "type": "function", "function": { "name": "...

When starting Navisworks development, you might encounter situations where DataProperty values appear as “unknown”. This scenario isn’t an error but rather a property type that requires specific handling in the code. The API provides separate conversion methods to retrieve the correct property values.  As shown in the image above, calling ToString() directly on a wall property in the code results in “unknown...

再开始接触Naviworks的开发时候会出现Dataproperty的值unknow的情况，这种场景不是报错而是在代码中需要判定值的属性，API会提供一个单独的转换方法，从这些方法中即可获取到正确的属性值。  像是上图一个墙体的属性在代码中直接ToString()，会获取到unknow，是因为没有使用单独的转化方法，通过下面的代码，对每个value进行单独的判定转换，即可获取到所有的值。 在Naviworks中依旧存在英寸与毫米的换算系数的情况，这种需要在进行一步转换。 12345678910111213141516171819202122232425262728293031323334353637383940414243var stringBuilder = new StringBuilder();var doc = Application.ActiveDocument;var selection = doc.Current...

大多数博客给出的方案都是如何在有自适应族的情况下进行修改定位点或是将数据传入自适应族,如何直接在族文件中创建自适应模型并将点转换为自适应点,连接自适应点成为自适应路径这种方式没有文章介绍.  下面的代码中给出了如何在自适应族文件中创建参照点并转换为自适应点连接成为路径,最终创建放样融合的办法. public Result Execute(ExternalCommandData commandData, ref string message, ElementSet elements) { var uiDoc = commandData.Application.ActiveUIDocument; // UIDocument var doc = uiDoc.Document; ...

Most blogs only cover how to modify positioning points or pass data into an existing adaptive family. There are no articles introducing how to directly create an adaptive model in a family file, convert points to adaptive points, and connect adaptive points to form an adaptive path.  The code below demonstrates how to create reference points in an adaptive family file, convert them to adaptive...