Hongji Zhu
Hongji Zhu
 如果web页面显示像上图一样,底部按钮不可点击, 表示没连接上model server,请先检查model server是否正常。 If the web page appears as shown in the image, with the button at the bottom greyed out and unclickable, it indicates that the model...
正常情况下,点击底部call通话按钮,会弹出mic和camera权限请求,点击允许即可。如果没有弹出,请确认使用Chrome浏览器。 Under normal circumstances, clicking the "Call" button at the bottom will trigger a request for microphone and camera permissions. Simply click "Allow" to proceed. Make sure you are using...
如果没有弹出权限请求,请点击左上角i按钮,并点击reset permission,然后刷新重试。 If the permission request does not appear, click the "i" button in the top left corner, select "Reset permissions," and then refresh the page to try again.
@cizhenshi 抱歉,我们发现使用ip地址访问会无法获取camera/mic权限,原因是没有https,浏览器会认为不安全而拒绝访问camera/mic。现已补充解决方案,请拉一下最新代码,执行make_ssl_cert.sh 创建自签名证书,如下: ``` cd web_demos/minicpm-o_2.6/web_server # create ssl cert for https, https is required to request camera and microphone permissions. bash ./make_ssl_cert.sh # output key.pem and cert.pem pnpm install...
> > 你好模型仓库里有的,  > > 也就是说我训练完后得把原始的模型文件夹下的东西合并过来对吗 是的
现在vllm支持直接`pip install vllm==0.5.4`使用,支持openbmb/MiniCPM-V-2_6及openbmb/MiniCPM-Llama3-V-2_5 > 假如要用vllm推理,两步就行了,很简单的 > > 1. 编译制作好.whl文件 > 2. 基于英伟达官方的推理镜像上制作,载入编译好的文件,然后安装各种包 > 注意编译的主机的nvcc版本、driver版本和英伟达推理镜像对齐
Thanks Gradio team,We will consider using the new features in our future version.
@zhouzhoumd 请问你是运行2.6的模型微调后出错吗,我们正在check为什么会缺文件
> > 营养、简单又美味的番茄炖牛腩。 牛腩切块,焯水去除血腥味。切块高压锅炖烂后,放入切好的番茄、土豆块,炖煮。 之后淋在饭上,非常之美味。 > > 得加八角香叶还有番茄酱吧? 不建议番茄酱,用番茄炒出就好,原汁原味原生态
For single image input, 24GB is enough for 1 instance, but for multi-images or video you may need 2x24GB using multi-gpus deployment like [web_demo_2.6.py#L44-L64](https://github.com/OpenBMB/MiniCPM-V/blob/main/web_demo_2.6.py#L44-L64) .