如何识别文件的真假 - 阮一峰的网络日志

link管理

链接快照平台

输入网页链接，自动生成快照
标签化管理网页链接

相关文章推荐

寂寞的芹菜 · Observable.TimeInterva ...· 1 年前 ·

无邪的小蝌蚪 · C#复杂XML反序列化为实体对象两种方式 ...· 1 年前 ·

慷慨的丝瓜 · 3分钟短文 | Laravel ...· 2 年前 ·

逆袭的打火机 · Java Swing ...· 2 年前 ·

while attempting to download the VIA Container Image and after:

docker login nvcr.io -u ‘$oauthtoken’ -p <NGC_API_KEY>

which obviously succeeded because I got following message:

WARNING! Using --password via the CLI is insecure. Use --password-stdin.
WARNING! Your password will be stored unencrypted in /home/koule/.docker/config.json.
Configure a credential helper to remove this warning.
Login Succeeded

then I tried to perform the download:

docker pull nvcr.io/metropolis/via-dp/via-engine:2.0-dp

and got following message:

permission denied while trying to connect to the Docker daemon socket at unix:///var/run/docker.sock: Post “http://%2Fvar%2Frun%2Fdocker.sock/v1.47/images/create? fromImage=nvcr.io %2Fmetropolis%2Fvia-dp%2Fvia-engine&tag=2.0-dp”: dial unix /var/run/docker.sock: connect: permission denied.

Is there a fix or workaround for this issue?

sudo docker pull nvcr.io/metropolis/via-dp/via-engine:2.0-dp
[sudo] password for user:

There’s one more thing I should probably mention: I didn’t go ahead with the key for OpenAI. I need to avoid using OpenAI because of the confidential data that we will process. And we want to process them only locally with our hardware.

I had to use sudo for both of following commands:

I don’t know whether it’s a good idea to use sudo. I also proceeded with the steps for rootless-mode but I still need to use sudo for the pull to begin.

when I use following command I get an error - no CUDA GPU available. Well, I have Gerofce RTX 4090 on this computer. Is it really the error message that I should get? Maybe I would expect something insufficient GPU memory or something related insufficient hardware.

Could you check your CUDA environment first by referring to the pytorch-says-that-cuda-is-not-available-on-ubuntu ?

The possible reason for this problem is that your preparation has not been done well. Please refer to our Guide step by step to install the driver, docker and NVIDIA Container Toolkit correctly.
Install the NVIDIA Driver .

Well, I didn’t want to uninstall the later version of driver that I had because I didn’t want to jeopardize the function of my Omniverse applications. Well I did it now and then I tried to carry out the installation of NVIDIA-Linux-x86_64-535.161.08.run using following commands:
and I got following error:

see also the log file:
log.txt (167.7 KB)

And now Installation of drivers for Geforce RTX 4090. I noticed that your drivers include cuda toolkit 12.2 so I found a page describing the installation steps for laptop version of Geforce RTX 4080. So, I hope the driver is going to suport the desktop version of RTX 4090 as well since desktop RTX4090 is in the list of supported devices for driver version 535.171.04:

Installing CUDA on Ubuntu 22.04 (RXT4080-laptop) Drivers - Linux, Windows, MacOS I tried to install nvcc on my Ubuntu 22.04 machine following the NVIDIA offical guides and broke my driver 3 times, with a hell lot of headache, and eventually found that this version works. Installing CUDA Toolkit 12.2 on Ubuntu 22.04 This guide will walk you through the process of installing the CUDA Toolkit 12.2 on Ubuntu 22.04, including setting up the environment variables and verifying the installation. Steps 1. Remove Previous CUDA Installations and Install the NVIDIA Driver Ensure that…

Plus, according to our guide, you need to guarantee the 65+ GB system memory and 40+ GB GPU memory.
Can you make sure that your laptop meets this requirement first?

I was out of office for some time now.
Well, as I already mentioned it’s Geforce RTX 4090 that I’ve been using so far. The other spec. is most likely OK. It’s a workstation with 2 CPUs of 48 CPU cores in total and RAM of 384 GB. So GPU memory is out of spec. because it’s only 24GB.

When your script finds out that GPU memory is less than 40 GB it switches the VLM batch size to 1. So, I was hoping that I cloud demo the VIA Microservices to other co-workers of mine even if I don’t get 30 FPS or if there is some other kind of latency that would be unacceptable in real application.
see also:

It seems to require deepstream plugin for gstreamer the this require TensorRT support. I found metropolis deepstream 6.4 installation instructions for Ubuntu 22.04 and it requires impossible combination of cuda toolkit 12.2 and TensorRT 8.6.1 (this version seem to be compatible with cuda toolkit 12.0 and 12.1)

So, it seems that I have to give up.

Since a very important event is coming very soon in our facility (November 6th and 7th). I don’t think I can get approval to rent a cloud server with appropriate GPU soon enough. It’s not only a purchase request. It also includes security approval which takes a lot of time in here.

So, is there any other way to make a demo for my co-workers. Would you be eventually willing to make a teams video conference via Microsoft Teams and make a short interactive introduction. Like you could let my co-workers give their own prompts to some video feeds.

And I also need to know whether you have a version of VIA Microservices for Nvidia Jetsons and which Jetson product would be the minimum requirement for a good operation results.

In case of the graphics card I prefer to purchase RTX A6000 because I also deal with animations and visualizations in Omniverse applications.

Nevertheless, before I can ask to purchase any hardware or enterprise account for VIA services I have to prove that it’s going to work and its useful.

So, if you have any idea how you can help me to convince the others that this is a very useful product, I’d be delighted to do my best to make it happen.