I successfully compiled the project from the following GitHub repository on an Ubuntu server with x86 architecture:
However, upon transferring the compiled files and models to a Jetson Nano 4GB B01 with ARM 64 architecture, I encountered issues running them. How can I address this compilation challenge?
As I do not have access to an x86 server, is it possible to compile the project on the Jetson Nano and upgrade the CUDA version to CUDA 11 or do you have any alternative methods to suggest?
You can compile directly on the Nano if you have the disk space. You won’t be able to use CUDA 11 on the older Nano though (I think feature development for the oldest Nano stopped at CUDA 10). If there is a release of the software capable of using CUDA 10 you could put the project on a USB thumb drive (for space) and just directly compile.
Cross compiling is possible, but with user space programs (which link to libraries in the environment) it gets rather complicated and the Jetson itself isn’t too bad for compile speed.
Thank you for your reply. I followed your advice and successfully compiled on Jetson. However, I now want to compile using the GPU, but unfortunately the compilation requires ggml which requires a minimum of CUDA 11.0. Do you have any good suggestions?
I couldn’t tell you how to do this, but if you need a newer release, then probably the only possibility (without going to Xavier or Orin) is to use a docker container. Probably you’d want to start a new forum thread and title it to include using docker with CUDA 11. However, the GPU architecture is part of this, and I don’t know if even docker will work with this since the CUDA 11 version might be for the wrong GPU architecture. Someone else would need to answer that.
Got it!Thank you so much for taking the time to reply to my question!
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.