0

Problem: I'm trying to copy a conda environment from one machine to another machine. The source machine is Internet-connected, CUDA-enabled Ubuntu, and the destination machine is an offline Ubuntu, has no CUDA machine. To avoid unexpected problems, I hope that there is a way to exclude CUDA settings when I create an environment in source machine.

What I've tried: Without considering CUDA availability, I've made a clone of a conda environment in the source machine and manually copied into the offline machine, following the link https://stackoverflow.com/a/58103362/10982975. It was able to activate the environment in the destination machine, but still I'm worrying potential problems which can be caused by the CUDA configuration mismatch.

What I've not tried yet: Physical removal of CUDA card from the source machine.

Summarize: Is there a way to exclude CUDA settings when I create an environment in source machine?

talonmies
  • 70,661
  • 34
  • 192
  • 269
overfit
  • 349
  • 1
  • 12
  • Do you actually need to replicate the environment? or do you just want to create *an* environment on the other machine without going online? – merv Dec 13 '22 at 02:25
  • @merv I want to create an environment on offline computer but there is no simple way to install required packages offline (including python upgrade). That's why I try to replicate the environment from online machine to offline machine.. – overfit Dec 13 '22 at 11:41
  • 1
    Make a new environment and don't install CUDA stuff in it. Like use `tensorflow-cpu` instead of `tensorflow-gpu`, etc. Conda is fully capable of virtually solving any environment across different hardware - no need to alter your machine. Otherwise, if you'd like more help, you need to specify what exactly you want to install on the machine. Try something, and give more details about the results (not prose descriptions). – merv Dec 13 '22 at 18:19
  • @merv You're right. Thank you for suggestions. I will try to tailor packages at first and if some details are obtained, I will describe them here. – overfit Dec 13 '22 at 23:42

0 Answers0