No root access on the remote server, but still want a newer PyTorch and transformers setup? These are my notes on using Conda to install CUDA and the tools I need without changing the system driver. I cover version choices, checking that PyTorch can use the GPU, pointing programs to the Conda CUDA installation, and watching out for xformers dependency changes.
After recently replacing my device with a new one, I was about to recompile an iOS jailbreak plugin called PassBy, and found that it frequently reported errors. After a bit of troubleshooting and searching, I determined that switching Xcode to an older version was the easiest solution, so I documented it.
I introduced in another article in the Linux remote server using conda configuration r-kernel jupyter notebook, but I found that conda for a variety of dependency conflicts are not handled well, often leading to unsuccessful installation. In macOS system locally using brew for installation is much easier and faster. Here I summarized steps for reference.
macOS has now replaced the default shell with zsh, which is a perfect match for the Oh My Zsh project to achieve an unimaginable terminal experience. With its colorful themes, powerful plug-ins, and active community, the "Black Windows" is no longer the cold terminal interface it used to be, but a unique my zsh of your own!
If you're like me and don't like the "bloat" of the official docker client, or simply want to try new things and tinker with new methods, then you're welcome to read this post. Since the blogger is using macOS, this is only a record of exploration under Mac. If there are any errors of omission, please feel free to criticize and point them out in the comment section.
Due to the author's frequent use of conda to configure the R kernel jupyter notebook on Linux remote servers, some steps are forgotten over time (the main problem is that the same pit will be stepped on differently every time). After summarizing the relevant tutorials and personal experience on the internet, a guide is provided for everyone and future me to refer to.