I'm done with Windows and I'm making the switch to Linux. However, I'm not sure which distribution to go with. I have some projects running fine on Windows (specifically related to LLM fine-tuning), but I'm hitting some dependency issues with reinforcement learning tasks like downloading the right libraries. I have some basic experience with Linux from working on a Raspberry Pi, but I really want to ensure everything I do works smoothly on Linux. I'm not considering Arch. What distro would you recommend, and are there specific packages I should install or avoid? By the way, I currently have CUDA version 13.0 and I'm using Build CUDA 12.8.
2 Answers
Linux and Windows are quite different, so you'll want to consider your workflow and the applications you rely on. Since you mentioned you're mostly using VS Code and Obsidian, you should be fine switching as those have great support on Linux. If you're worried about compatibility, maybe try running some tests in a virtual machine or creating a dual-boot setup initially before fully committing to Linux.
For standard users, I'd recommend going with Ubuntu or any of its derivatives like Zorin, Mint, or Kubuntu. They’re user-friendly and have strong community support. Just keep in mind, the specific needs of your projects will also play a big role in deciding if Linux can meet those needs. If your projects require certain Windows applications, you'll want to ensure they can run on Linux or find suitable alternatives.
Thanks for the guidance! I’m not too concerned about mainstream Windows apps like Office or Photoshop since I mainly use open-source software, but I’ve heard mixed things about Fedora. I've considered Debian too, given its stability. Would either of those be suitable for fine-tuning tasks?