LitGPT gives you clean, readable implementations of 20+ pretrained language models without the usual framework bloat. You can load models like Microsoft's Phi-2 with just a few lines of Python and get access to production-ready training workflows. It's maintained by the community with 9.2K GitHub stars, so you're not dealing with abandoned code. The appeal here is simplicity: if you need to actually understand what's happening under the hood or customize model behavior without fighting through abstraction layers, this is a solid choice. Just note that Snyk flagged some warnings in the security audit, so review dependencies before using in production.
npx skills add https://github.com/orchestra-research/ai-research-skills --skill implementing-llms-litgpt