The current gold standard for running an LLM locally on a 3090 24GB GPU would be using Oobabooga coupled with Sillytavern. These tools are well-regarded for their ease of use and efficiency. For the latest models and advice, consulting the "best current local model" thread on the LocalLLaMA subreddit is recommended.