https://old.reddit.com/r/LocalLLaMA/

Bit of a hype madhouse whenever a new model is released, but it's pretty easy to filter out simple hype from people showing reproducible experiments, specific configs for llama.cpp, github links etc.