I want to try some open-source and/or local LLMs. Which models do you use and what are they best at? I've looked on GitHub for "awesome-lists" but nothing was really maintained.

It would be useful if you included the number of tokens per second and your CPU/GPU so that other readers know what to expect, too.