You need ollama[1][2] and hardware to run 20-70B models with quantization of Q4 at least to have similar experience to commercially hosted models. I use codestr

You need ollama[1][2] and hardware to run 20-70B models with quantization of Q4 at least to have similar experience to commercially hosted models. I use codestral:22b, gemma2:27b, gemma2:27b-instruct, aya:35b. Smaller models are useless for me, because my native language is Ukrainian (it's easier to spot mistakes made by model in a language with more complex grammar rules). As GUI, I use Page Assist[3] plugin for Firefox, or aichat[4] commandline and WebUI tool. [1]: https://github.com/ollama/ollama/releases https://github.com/ollama/ollama/releases [2]: https://ollama.com/ https://ollama.com/ [3]: https://github.com/n4ze3m/page-assist https://github.com/n4ze3m/page-assist [4]: https://github.com/sigoden/aichat https://github.com/sigoden/aichat