possible to do a qwq32b equivalent and 70b and 683b deepseek equivalent?

[sprappcom]

current 2b hallucination is high and token count is very little. it's very good considering all other factors but just too small.