I use a small non-reasoning local model for article summarization (llama3.2:3b), but wouldn't use that for translation. Maybe splitting it up even further helps?
Yes, splitting further is the plan. I’d go with something smaller in the first place if it was for personal use, but I’m working on a solution for the community, so I want to get best results and then see where I can downgrade without losing quality.
llama3.2:3b), but wouldn't use that for translation. Maybe splitting it up even further helps?