What size is useful for what tasks?
Looking for some usecases for max 12b models (works on my Air).
Found just one - summarisation of meetings transcriptions.
Even MCP tools calling is very unrelia...
Unfortunately not much. Can't you run even quantized models?