Why these models?

Coding:

codegemma:2b → lightweight, good for completions.

codellama:7b → solid for structured code (like Docker Compose).

mistral:7b → generalist, also good with logic in code.

Writing (tech docs & emails):

llama3.2:3b → smaller generalist.

gemma:7b → more natural writing.

neural-chat:7b → conversational, good for email tone.