Juraj on nostr

Open in Damus

nightwatch · 11w

Interesting. Have you found a good solution for a "coding LLM" but locally ?

Juraj @Juraj 1767573713

Qwen models are nice for local coding, but the sizes you can realistically run locally are at least two generations behind. But they are not completely useless.

1❤️1

Honza Pobořil · 10w

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 unreliable with models of this size.