Good results fine tuning a local LLM like Qwen 3:0.6B to categorize questions 197 points by dev-experiments 19 hours ago 43 comments story