Good results fine tuning a local LLM like Qwen 3:0.6B to categorize questions

A developer fine-tuned the tiny Qwen 3:0.6B language model (600M parameters) to categorize household questions for a RAG-based chatbot, achieving significantly better results than the baseline. The baseline model using only prompting achieved only ~10% accuracy on 131 test questions, while the fine-tuned version was trained on ~850 labeled household questions using the Unsloth framework to improve question categorization into metadata categories like "pool," "hvac," and "cooking." This experiment demonstrates that even very small local LLMs can be effectively fine-tuned to perform reliable classification tasks when given sufficient training data.