I Fine-Tuned a 4B Model to Write Etsy Listings That Actually Rank

I fine-tuned a 4B model to create optimized Etsy listings, proving that a smaller fine-tuned model can outperform a larger base model at this specific task. This side hustle allows Etsy sellers to generate well-structured titles, descriptions, and tags in a cost-effective, efficient manner. Using the LoRA fine-tune of Qwen3.5-4B on Tinker, Thinking Machines' fine-tuning API, the project measured the improvement in performance through a blind test, where the fine-tuned model outperformed the base model in title SEO, tag relevance, and overall judgment.

However, the fine-tuned model does generate more tags than required, indicating that there is still room for improvement in the model's output. All aspects of this project, including the base model weights, training and evaluation scripts, and metrics, are openly available, allowing others to reproduce and build upon this work.