acidtib/dispensary-product-name-gemma3
The acidtib/dispensary-product-name-gemma3 is a 0.3 billion parameter fine-tuned Gemma-3 model developed by acidtib. It is specifically designed to rewrite raw Dutchie POS listing names into a canonical "Strain - Type" format for internal dispensary inventory tracking. This model achieves a 98.8% exact match rate on its specialized product name extraction task, making it highly accurate for its niche application. It is not intended for general-purpose use outside of this specific product name normalization task.
Loading preview...
Model Overview
acidtib/dispensary-product-name-gemma3 is a specialized fine-tune of the google/gemma-3-270m-it model, developed by acidtib. Its primary function is to transform raw product names from Dutchie POS listings into a standardized "Strain - Type" format. This model is highly focused and not intended for general-purpose natural language processing tasks.
Key Capabilities
- Product Name Normalization: Rewrites complex, raw dispensary product names into a clean, consistent format (e.g., "Eureka Live Resin Reload 1g; Super Skunk" becomes "Super Skunk - Live Resin").
- High Accuracy: Achieves an impressive 98.8% exact match rate on a held-out evaluation set of 245 examples, demonstrating its effectiveness for its specific task.
- Compact Size: Built on a 0.3 billion parameter Gemma-3 base, making it efficient for deployment in specialized applications.
Training Details
The model was trained on 1697 examples from the acidtib/dispensary-product-name-extraction dataset, with 1488 examples used for training and 245 for evaluation. The training involved 10 epochs with a learning rate of 3e-05, resulting in a best evaluation loss of 0.003563.
Intended Use
This model is specifically designed for internal dispensary inventory tracking systems that require standardized product naming. It is not recommended for general-purpose text generation or other broad NLP applications due to its highly specialized fine-tuning.