Beverage Parser Model

A specialized fine-tuned Ollama model that extracts structured data from beverage descriptions. This model can identify and parse:

• Beverage types and brands
• Container volumes
• Alcohol content
• Packaging formats

For our beverage parser use case, I chose llama3.2:1b as it provides a good balance of efficiency and capability - it can handle the text parsing requirements while being lightweight enough to run efficiently in various environments.

Installation

Create the model using Ollama:

$ ollama create "beverage-parser" -f Modelfile

Usage

Run the model with a beverage description:

$ ollama run beverage-parser "A bottle of Heineken beer, 330ml, 5% alcohol"
{
  "quantity": 1,
  "type": "beer",
  "name": "Heineken",
  "alcohol_percentage": 0.05,
  "volume_ml": 330
}

Example Inputs

Simple Descriptions

• "A bottle of Heineken beer, 330ml, 5% alcohol"
• "2 cans of Coca-Cola, 355ml each"
• "750ml bottle of Chardonnay wine, 13.5% ABV"

Complex Descriptions

• "Dom Pérignon Vintage 2010 Champagne (12.5°), 750ml"
• "Six-pack of Sierra Nevada Pale Ale, 12 fl oz bottles, 5.6% alcohol"
• "1.75L Grey Goose vodka, 40% alcohol by volume"

Non-Alcoholic Beverages

• "A glass of tap water"
• "Monster Energy Drink"
• "3 shots of espresso"

Alternative Formats

• "75cl Bordeaux red wine, 14 degrees"
• "Handle of Jack Daniel's (1.75L, 80 proof)"
• "500 milliliters of craft IPA at 6.8%"

Multi-Unit Packages

• "2x 330ml Red Bull sugar free"
• "Magnum of Moët & Chandon (1.5L), 12% ABV"
• "4-pack of 250ml Gin & Tonic cans, 5% alcohol"

Additional Resources

For more information about Ollama model files:

• Ollama Modelfile Documentation