sanidhya1910/recipe-lens-4b
sanidhya1910/recipe-lens-4b is a 4 billion parameter recipe assistant model fine-tuned to provide structured JSON output for recipe generation, scaling, and ingredient substitution. Developed by sanidhya1910, it excels at generating recipes from available ingredients, adjusting serving sizes, and suggesting ingredient swaps, ensuring high accuracy in format validity, containment, and scaling. This model is specifically designed for applications requiring precise, machine-readable recipe data rather than free-form text.
Loading preview...
Recipe Lens: A Structured Recipe Assistant
sanidhya1910/recipe-lens-4b is a 4 billion parameter model specifically fine-tuned to act as a precise recipe assistant. Unlike general-purpose LLMs, this model focuses on generating structured JSON output for specific recipe-related queries, ensuring high reliability and machine-readability.
Key Capabilities
- Ingredient-based Recipe Generation: Given a list of ingredients, it generates a recipe using only those items (plus pantry staples), declining if insufficient ingredients are provided.
- Recipe Rescaling: Accurately adjusts all ingredient quantities in a given recipe to match a new serving size.
- Ingredient Substitution: Suggests valid ingredient swaps and adjusts amounts when a specific ingredient is unavailable.
- Structured Output: All responses are provided as a single JSON object containing
title,servings,ingredients(withitem,qty,unit), andsteps. - High Accuracy: Achieves 98.7% format validity, 98.0% containment (uses only listed ingredients), and 100.0% scaling accuracy on held-out test sets, with metrics computed by rules, not LLM judges.
- Zero Memorization: Demonstrates 0.0% memorization, indicating it generates novel recipes rather than reproducing training data verbatim.
What Makes This Model Different?
This model stands out due to its specialized focus on structured recipe generation and manipulation with verifiable accuracy. While other LLMs can generate recipes, Recipe Lens is explicitly trained to produce parseable JSON output and adhere to strict constraints like ingredient containment and precise scaling. Its evaluation relies on rule-based metrics (e.g., arithmetic for scaling, set operations for containment) rather than subjective LLM judges, ensuring objective performance measurement. It also explicitly handles cases where a recipe cannot be made, declining rather than hallucinating.
Should You Use This for Your Use Case?
This model is ideal for applications requiring reliable, structured recipe data where precision in ingredient usage, scaling, and format is critical. Use it if your application needs:
- Automated recipe generation from user-provided ingredients.
- Dynamic adjustment of recipe serving sizes.
- Intelligent ingredient substitution in cooking apps.
- Integration with systems that consume JSON-formatted recipe data.
Avoid if you need a general-purpose conversational AI or a model that generates free-form, creative culinary text without strict structural requirements.