thealper2/gemma-3-1b-turkish-recipes

Overview

This model, thealper2/gemma-3-1b-turkish-recipes, is a 1 billion parameter Gemma-based language model, specifically fine-tuned to generate Turkish recipe preparation steps. It was developed by thealper2 through a LoRA fine-tuning process on unsloth/gemma-3-1b-pt.

Key Capabilities

• Turkish Recipe Generation: Generates detailed, numbered preparation steps in Turkish based on a recipe title, ingredient list, and optional serving/time details.
• Specialized Fine-tuning: Utilizes a LoRA fine-tune (merged, bf16) approach, trained on a subset of the mmkocak/turkish-recipes-175K dataset.
• Optimized for Completion: Achieves a perplexity of 3.957 on completion tokens, significantly outperforming the base model's 5.539.
• Structured Output: Produces output in a clear, numbered list format for preparation steps, ending with <eos>.

Good for

• Automated Recipe Instruction Generation: Ideal for applications requiring the automatic creation of cooking instructions in Turkish.
• Culinary Content Creation: Useful for generating content for recipe websites, cooking apps, or digital cookbooks focused on Turkish cuisine.
• Language-Specific NLP Tasks: Demonstrates effective fine-tuning for highly specific, language-dependent text generation tasks within a niche domain.

Limitations

• The model generates only preparation steps; it does not validate quantities, times, or temperatures, which can be invented.
• It does not provide nutrition information.
• The style and quality of generated recipes reflect the scraped home-cooking sources of the training data.