---

base_model: Qwen/Qwen2.5-7B-Instruct

tags:

- gguf

- recipe-generation

- unsloth

- lora

- qlora

language:

- en

pipeline_tag: text-generation

license: apache-2.0

---

CulinaAI - Qwen2.5-7B Recipe Generation (GGUF, Q4_K_M)

CulinaAI is a fine-tuned version of Qwen2.5-7B-Instruct, specialized for generating cooking recipes. It was trained using Unsloth Studio and exported to GGUF format for easy local inference.

Model Details

• Base model: Qwen/Qwen2.5-7B-Instruct
• Fine-tuning method: LoRA (QLoRA-style, 4-bit base model)
• Training framework: Unsloth Studio
• Format: GGUF, quantized to Q4_K_M (~4.68 GB)

Training Configuration

| Hyperparameter | Value |

|---|---|

| Epochs | 3 |

| Batch size | 2 |

| Learning rate | 2e-4 |

| Optimizer | AdamW 8-bit |

| Context length | 1024 |

| Warmup steps | 380 |

| LoRA rank (r) | 16 |

| LoRA alpha | 32 |

| LoRA dropout | 0.05 |

Final metrics:

• Smoothed training loss: 0.941
• Best eval loss: 0.719

Usage

This model is in GGUF format and works with:

• llama.cpp
• Ollama
• LM Studio
• Unsloth Studio itself

Ollama

ollama run hf.co/Shlok0011/CulinaAI-Qwen2.5-7B-GGUF

Example Prompt

### Instruction:
Give me a recipe for a quick vegetarian dinner using chickpeas.

### Response:
[model generates a recipe]

(Update this with an actual example from your model's output.)

Limitations

• Generated recipes may contain inaccurate quantities, cooking times, or steps. Always use judgment and review before cooking.
• The model may occasionally hallucinate ingredients or techniques that aren't standard.
• Trained with a 1024-token context window, so very long recipes or multi-recipe requests may be truncated.

Training Data

Fine-tuned on a 22,000-recipe subset drawn from an enhanced version of the Recipe1M+ dataset (MIT CSAIL). The full enhanced source dataset contains 1,023,489 recipes with the following columns, from which this training subset was sampled:

• title,
• meal_type, primary_classification, secondary_classification, context_dependent_classification — recipe categorization
• ingredients, NER — ingredient lists (raw and NER-extracted entities)
• instructions — step-by-step cooking instructions

> Note: This repository contains only the fine-tuned model weights, not the training dataset itself. Recipe1M+ is distributed under MIT's own terms of access — refer to the official Recipe1M+ site for dataset licensing if you plan to redistribute the data.