You can deploy this model and start earning money today!

Amey9766/
qwen2.5-1.5b-culinary-sft

Name: qwen2.5-1.5b-culinary-sft
Availability: InStock

safetensors

qwen2

2407.10671

Qwen/Qwen2.5-1.5B-Instruct

finetune

mit

Views

No views yet

🍽️ Qwen2.5-1.5B Culinary SFT

A domain-specialized instruction-tuned language model for culinary education, professional kitchen practices, and food & beverage theory.

🔍 Model Overview

Qwen2.5-1.5B Culinary SFT is a fine-tuned version of Qwen2.5-1.5B-Instruct, adapted specifically for:

Culinary students and hospitality learners
Professional chefs and kitchen staff
Food safety, hygiene, and kitchen operations education
Structured recipe generation and culinary theory explanation

The model is designed to consistently produce structured, practical, and industry-aligned outputs, following professional culinary standards.

🧠 Base Model

Base model: Qwen/Qwen2.5-1.5B-Instruct
Architecture: Transformer (RoPE, SwiGLU, RMSNorm, GQA)
Context length: up to 32,768 tokens
Parameters: ~1.54B
License: Apache 2.0 (inherits base model license)

🧪 Fine-Tuning Methodology

Technique: Supervised Fine-Tuning (SFT)
Approach: Parameter-Efficient Fine-Tuning (LoRA / QLoRA)
Quantization: 4-bit NF4 during training
Optimizer: Paged AdamW (8-bit)
Training framework: Hugging Face Transformers
Experiment tracking: Weights & Biases (wandb)

The final model was merged into a standalone full model for inference and deployment.

📚 Dataset Description

The fine-tuning dataset consists of culinary-domain instructional data, including:

Content Types

Culinary theory and fundamentals
Kitchen hygiene and food safety protocols
Cooking techniques and terminology
Professional kitchen workflows
Structured recipe knowledge
Common mistakes and best practices

Data Sources

Public-domain culinary textbooks
Openly available educational PDFs
Web-based culinary education resources
Public instructional material for chefs and culinary students

⚠️ No proprietary, private, or copyrighted content was intentionally included.
All data was sourced from public-domain or openly accessible educational materials.

🎯 Model Behavior & Output Style

The model is trained to behave as a culinary tutor and follows these principles:

When the input is a recipe request

The model outputs a standardized recipe card with:

Title
Servings / Yield
Mise en place (first)
Ingredients (prefer grams)
Step-by-step method
Timing
Allergen information
Chef notes

When the input is theory or concepts

The model produces structured study notes including:

Clear definitions
Key concepts and scope
Practical kitchen examples
Safety and hygiene notes
Common mistakes to avoid

🧪 Evaluation Summary

Training loss: 1.07
Validation loss: 1.10
Perplexity: 2.62
Qualitative evaluation: Manual testing confirms improved domain alignment, structured responses, and reduced hallucinations for culinary topics.

This model is optimized for educational clarity and professional correctness, not creative writing.

Citation for base model used-

@misc{qwen2.5, title = {Qwen2.5: A Party of Foundation Models}, url = {https://qwenlm.github.io/blog/qwen2.5/}, author = {Qwen Team}, month = {September}, year = {2024} }

@article{qwen2, title={Qwen2 Technical Report}, author={An Yang et al.}, journal={arXiv preprint arXiv:2407.10671}, year={2024} }

Citation for base model-

@misc{qwen25_culinary_sft, title = {Qwen2.5-1.5B Culinary SFT}, author = {Amey Tillu}, year = {2026}, note = {Domain-specific supervised fine-tuning on culinary education and professional kitchen practices}, howpublished = {\url{https://huggingface.co/Amey9766/qwen2.5-1.5b-culinary-sft}} }

👤 Author

Amey Tillu Hospitality & Tourism Data Analytics | AI & ML Hobbyist

🚀 Inference Example

python

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "Amey9766/qwen2.5-1.5b-culinary-sft"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)

prompt = "Explain what culinary arts is and describe key hygiene practices in a professional kitchen."

messages = [
    {"role": "system", "content": "You are a culinary tutor for chefs and culinary students."},
    {"role": "user", "content": prompt}
]

input_ids = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

attention_mask = torch.ones_like(input_ids)

with torch.no_grad():
    output = model.generate(
        input_ids=input_ids,
        attention_mask=attention_mask,
        max_new_tokens=500,
        temperature=0.7,
        top_p=0.9
    )

response = tokenizer.decode(
    output[0][input_ids.shape[-1]:],
    skip_special_tokens=True
)

print(response)