domain-name-generator

Domain Name Generator

Model Details

• Base model: Mistral-7B-v0.1
• Fine-tuning method: LoRA (via TRL’s SFT Trainer)
• Dataset: Fully synthetic, generated with LLaMA 3 (70B) using the GROQ API
• Languages: English
• Task: Domain name suggestion from business descriptions

Intended Use

• Primary use case: Generate domain name ideas given a short business description.
• Input: 1–5 sentence business description.
• Output: 3–5 candidate domain names, each with a confidence score.

Example:

{
  "business_description": "A startup offering eco-friendly clothing.",
  "n_candidates": 3
}

{
  "suggestions": [
    { "domain": "greenwear.com", "confidence": 0.81 },
    { "domain": "ecofit.net", "confidence": 0.72 },
    { "domain": "sustainstyle.org", "confidence": 0.66 }
  ]
}

Training Data

• Number of training samples: 14,849
• Number of evaluation samples: 3,291
• Total samples: 18,140

Training Parameters

The model was trained with the following configuration:

Evaluation

• Cosine similarity: Embedding-based metric to measure closeness between domain names and descriptions.
• GPT-4 scoring: Used on ~10% of the dataset for qualitative evaluation.

Limitations

• Synthetic data: Generated dataset may not capture all real-world naming conventions.
• English only: Not trained on other languages.
• Domain availability: Model does not check if suggested domains are already taken.

Citation

If you use this model, please cite:

@misc{domain-name-generator,
  title = {Domain Name Generator},
  author = {Jawaheer, Hassani},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/hassanij/domain-name-generator}}
}