Turkishlaw

🚀 OrionCAF/TurkishLaw

A specialized language model tailored for Turkish Law (Unsloth 4-bit + LoRA, trained only on Turkish legal texts)

📚 General Description

Qwen3-14B-Merged-Law-TR is a specialized language model. It is fine - tuned using the LoRA method on Turkish Civil, Obligations, and Commercial Law texts, based on Alibaba's Qwen3 model. It is then embedded into a single file for efficient 4 - bit operation. This model is designed for lawyers, law students, academics, and anyone who needs quick access to legal information. It can generate explanatory and accurate responses supported by article numbers about the Turkish legal system.

⚠️ Important Note

This model does not replace legal advice. It is always recommended to consult a lawyer on professional legal matters.

✨ Features

• Provide detailed information about the Turkish Civil Code, Obligations Code, and Commercial Code.
• Refer to legal articles with correct numbers.
• Explain and interpret legal concepts.
• Answer legal questions within the framework of relevant legislation.
• Low memory usage (thanks to 4 - bit quantization).
• Generate quick responses.

🚀 Quick Start

The following is a code snippet demonstrating how the model generates content based on given inputs.

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
        model_name = "OrionCAF/TurkishLaw",
        max_seq_length = max_seq_length,
        #load_in_4bit = False,
        load_in_8bit = False,
    )

FastLanguageModel.for_inference(model)

messages = [{"role": "user", "content": "Türk Medeni Kanunu'na göre evlenme engelleri nelerdir? }]
    inputs = tokenizer.apply_chat_template(
        messages, tokenize=True, add_generation_prompt=True, return_tensors="pt"
    ).to(device)

    
    output = model.generate(
        input_ids=inputs, max_new_tokens=1500, use_cache=True,
        temperature=0.6, min_p=0.1
    )

    cikti_text = tokenizer.decode(output[0], skip_special_tokens=True)

🔧 Model Details

👥 Our Team

This model was developed by the following team members:

• Ferhat Kürkçüoğlu (https://www.linkedin.com/in/ferhatkurkcuoglu/)
• Cengizhan Bayram (https://www.linkedin.com/in/cengizhan-bayram-a66009223/)
• Cevdet Ahmet Turan (https://www.linkedin.com/in/cevdet-ahmet-turan/)
• Volkan Altıntaş (https://www.linkedin.com/in/volkanaltintas/)

Our team focuses on developing specialized AI models.

⚠️ Limitations

• The model is only trained in the fields of Turkish Civil, Obligations, and Commercial Law. It may have limited knowledge in other legal fields (such as criminal law, administrative law, etc.).
• The model's knowledge is limited by the date of the training data and may not include the latest legal changes.
• While 4 - bit quantization improves performance on low - end devices, it may cause a loss of performance compared to full precision in some cases.
• It should be used for information purposes, not for providing legal advice.

📄 License

This model is licensed under the Apache License 2.0. It can be freely used, including for commercial purposes, as long as proper attribution is made.

_{❓ For questions and feedback: GitHub | Hugging Face}

_{Developed by Ferhat, Cengizhan, Cevdet, and Dr. Volkan Altıntaş as part of NLP research.}