Ayushi054/BNS-Legal-Phi2

TEXT GENERATIONConcurrency Cost:1Model Size:3BQuant:BF16Ctx Length:2kPublished:Apr 26, 2026Architecture:Transformer0.0K Cold

Ayushi054/BNS-Legal-Phi2 is a 2.7 billion parameter language model, fine-tuned from Microsoft's Phi-2 using LoRA. It specializes in the Bharatiya Nyaya Sanhita (BNS) 2023, India's new criminal law code, trained on 3,000 BNS-specific question-answer pairs. This model is designed for legal research and educational purposes related to Indian criminal law.

Loading preview...

BNS Legal Phi-2: Specialized Indian Criminal Law Model

Ayushi054/BNS-Legal-Phi2 is a specialized language model built upon microsoft/phi-2, fine-tuned using the LoRA method. Its primary focus is the Bharatiya Nyaya Sanhita (BNS) 2023, India's new criminal law code, which replaced the Indian Penal Code 1860. This model is specifically trained to understand and respond to queries related to BNS 2023.

Key Capabilities

  • BNS 2023 Expertise: Fine-tuned on a custom dataset of 3,000 question-answer pairs derived directly from the BNS 2023 document.
  • Efficient Fine-tuning: Utilizes LoRA (rank=16, bf16) for efficient adaptation of the base Phi-2 model.
  • Focused Legal Knowledge: Capable of providing information on definitions, punishments, and scenarios as outlined in the BNS 2023.

Training Details

The model was trained for 3 epochs on a Kaggle T4 x2 GPU for approximately 6.5 hours, achieving a best evaluation loss of 0.1489. The training dataset includes a mix of definition, punishment, scenario, and augmented samples from the BNS 2023 PDF.

Intended Use

This model is intended for educational purposes and legal research/reference concerning the BNS 2023. It serves as a tool for understanding the provisions of the new Indian criminal law code. Users should note that this model does not provide legal advice and a qualified legal professional should always be consulted for legal matters.