⚖️ Legal-LLaMA-3B (Fine-tuned on Indian Legal QA)

Model Description

This is a LLaMA-3B model fine-tuned using LoRA (merged) with Unsloth on ~14.5K Indian legal question-answer pairs.
The model is designed to act as a legal assistant chatbot specialized in Indian law (contracts, consumer protection, family law, etc.).

Developed by: DeathBlade020
Model type: Causal LM (decoder-only)
Language(s): English (with Indian legal terminology)
Finetuned from: LLaMA-3B base
License: LLaMA license (Meta AI)

Uses

Direct Use

Educational / research purposes for Indian law Q&A
Chatbot-style applications in legal learning

Out-of-Scope Use

❌ Not a substitute for professional legal advice
❌ Not intended for real-world legal decision making

Training Details

Dataset: ~14,543 Indian legal QA pairs
Training split: ~13,815 train / ~728 validation
Method: LoRA fine-tuning with Unsloth
Epochs: 3
Max seq length: 2048
Optimizer: AdamW, lr=2e-4
Hardware: Google Colab T4 GPU(Free)

Installation

Make sure you have the required libraries installed:

pip install unsloth transformers accelerate torch

Example Usage


from unsloth import FastLanguageModel

model_id = "DeathBlade020/legal-llama-3b"
model, tokenizer = FastLanguageModel.from_pretrained(model_id, max_seq_length=2048)
FastLanguageModel.for_inference(model)

messages = [
    {"role": "system", "content": "You are a legal expert specializing in Indian law."},
    {"role": "user", "content": "What are the essential elements of a valid contract under Indian law?"},
]

inputs = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to(model.device)
outputs = model.generate(input_ids=inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))