LegalSafe-Falcon-7B is a safety-aligned large language model fine-tuned for the Indian legal domain. The model focuses on reducing toxic, biased, and unsafe outputs while maintaining high accuracy in legal reasoning and responses. It integrates Constitutional AI principles and Reinforcement Learning from AI Feedback (RLAIF) to improve ethical alignment.

Developed by: Gopi M
Model type: Causal Language Model (LLM)
Language(s): English (Legal Domain - Indian Context)
License: Apache 2.0 (inherits from base model)
Finetuned from model: ybelkada/falcon-7b-sharded-bf16

Uses

Direct Use

Legal question answering
Legal document summarization
Safe text generation in sensitive domains
Educational tools for legal studies

Downstream Use

Integration into legal chatbots
AI-based legal assistants
Compliance and advisory systems
Safety-critical NLP applications

Out-of-Scope Use

Providing official legal advice
Use in high-risk legal decision-making without human oversight
Generating harmful, biased, or illegal content

Bias, Risks, and Limitations

May still exhibit residual bias from training data
Performance may vary outside Indian legal domain
Not a substitute for professional legal consultation
Risk of hallucinations in complex legal scenarios

Recommendations

Use with human oversight in legal applications
Validate outputs before deployment in critical systems
Avoid relying on model for final legal decisions

How to Get Started

python

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "gopi30/rlaif-safety-alligned"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

prompt = "Explain the concept of fundamental rights in Indian law."
inputs = tokenizer(prompt, return_tensors="pt")

outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0]))