README.md

metadata

license: apache-2.0
language:
  - en
base_model:
  - mistralai/Mistral-7B-v0.1
tags:
  - legal

reglab-rrc/mistral-rrc

Paper: AI for Scaling Legal Reform: Mapping and Redacting Racial Covenants in Santa Clara County

Usage

Here is an example of how to use the model to find racial covenants in a page of text:

from transformers import AutoTokenizer, AutoModelForCausalLM
import re

# Load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("reglab/mistral-rrc")
model = AutoModelForCausalLM.from_pretrained("reglab/mistral-rrc")

def format_prompt(document):
    return f"""### Instruction:
Determine whether the property deed contains a racial covenant. A racial covenant is a clause in a document that \
restricts who can reside, own, or occupy a property on the basis of race, ethnicity, national origin, or religion. \
Answer "Yes" or "No". If "Yes", provide the exact text of the relevant passage and then a quotation of the passage \
with spelling and formatting errors fixed.

### Input:
{document}

### Response:"""

def parse_output(output):
    answer_match = re.search(r"\[ANSWER\](.*?)\[/ANSWER\]", output, re.DOTALL)
    raw_passage_match = re.search(r"\[RAW PASSAGE\](.*?)\[/RAW PASSAGE\]", output, re.DOTALL)
    quotation_match = re.search(r"\[CORRECTED QUOTATION\](.*?)\[/CORRECTED QUOTATION\]", output, re.DOTALL)
    
    answer = answer_match.group(1).strip() if answer_match else None
    raw_passage = raw_passage_match.group(1).strip() if raw_passage_match else None
    quotation = quotation_match.group(1).strip() if quotation_match else None
    
    return {
        "answer": answer == "Yes",
        "raw_passage": raw_passage,
        "quotation": quotation
    }

# Example usage
document = "Your property deed text here..."
prompt = format_prompt(document)
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=512)
result = tokenizer.decode(outputs[0])
parsed_result = parse_output(result)

print(parsed_result)

The model was trained with the given input and output formats, so be sure to use them when performing inference.

Intended Use

This model is designed to detect racial covenants in property deeds.

Training Data

Performance

Limitations

Ethical Considerations

Citation

@article{suranisuzgun2024,
  title={AI for Scaling Legal Reform: Mapping and Redacting Racial Covenants in Santa Clara County},
  author={Surani, Faiz and Suzgun, Mirac and Raman, Vyoma and Manning, Christopher D. and Henderson, Peter and Ho, Daniel E.},
  journal={},
  year={2024}
}