| | --- |
| | title: Legaleaze |
| | emoji: ⚖️ |
| | colorFrom: blue |
| | colorTo: purple |
| | sdk: streamlit |
| | sdk_version: 1.39.0 |
| | app_file: app.py |
| | pinned: false |
| | models: |
| | - ckharche/legaleaze-bart-121k |
| | --- |
| | |
| | # Legaleaze: Legal Text Simplifier |
| |
|
| | Fine-tuned BART-Large-CNN with LoRA adapters to simplify complex legal language into plain English. |
| |
|
| | ## Training |
| | - **Steps**: 121,000 |
| | - **Dataset**: 53k asylum case documents (Canadian legal corpus) |
| | - **Hardware**: Northeastern University H100/H200 GPUs |
| | - **Method**: LoRA (r=16, α=32) for parameter-efficient fine-tuning |
| |
|
| | ## Performance |
| | | Metric | Score | |
| | |--------|-------| |
| | | FKGL Reduction | 35% | |
| | | BERTScore F1 | 0.89 | |
| | | ROUGE-L | 0.48 | |
| |
|
| | Built by [@ckharche](https://github.com/chaitanyaharche) |
