39 lines
1.6 KiB
Markdown
39 lines
1.6 KiB
Markdown
---
|
|
license: apache-2.0
|
|
base_model: unsloth/meta-llama-3.1-8b-instruct-bnb-4bit
|
|
tags:
|
|
- legal
|
|
- text-simplification
|
|
- llama-3.1
|
|
- justice-tech
|
|
- unsloth
|
|
- impacthacks
|
|
- featherless-ai
|
|
model_name: Legal-Lens-500
|
|
language:
|
|
- en
|
|
pipeline_tag: text-generation
|
|
---
|
|
|
|
# Legal-Lens-500: Llama-3.1 Fine-Tuned for Legal Clarity
|
|
|
|
Developed by **Akhil Reddy & Ben Gideon** for **ImpactHacks 2026**, **Legal-Lens-500** is a specialized 8B parameter model designed to bridge the "Justice Gap." It simplifies dense legal contracts into plain English and identifies potential "red flag" clauses for non-experts.
|
|
|
|
## 🚀 Impact & Purpose
|
|
Inaccessible legal language (legalese) often leads to the exploitation of tenants, students, and freelancers. This model serves as a "first-pass" review tool to:
|
|
- **Democratize Legal Knowledge:** Translate jargon into a 6th-grade reading level.
|
|
- **Identify Risks:** Surface hidden obligations in rental and employment agreements.
|
|
- **Empower Individuals:** Provide clarity before users sign binding documents.
|
|
|
|
## 🛠 Technical Specifications
|
|
- **Base Model:** `meta-llama-3.1-8b-instruct`
|
|
- **Fine-tuning Method:** QLoRA (via Unsloth) - Trained 2x faster with optimized memory usage.
|
|
- **Export Format:** **FP16 Safetensors** (Optimized for high-speed inference on [Featherless.ai](https://featherless.ai)).
|
|
- **Architecture:** Llama 3.1 (8B parameters).
|
|
|
|
## 💡 How to Use via Featherless API
|
|
Once indexed, you can call this model using the OpenAI-compatible Featherless endpoint:
|
|
|
|
|
|
Vote for API support in the Discussions tab!
|
|
https://huggingface.co/spaces/huggingface/InferenceSupport/discussions/8980#69cd4d981be3e071495e5581 |