README.md

---
language:
- en
license: llama3.1
library_name: ollama
tags:
- legal
- singapore
- law
- assistant
- llama
- quantized
pipeline_tag: text-generation
base_model: meta-llama/Meta-Llama-3.1-8B-Instruct
base_model_relation: quantized
model-index:
- name: LexSG
  results: []
---

# LexSG - Singapore Legal Assistant Model

A specialized AI assistant trained on Singapore statutes and subsidiary legislation, built on the Llama 3.1 8B Instruct architecture and optimized for legal text generation.

## Model Details

### Model Description

LexSG is a fine-tuned and quantized language model designed specifically to assist with Singapore legal matters. It provides accurate, contextual responses about Singapore's legal framework and helps users understand complex legal provisions.

- **Developed by:** Chang Sau Sheong
- **Model type:** Causal Language Model
- **Language(s) (NLP):** English
- **License:** Llama 3.1 License
- **Finetuned from model:** meta-llama/Meta-Llama-3.1-8B-Instruct

### Model Sources

- **Repository:** (https://huggingface.co/sausheong/lexsg)
- **Base Model:** [meta-llama/Meta-Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct)

## Uses

### Direct Use

This model is intended for educational and informational purposes to help users understand Singapore legal provisions and statutes. It can be used to:

- Explain legal sections and provisions from Singapore acts
- Answer questions about Singapore's legal framework
- Provide context for legal documents
- Help interpret legal language and terminology
- Assist with understanding regulatory requirements

### Downstream Use

The model can be integrated into legal research tools, educational platforms, or chatbot applications focused on Singapore law.

### Out-of-Scope Use

- **Not for legal advice:** This model should not be used as a substitute for professional legal counsel
- **Not for other jurisdictions:** Specifically trained on Singapore law and may not be accurate for other legal systems
- **Not for critical decisions:** Should not be used for making important legal or business decisions without professional verification

## Bias, Risks, and Limitations

- **Training data limitations:** Responses are based on training data and may not reflect the most recent legal changes
- **Legislation only:** Training data is Singapore statutes and subsidiary legislation only, without any Singapore legal cases
- **Legal complexity:** Legal interpretations can be highly context-dependent and nuanced
- **Professional consultation required:** Complex legal matters require consultation with qualified legal professionals
- **Potential biases:** May reflect biases present in legal training data

### Recommendations

Users should be made aware of the risks, biases and limitations of the model. Always consult with qualified legal professionals for specific legal matters.

## How to Get Started with the Model

### llama.cpp/Ollama

The model file `llama-3.1-8b-lexsg-q4_k_m.gguf` is formatted in GGUF and can be used in any llama.cpp compatible library or application. 
Specifically it has been tested in Ollama [Ollama](https://ollama.com/), with the given Modelfile

### Running the Model

To use this with Ollama:

1. Build the model from the Modelfile:
   ```bash
   ollama create lexsg -f Modelfile
   ```

   or even simpler just do this:
   ```bash
   ./setup_ollama_model.sh
   ```

2. Run the model:
   ```bash
   ollama run lexsg
   ```

3. Start asking questions about Singapore law:
   ```
   > What does Section 73 of the Companies Act cover?
   > Explain the requirements for setting up a private limited company in Singapore
   > What are the penalties for non-compliance with PDPA?
   ```

## Training Details

### Training Data

The model was fine-tuned on Singapore legal documents and statutes, including but not limited to:
- Singapore Acts and Statutes
- Legal provisions and regulations
- Case law references
- Regulatory guidelines

### Training Procedure

#### Training Hyperparameters

- **Training regime:** Fine-tuned from Llama 3.1 8B Instruct
- **Quantization:** Q4_K_M (4-bit quantized for efficient inference)

#### Speeds, Sizes, Times

- **Model size:** ~4.8GB (quantized)
- **Context length:** 4,096 tokens
- **Max generation:** 1,024 tokens


## Technical Specifications

### Model Architecture and Objective

- **Architecture:** Llama 3.1 transformer architecture
- **Training objective:** Causal language modeling

### Hardware

- **Memory requirements:** ~6GB RAM recommended for inference
- **Platform support:** Cross-platform via Ollama

### Inference parameters

The following are the inference parameters in the model file. You can change it accordingly.

- Temperature: 0.3 (conservative, factual responses)
- Top-p: 0.9 (nucleus sampling for quality)
- Top-k: 40 (controlled vocabulary selection)
- Repeat penalty: 1.1 (reduces repetition)

## Model Card Authors

Chang Sau Sheong

## More Information

For more details about Singapore legislation, refer to [Singapore Statutes Online](https://sso.agc.gov.sg/)

---

**Legal Disclaimer:** This model is designed to provide general information about Singapore law and should not be considered as legal advice. For specific legal matters, always consult with a qualified legal professional licensed to practice in Singapore.