A powerful Thai legal assistant with RAG techniques.
🚀 Installation
- Create an .env file with the following content:
OPENAI_API_KEY=YOUR_OPENAI_API_KEY TYPHOON_API_KEY=YOUR_TYPHOON_API_KEY
- Install the package using pip:
git clone https://github.com/BetterACS/LawBuddy cd LawBuddy pip install -r requirements.txt pip install -e .
💡 Quick Start Using OpenAI Model
from lawbuddy.rag import SimpleRagPipeline, Hybrid, Graph, Hyde, QueryTransformType
pipeline = SimpleRagPipeline.from_openai_model(model="gpt-3.5-turbo")
pipeline.create_vector_store(
csv_paths=["laws.csv"],
save_dir="spaces/hybrid_rag"
)
query = "โดนโกง 300 ล้านบาทไทย แต่คนโกงไม่โดนฟ้องควรทำยังไง"
response = pipeline.query(query, verbose=True)📚 Vector Store Management
-
Creating a New Vector Store
# Create vector store from CSV files pipeline.create_vector_store( csv_paths=["laws.csv"], save_dir="spaces/hybrid_rag" )
-
Loading Existing Vector Store
pipeline.load_vector_store(path="spaces/hybrid_rag")
🚌 Query Transforms
-
Simple Query Transform (default)
from lawbuddy.rag import QueryTransformType pipeline.query( query="โดนโกง 300 ล้านบาทไทย แต่คนโกงไม่โดนฟ้องควรทำยังไง", query_transform_mode=QueryTransformType.SIMPLE )
No query transformation is applied.
-
Chunk Query Transform
from lawbuddy.rag import QueryTransformType pipeline.query( query="โดนโกง 300 ล้านบาทไทย แต่คนโกงไม่โดนฟ้องควรทำยังไง", query_transform_mode=QueryTransformType.CHUNK )
Chunk the query into smaller parts for better processing. Each chunk is being retrieved separately from the vector store and then concatenated later.
🤖 Load model
- Local model
from transformers import AutoModelForCausalLM, AutoTokenizer from lawbuddy.rag import Hybrid # Load model and tokenizer model = AutoModelForCausalLM.from_pretrained("openthaigpt/openthaigpt1.5-7b-instruct") tokenizer = AutoTokenizer.from_pretrained("openthaigpt/openthaigpt1.5-7b-instruct") # Load specialized legal adapter model.load_adapter("betteracs/lawbuddy-7b") # Initialize pipeline with local model pipeline = Hybrid.from_local_model( model_name="openthaigpt/openthaigpt1.5-7b-instruct", model=model )
- API
from lawbuddy.rag import SimpleRagPipeline pipeline = SimpleRagPipeline.from_api( model="typhoon-v1.5-instruct", api_base="https://api.opentyphoon.ai/v1", context_window=8192, is_chat_model=True, max_tokens=768, is_function_calling_model=False, api_key="...." )
💹 Graph RAG For the Graph RAG model, you need to install the Neo4j database and run the following commands:
-
Installation
pip install neo4j pip install llama-index-vector-stores-neo4jvector
-
Start Neo4j database with docker
docker run \ -p 7474:7474 -p 7687:7687 \ -v $PWD/data:/data -v $PWD/plugins:/plugins \ --name neo4j-apoc \ -e NEO4J_apoc_export_file_enabled=true \ -e NEO4J_apoc_import_file_enabled=true \ -e NEO4J_apoc_import_file_use__neo4j__config=true \ -e NEO4JLABS_PLUGINS=\[\"apoc\"\] \ neo4j:latest -
Authenticate with the default username and password (neo4j/neo4j) or changing the password in browser at http://localhost:7474
-
Create a new graph
from lawbuddy.rag import Graph import nest_asyncio nest_asyncio.apply() # Required for preventing asyncio conflicts graph = Graph.from_openai_model(model="gpt-3.5-turbo") graph.create_graph( csv_paths=["laws.csv"], url="neo4j://localhost:7687", username="neo4j", # default username password="neo4j" # default password ) # graph.load_graph( # url="neo4j://localhost:7687", # username="neo4j", # password="neo4j" # )
This might take a while to create the graph.
To evaluate the model performance on specific tasks or legal document types, use the following script. This example shows how to evaluate on the Civil (แพ่ง) law type.
import os
from dotenv import load_dotenv
from lawbuddy.eval import evaluate
from lawbuddy.rag import Hybrid
# Load pipeline
pipeline = Hybrid.from_openai_model(model="gpt-3.5-turbo")
# Load existing vector store
pipeline.load_vector_store(path="spaces/iterative_query_chunking")
# Get OpenAI API key
openai_key = os.getenv('OPENAI_API_KEY')
# Run evaluation
evaluate(pipeline, type_name='แพ่ง', model='gpt-3.5-turbo', openai_key=openai_key)The system supports various configurations for both OpenAI and local models. You can customize:
Chunk sizes for document processing Vector store parameters Model-specific settings Query processing parameters
🤝 Contributing Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change. 📝 License MIT License 📬 Contact For support or queries, please open an issue in the GitHub repository.
Made with ❤️ for the LawBuddy team.