Skip to content

Latest commit

 

History

History
98 lines (76 loc) · 6.48 KB

DeveloperGuide.md

File metadata and controls

98 lines (76 loc) · 6.48 KB

Developer Guide

Introduction

Welcome to the developer guide for TigerGraph GraphRAG. This guide provides information on how to add a new LangChain tool, embedding service, or LLM generation service to GraphRAG.

Table of Contents

Adding a New LangChain Tool

If you want your agent to connect to other data sources or be able to perform custom logic, you can add a new LangChain tool to TigerGraph GraphRAG. To add a new LangChain tool, follow these steps:

  1. In the app/tools directory, create a new file for your tool. The file should be named toolname.py where toolname is the name of your tool.
  2. Define your tool. The tool should a valid Python class that inherits from the LangChain BaseTool class. For more information refer to the LangChain documentation.
  3. Add your tool to the app/tools/__init__.py file. This file should contain an import statement for your tool. For example:
from .generate_function import GenerateFunction
  1. Enable your tool to be used by the agent. To do this, import and instantiate your tool in the app/agent.py file. For example:
from tools import GenerateFunction
generate_function = GenerateFunction()

Then add the tool to the tools list in the Agent class. For example:

tools = [mq2s, gen_func, new_tool]

  1. Test your tool. Run the service and test your tool to ensure that it works as expected.

  2. (Optional): Think that your tool could be useful for others? Consider contributing it! To contribute your tool, submit a pull request to the TigerGraph GraphRAG repository and checkout our contributing guidelines.

Add a New Embedding Service

One might want to add a new embedding service to TigerGraph GraphRAG to better fit their deployment environment. To do this, follow these steps:

  1. In app/embeddings/embedding_service.py and create a new class that inherits from the BaseEmbeddingService class. For example:
class MyEmbeddingService(BaseEmbeddingService):
    def __init__(self, config):
        super().__init__(config)
        # Add your custom initialization code here
  1. Implement the needed methods for your service. If you utilize a LangChain-supported embedding service, you can use the BaseEmbeddingService class as a reference. If you are using a custom endpoint, you will need to implement the embed_documents and embed_query methods accordingly.
  2. Import your service and dd your service to the app/main.py file where the EmbeddingService class is instantiated. For example:
from common.embeddings.embedding_service import MyembeddingService

if llm_config["embedding_service"]["embedding_model_service"].lower() == "MyEmbeddingService":
    embedding_service = MyEmbeddingService(llm_config["embedding_service"])
  1. Test your service. Run the service and test your service to ensure that it works as expected.
  2. (Optional): Think that your service could be useful for others? Consider contributing it! To contribute your service, submit a pull request to the TigerGraph GraphRAG repository and checkout our contributing guidelines.

Add a New LLM Generation Service

To add a new LLM generation service to TigerGraph GraphRAG, follow these steps:

  1. Create a new file in the app/llm_services directory. The file should be named service_name.py where service_name is the name of your service.
  2. Define your service. The service should be a valid Python class that inherits from the LLM_Model class defined in the app/llm_services/base_llm.py file.
  3. Add your service to the app/llm_services/__init__.py file. This file should contain an import statement for your service. For example:
from .service_name import ServiceName
  1. Import and instantiate your service in the app/main.py file. For example:
from common.llm_services import ServiceName

# Within the instantiation of the Agent class elif block
elif llm_config["completion_service"]["llm_service"].lower() == "my_service":
    logger.debug(f"/{graphname}/query request_id={req_id_cv.get()} llm_service=my_service agent created")
    agent = TigerGraphAgent(AzureOpenAI(llm_config["completion_service"]), conn, embedding_service, embedding_store)
  1. Test your service. Run the service and test your service to ensure that it works as expected.
  2. (Optional): Think that your service could be useful for others? Consider contributing it! To contribute your service, submit a pull request to the TigerGraph GraphRAG repository and checkout our contributing guidelines.

Adding a New Test Suite

To add a new InquiryAI test suite to TigerGraph GraphRAG, follow these steps:

  1. Download the InquiryAI test template from here in .tsv format.

  2. Create a new directory in the tests/test_questions directory. The directory should be named suite_name where suite_name is the name of your test suite.

  3. Add the .tsv file to the new directory, populated with your example questions and expected answers.

  4. (Optional): Add the necessary GSQL and setup script to the tests/test_questions/suite_name directory to support your test suite. The setup scripts are not run with the test suite, but help to set up the graph for the test suite. The tests assume that the graph is already set up.

  5. Add necessary query descriptors to the tests/test_questions/suite_name directory. Within a directory named after the query, add a .json file with the query descriptor. Optionally add a .gsql file with the query itself.

  6. Add the test suite to the tests/test_questions/parse_test_config.py file by adding an available schema to the schema argument list.

  7. Test your test suite. Run the test suite and ensure that it works as expected. Run the tests with the following command (and add desired options described here):

./run_tests.sh
  1. (Optional): Think that your test suite could be useful for others? Consider contributing it! To contribute your test suite, submit a pull request to the TigerGraph GraphRAG repository and checkout our contributing guidelines.