repo-chat 0.1.0

Using LLMS to facilitate document retrieval and chat operations

Repo Chat

Repo Chat is a software package for answering queries about code repositories using large language models with document retrieval. It leverages the langchain library to facilitate the retrieval and chat operations.

Installation

To install the required libraries, run:

pip install -r requirements.txt

Usage

Basic Usage

In order to use Repo Chat, you need to set the OPENAI_API_KEY in a .env file as shown below:

OPENAI_API_KEY=your_api_key_here

After setting the API key, you can use the chain_manager module to create a chain:

from repo_chat import chain_manager

chain = chain_manager.get_chain("UPGRADE_QUERY")

You can then call the chain with input data:

chain_input_data = {
    "query": "What does the function do?",
    "repo": "https://github.com/my-org/my-repo",
}

output = chain(chain_input_data)

Advanced Usage

Please refer to the source code examples in the following files to learn more about Repo Chat's advanced features:

• repo_chat\chain_manager.py
• repo_chat\chat_utils.py
• repo_chat\eval_utils.py
• repo_chat\templates.py

Chain Templates

Repo Chat uses multiple chain templates for different purposes. The templates are defined in repo_chat\templates.py. Here is a brief explanation of the available templates:

• UPGRADE_QUERY: Expand or refine a given query to improve the retrieved documents' quality and relevance.
• RUN_QUERY_RAG: Process the given query and its similar documents to provide the most accurate and relevant response possible.
• CONTEXT_VALIDATOR: Validate the context sufficiency of the given query and similar documents.
• CRITIC: Score a given response for a query by evaluating relevancy and accuracy.
• RAW_CODE: Process a given query for a specific code repository.

Evaluation

Repo Chat provides evaluation utilities in the repo_chat\eval_utils.py module. The CriticChain class is used to evaluate a response generated by the model. The QueryEvaluator class is used to evaluate a single query against the Repo Chat chain, and MultiQueryEvaluator evaluates multiple queries in parallel.

Contributing

Feel free to submit issues or pull requests if you have any suggestions or improvements you would like to add to Repo Chat.

License

Repo Chat is released under the MIT License.