使用 GPT 交互式学习 Repo
Learn A Repo Interactively with GPT. Ask questions to let the GPT explain the code to you.
GPT-Code-Learner supports running the LLM models locally. In general, GPT-Code-Learner uses LocalAI for local private LLM and Sentence Transformers for local embedding.
Please refer to Local LLM for more details.
Note: Due to the current capability of local LLM, the performance of GPT-Code-Learner is not as good as the online version.
git clone https://github.com/JinghaoZhao/GPT-Code-Learner.git
pip install -r requirements.txt.env file to put your API key:OPENAI_API_KEY=sk-xxxxxx
LLM_TYPE="OpenAI"
EMBEDDING_TYPE="OpenAI"If you want to run the whole program locally, please change the following line in the .env file:
LLM_TYPE="local"
EMBEDDING_TYPE="local"Repo Link textbox and click Analyze Code Repo button in the GUI. Or manually clone the repo you want to learn into code_repo folder:cd code_repo
git clone python run.pyGPT-Code-Learner generates vector database from the code repo as a knowledge base to answer repo-related questions. By default, it will use the source codes as the knowledge base. More details can be found in Knowledge Base.
The core of the GPT-Code-Learner is the tool planner. It leverages available tools to process the input to provide contexts.
Currently, the tool planner supports the following tools:
Code_Searcher: This tool searches keywords (e.g., specific functions or variables) extracted from user query in the code repository
Repo_Parser: This tool performs a fuzzy search with vector database of the code repo. It provides contexts for questions about the general procedures in the repo.
More tools are under development. Feel free to contribute to this project!
暂无开放 Issues,或尚未同步最近议题。