OR-Assistant

Introduction

The OpenROAD chat assistant aims to provide easy and quick access to information regarding tools, responses to questions and commonly occurring problems in OpenROAD and its native flow OpenROAD-flow-scripts.

The current architecture uses certain retrieval techniques on OpenROAD documentation and other online data sources. We aim to continuously improve the architecture and the associated the dataset to improve accuracy, coverage and robustness.

Use Cases

• Installation and Troubleshooting Assistance: The chatbot will provide users with quick and accurate solutions to common installation issues and troubleshooting steps.
• Easy Access to Existing Resources: The chatbot will be able to summarize relevant information from OpenROAD documentation, user guides, and online resources to provide concise and actionable answers to user queries.

Components

We have divided our app into three components, each of which can be hosted on a separate machine for scalability.

• Backend: Generates the necessary chat endpoints for users to communicate with.
• Frontend: We use Streamlit to communicate with a chat endpoint, providing a user-friendly chat interface.
• Evaluation: Besides the vanilla chat interface, we also have a human evaluation interface for research and development.

Setup

This setup involves the setting of both the frontend and backend components. We shall begin with backend:

Backend Setup

Option 1 - Docker

Ensure you have docker and docker-compose installed in your system.

Step 1: Clone the repository:

  git clone https://github.com/The-OpenROAD-Project/ORAssistant.git

Step 2: Copy the .env.example file, and update your .env file with the appropriate API keys.

  cd backend
  cp .env.example .env

Step 3: Start the docker container by running the following command:

  docker compose up

Option 2 - Local Install

Prerequisites:

• Python 3.12, recommended using a virtual environment like conda.
• wget
• pandoc
• git

Step 1: Install the required dependencies.

pip install -r backend/requirements.txt

Step 2: Copy the .env.example file, and update your .env file with the appropriate API keys.

  cd backend
  cp .env.example .env

Step 3: For populating the data folder with OR/ORFS docs, OpenSTA docs and Yosys docs, run:

  cd backend && python build_docs.py

Step 4: To run the server:

  python main.py

Optionally: To interact with the chatbot in your terminal, run:

python chatbot.py

The backend will then be hosted at http://0.0.0.0:8000.

Open http://0.0.0.0:8000/docs for the API docs.

Frontend Setup

Note: Please refer to the frontend README for more detailed instructions.

• Step 1: Set up the .env as per the instructions in the frontend README. Get the Google Sheet API Key

cd frontend
cp .env.example .env

• Step 2: Install the necessary requirements. You are encouraged to use a virtual environment for this.

pip install -r requirements.txt

• Step 3: Run streamlit application

streamlit run streamlit_app.py

Architecture Overview

OpenROAD documentation, OpenROAD-flow-scripts documentation, manpages and OpenSTA documentation is chunked and embedded into FAISS Vector Databases.

Documents are first retrieved from the vectorstore using a hybrid retriever, combining vector and semantic search methods. These retrieved documents undergo re-ranking using a cross-encoder re-ranker model.

flowchart LR
    id0([Query]) --> id1

    id1([Vectorstore]) --- id2([Semantic Retriever])
    id1([Vectorstore]) --- id3([MMR Retriever])
    id1([Vectorstore]) --- id4([BM25 Retriever])

    id2([Semantic Retriever]) -- Retrieved Docs ---> id5([Reranking]) 
    id3([MMR Retriever]) -- Retrieved Docs ---> id5([Reranking])
    id4([BM25 Retriever]) -- Retrieved Docs ---> id5([Reranking])

    id5([Reranking]) ---> id6(top-n docs)
 

Loading

Depending on the input query, each query can be forwarded to any one of the following retrievers,

1. General OR/ORFS information
2. OR tools and commands
3. OR/ORFS installation
4. OR Error Messages
5. OpenSTA docs
6. Yosys docs

The retrievers act as separate tools and can be accessed by the LLM's tool-calling capabilities.

The langgraph framework has been used to make effective use of the multiple retriever tools. Upon receiving a query, a routing LLM call classifies the query and forwards it to the corresponding retriever tool. Relevant documents are the queried from the vectorstore by the tool and sent to the LLM for response generation.

graph TD
    __start__ --> router_agent
    router_agent -.-> retrieve_general
    router_agent -.-> retrieve_cmds
    router_agent -.-> retrieve_install
    router_agent -.-> retrieve_errinfo
    router_agent -.-> retrieve_opensta
    router_agent -.-> retrieve_yosys
    retrieve_general --> generate
    retrieve_cmds --> generate
    retrieve_install --> generate
    retrieve_errinfo --> generate
    retrieve_opensta --> generate
    retrieve_yosys --> generate
    generate --> __end__

Loading

Tests

1. Ruff: Auto-formatter and checker for python

pip install ruff
ruff format && ruff check

2. Mypy: A static type checker for python

pip install mypy 
mypy .

To install it as a pre-commit hook:

pip install pre-commit
pre-commit install