Inference Quick Start

This guide shows how to set up the environment and run inference.

1. Environment Setup

• Recommended Python version: 3.12.0 (using other versions may cause dependency issues).

# Example with Conda
conda create -n react_infer_env python=3.12.0
conda activate react_infer_env

2. Installation

Install the required dependencies:

pip install -r requirements.txt

3. Environment Configuration

Environment Configuration

Edit the .env file to configure your API keys and tool-related settings:

• SERPER_KEY_ID: Get your key from Serper.dev for web search and Google Scholar
• JINA_API_KEYS: Get your key from Jina.ai for web page reading
• API_KEY/API_BASE: OpenAI-compatible API for page summarization from OpenAI
• Proxy(Optional): Configure the proxy settings

4. Run the Inference Script

You can modify the settings in run_demo.sh.

• MODEL_PATH: Path to your model weights
• QUESTION: Run DeepResearch inference on the given QUESTION
• OUTPUT_FILE: Directory for saving results

Use the following command to start inference:

bash run_demo.sh

Web Demo (Optional)

If you prefer a web-based interface for visualizing and interacting with DeepResearch, you can launch the web demo with a single command:

bash run_web_demo.sh

Acknowledgements

The tool-call in this project is partially inspired by the Tongyi DeepResearch work.