# KnowPath
**Repository Path**: tize/KnowPath
## Basic Information
- **Project Name**: KnowPath
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No
## Statistics
- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-05-22
- **Last Updated**: 2025-05-23
## Categories & Tags
**Categories**: Uncategorized
**Tags**: None
## README
# KnowPath
# About
`KnowPath` is a knowledge-enhanced large language model framework that effectively integrates both internal and external knowledge of LLMs to achieve superior knowledge-based question answering. Its notable advantages include:
- `Reduced LLM invocations` (fewer API calls to LLMs)
- `Lower token consumption` (decreased usage of tokens)
- Capability to generate `fully directed connected knowledge subgraphs`
The framework helps mitigate hallucination phenomena in LLM responses while significantly improving answer interpretability.
# The pipeline of KnowPath
# Project Structure
```
├── CoT
│ ├── cot_io.py # COT method run file
│ ├── prompt_list.py # Prompt of cot method
│ └── utils.py # Utility functions
├── data # The dataset used in this paper
│ ├── cwq.json
│ ├── README.md # See this for details of thest datasets
│ ├── SimpleQA.json
│ ├── WebQSP.json
│ └── WebQuestions.json
├── eval
│ ├── eval.sh # Evaluation method script
│ ├── knowpath_eval.py
│ └── utils.py
├── knowpath # Source code of KnowPath
│ ├── freebase_func_knowpath.py # Subgraph Exploration Method
│ ├── knowpath_prompt.py # Prompt of our proposed KnowPath
│ ├── main.py # Entry file
│ ├── reason_path_visualize.py # Subgraph visualization
│ ├── tempreture.sh # Analyzing Temperature Hyperparameters
│ └── utils_knowpath.py # Utility functions of KnowPath
├── README.md
└── requirements.txt # Pip environment file
```
# Get Start
- ## 1.Deploy FreeBase locally
Here we would like to acknowledge [ToG](https://github.com/GasolSun36/ToG/tree/main/Freebase) for `providing the FreeBase initialization method`. They have offered a relatively comprehensive installation and startup procedure for FreeBase, so this step can be followed by referring to their provided process.
- ## 2.Install the necessary pip packages
Run the following command to install the required dependencies:
```
pip install -r requirements.txt
```
# How to run KnowPath
```
cd knowpath
python main.py --dataset cwq \ # choose one of the four datasets
--LLM_type qwen2 \ # you can use local model via ollama or api
--max_length 512 \
--max_depth 3 \ # Maximum exploration depth, default is 3
--random_seed 42 \ Random seed, default is 42
--temperature_reasoning 0 \ Temperature for model inference, default is 0
```
# How to eval
```
cd eval
bash eval.sh
```
# Subgraph visualization
```
cd knowpath
python reason_path_visualize.py
```
Then you can get:smile::
# Result
# Case study
