InternLM/HuixiangDou

🎚️ Upgrade

HuixiangDou2(ACL25) is a GraphRAG solution whose effectiveness has been demonstrated in the plant-science domain and that contributed to the cover paper in Cell Molecular Plant. If you work outside computer science, give the new release a try.

English | 简体中文

HuixiangDou1 is a professional knowledge assistant based on LLM.

Advantages:

1. Design three-stage pipelines of preprocess, rejection and response
   • chat_in_group copes with group chat scenario, answer user questions without message flooding, see 2401.08772, 2405.02817, Hybrid Retrieval and Precision Report
   • chat_with_repo for real-time streaming chat
2. No training required, with CPU-only, 2G, 10G configuration
3. Offers a complete suite of Web, Android, and pipeline source code, industrial-grade and commercially viable

Check out the scenes in which HuixiangDou are running and current public service status:

• readthedocs ChatWithAI (cpu-only) is available
• OpenXLab is using GPU and under continuous maintenance
• WeChat bot has a cost associated with WeChat integration. All code has been verified to be functional for one year. Please deploy it on your own for either the free or commercial version.

If this helps you, please give it a star ⭐

🔆 New Features

Our Web version has been released to OpenXLab, where you can create knowledge base, update positive and negative examples, turn on web search, test chat, and integrate into Feishu/WeChat groups. See BiliBili and YouTube !

The Web version's API for Android also supports other devices. See Python sample code.

• [2025/03] Simplify deployment and removing --standalone
• [2025/03] Forwarding multiple wechat group message
• [2024/09] Inverted indexer makes LLM prefer knowledge base🎯
• [2024/09] Code retrieval
• [2024/08] chat_with_readthedocs, see how to integrate 👍
• [2024/07] Image and text retrieval & Removal of langchain 👍
• [2024/07] Hybrid Knowledge Graph and Dense Retrieval improve 1.7% F1 score 🎯
• [2024/06] Evaluation of chunksize, splitter, and text2vec model 🎯
• [2024/05] wkteam WeChat access, parsing image & URL, support coreference resolution
• [2024/05] SFT LLM on NLP task, F1 increased by 29% 🎯

  🤗

  LoRA-Qwen1.5-14B

  LoRA-Qwen1.5-32B

  alpaca data

  arXiv

• [2024/04] RAG Annotation SFT Q&A Data and Examples
• [2024/04] Release Web Front and Back End Service Source Code 👍
• [2024/03] New Personal WeChat Integration and Prebuilt APK !
• [2024/02] [Experimental Feature] WeChat Group Integration of multimodal to achieve OCR

📖 Support Status

📦 Hardware Requirements

The following are the GPU memory requirements for different features, the difference lies only in whether the options are turned on.

Configuration Example	GPU mem Requirements	Description	Verified on Linux
config-cpu.ini	-	Use siliconcloud API for text only
[Standard Edition]config.ini	2GB	Use openai API (such as kimi, deepseek and stepfun to search for text only
config-multimodal.ini	10GB	Use openai API for LLM, image and text retrieval

🔥 Running the Standard Edition

We take the standard edition (local running LLM, text retrieval) as an introduction example. Other versions are just different in configuration options.

I. Download and install dependencies

Click to agree to the BCE model agreement, log in huggingface

huggingface-cli login

Install dependencies

# parsing `word` format requirements
apt update
apt install python-dev libxml2-dev libxslt1-dev antiword unrtf poppler-utils pstotext tesseract-ocr flac ffmpeg lame libmad0 libsox-fmt-mp3 sox libjpeg-dev swig libpulse-dev
# python requirements
pip install -r requirements.txt
# For python3.8, install faiss-gpu instead of faiss

II. Create knowledge base

We use some novels to build knowledge base and filtering questions. If you have your own documents, just put them under repodir.

Copy and execute all the following commands (including the '#' symbol).

# Download the knowledge base, we only take the some documents as example. You can put any of your own documents under `repodir`
cd HuixiangDou
mkdir repodir
cp -rf resource/data* repodir/

# Build knowledge base, this will save the features of repodir to workdir, and update the positive and negative example thresholds into `config.ini`
mkdir workdir
python3 -m huixiangdou.services.store

# You can also build knowledge base from QA pairs (CSV or JSON format)
# CSV: First column is key (question), second column is value (answer)
# JSON: {"question1": "answer1", "question2": "answer2", ...}
# python3 -m huixiangdou.services.store --qa-pair resource/data/qa_pair.csv

III. Setup LLM API and test

Set the model and api-key in config.ini. If running LLM locally, we recommend using vllm.

vllm serve /path/to/Qwen-2.5-7B-Instruct --served-model-name vllm --enable-prefix-caching --served-model-name Qwen-2.5-7B-Instruct

Here is an example of the configured config.ini:

[llm.server]
remote_type = "kimi"
remote_api_key = "sk-dp3GriuhhLXnYo0KUuWbFUWWKOXXXXXXXXXX"
remote_llm_model = "auto"

# remote_type = "step"
# remote_api_key = "5CpPyYNPhQMkIzs5SYfcdbTHXq3a72H5XXXXXXXXXXXXX"
# remote_llm_model = "auto"

# remote_type = "deepseek"
# remote_api_key = "sk-86db9a205aa9422XXXXXXXXXXXXXX"
# remote_llm_model = "deepseek-chat"

# remote_type = "vllm"
# remote_api_key = "EMPTY"
# remote_llm_model = "Qwen2.5-7B-Instruct"

# remote_type = "siliconcloud"
# remote_api_key = "sk-xxxxxxxxxxxxx"
# remote_llm_model = "alibaba/Qwen1.5-110B-Chat"

# remote_type = "ppio"
# remote_api_key = "sk-xxxxxxxxxxxxx"
# remote_llm_model = "thudm/glm-4-9b-chat"

Then run the test:

# Respond to questions related to the Hundred-Plant Garden (related to the knowledge base), but do not respond to weather questions.
python3 -m huixiangdou.main

+-----------------------+---------+--------------------------------+-----------------+
|         Query         |  State  |         Reply                  |   References    |
+=======================+=========+================================+=================+
| What is in the Hundred-Plant Garden? | success | The Hundred-Plant Garden has a rich variety of natural landscapes and life... | installation.md |
--------------------------------------------------------------------------------------
| How is the weather today?         | Init state| ..                           |                 |
+-----------------------+---------+--------------------------------+-----------------+
🔆 Input your question here, type `bye` for exit:
..

💡 Also run a simple Web UI with gradio:

python3 -m huixiangdou.gradio_ui

Or run a server to listen 23333, default pipeline is chat_with_repo:

python3 -m huixiangdou.api_server

# test async API 
curl -X POST http://127.0.0.1:23333/huixiangdou_stream  -H "Content-Type: application/json" -d '{"text": "how to install mmpose","image": ""}'
# cURL sync API
curl -X POST http://127.0.0.1:23333/huixiangdou_inference  -H "Content-Type: application/json" -d '{"text": "how to install mmpose","image": ""}'

Please update the repodir documents, good_questions and bad_questions, and try your own domain knowledge (medical, financial, power, etc.).

IV. Integration

To Feishu, WeChat group

• One-way sending to Feishu group
• Two-way Feishu group receiving and sending, recalling
• Personal WeChat Android access and Android tool
• Personal WeChat wkteam access

To web front and backend

We provide typescript front-end and python back-end source code:

• Multi-tenant management supported
• Zero programming access to Feishu and WeChat
• k8s friendly

Same as OpenXlab APP, please read the web deployment document.

To readthedocs.io

Try right-bottom button on the page and document.

🍴 Other Configurations

CPU-only Edition

If there is no GPU available, model inference can be completed using the siliconcloud API.

Taking docker miniconda+Python3.11 as an example, install CPU dependencies and run:

# Start container
docker run -v /path/to/huixiangdou:/huixiangdou -p 7860:7860 -p 23333:23333 -it continuumio/miniconda3 /bin/bash
# Install dependencies
apt update
apt install python-dev libxml2-dev libxslt1-dev antiword unrtf poppler-utils pstotext tesseract-ocr flac ffmpeg lame libmad0 libsox-fmt-mp3 sox libjpeg-dev swig libpulse-dev
python3 -m pip install -r requirements-cpu.txt
# Establish knowledge base
python3 -m huixiangdou.services.store --config_path config-cpu.ini
# Q&A test
python3 -m huixiangdou.main --config_path config-cpu.ini
# gradio UI
python3 -m huixiangdou.gradio_ui --config_path config-cpu.ini

If you find the installation too slow, a pre-installed image is provided in Docker Hub. Simply replace it when starting the docker.

10G Multimodal Edition

If you have 10G GPU mem, you can further support image and text retrieval. Just modify the model used in config.ini.

# config-multimodal.ini
# !!! Download `https://huggingface.co/BAAI/bge-visualized/blob/main/Visualized_m3.pth`    to `bge-m3` folder !!!
embedding_model_path = "BAAI/bge-m3"
reranker_model_path = "BAAI/bge-reranker-v2-minicpm-layerwise"

Note:

• You need to manually download Visualized_m3.pth to the bge-m3 directory
• Install FlagEmbedding on main branch, we have made bugfix. Here you can download bpe_simple_vocab_16e6.txt.gz
• Install requirements/multimodal.txt

Run gradio to test, see the image and text retrieval result here.

python3 tests/test_query_gradio.py

Furthermore

Please read the following topics:

• Hybrid knowledge graph and dense retrieval
• Refer to config-advanced.ini configuration to improve effects
• Group chat scenario anaphora resolution training
• Use wkteam WeChat access, integrate images, public account parsing, and anaphora resolution
• Use rag.py to annotate SFT training data

🛠️ FAQ

1. What if the robot is too cold/too chatty?

   • Fill in the questions that should be answered in the real scenario into resource/good_questions.json, and fill the ones that should be rejected into resource/bad_questions.json.
   • Adjust the theme content in repodir to ensure that the markdown documents in the main library do not contain irrelevant content.

   Re-run feature_store to update thresholds and feature libraries.

   ⚠️ You can directly modify reject_throttle in config.ini. Generally speaking, 0.5 is a high value; 0.2 is too low.

2. Launch is normal, but out of memory during runtime?

   LLM long text based on transformers structure requires more memory. At this time, kv cache quantization needs to be done on the model, such as lmdeploy quantization description. Then use docker to independently deploy Hybrid LLM Service.

3. No module named 'faiss.swigfaiss_avx2'

   locate installed faiss package

   import faiss
   print(faiss.__file__)
   # /root/.conda/envs/InternLM2_Huixiangdou/lib/python3.10/site-packages/faiss/__init__.py
   

   add soft link

   # cd your_python_path/site-packages/faiss
   cd /root/.conda/envs/InternLM2_Huixiangdou/lib/python3.10/site-packages/faiss/
   ln -s swigfaiss.py swigfaiss_avx2.py
   

🍀 Acknowledgements

• KIMI: Long text LLM, supports direct file upload
• FlagEmbedding: BAAI RAG group
• BCEmbedding: Chinese-English bilingual feature model
• Langchain-ChatChat: Application of Langchain and ChatGLM
• GrabRedEnvelope: WeChat red packet grab

📝 Citation

@misc{kong2024huixiangdou,
      title={HuiXiangDou: Overcoming Group Chat Scenarios with LLM-based Technical Assistance},
      author={Huanjun Kong and Songyang Zhang and Jiaying Li and Min Xiao and Jun Xu and Kai Chen},
      year={2024},
      eprint={2401.08772},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

@misc{kong2024labelingsupervisedfinetuningdata,
      title={Labeling supervised fine-tuning data with the scaling law}, 
      author={Huanjun Kong},
      year={2024},
      eprint={2405.02817},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2405.02817}, 
}

@misc{kong2025huixiangdou2robustlyoptimizedgraphrag,
      title={HuixiangDou2: A Robustly Optimized GraphRAG Approach}, 
      author={Huanjun Kong and Zhefan Wang and Chenyang Wang and Zhe Ma and Nanqing Dong},
      year={2025},
      eprint={2503.06474},
      archivePrefix={arXiv},
      primaryClass={cs.IR},
      url={https://arxiv.org/abs/2503.06474}, 
}