shane-huang/langchain-chatchat

Knowledge Base QA using RAG pipeline on Intel GPU (e.g., local PC with iGPU, discrete GPU such as Arc, Flex and Max) with BigDL-LLM

Langchain-Chatchat with BigDL-LLM Acceleration on Intel GPUs

Langchain-Chatchat is a RAG (Retrieval Augmented Generation) application that implements knowledge and search engine based QA. This repo is a fork of chatchat-space/Langchain-Chatchat, and includes BigDL-LLM optimizations to run it on Intel GPU (e.g., local PC with iGPU, discrete GPU such as Arc, Flex and Max).

You can change the UI language in the left-side menu. We currently support English and 简体中文 (see video demos below).

English	简体中文
Langchain-chatchat-en.mp4	Langchain-chatchat-chs.mp4

The following sections introduce how to install and run Langchain-chatchat on Intel Core Ultra platform (MTL), utilizing the iGPU to run both LLMs and embedding models.

Table of Contents

1. RAG Architecture
2. Installation
3. One-time Warmup
4. Start the Service
5. How to Use

Langchain-Chatchat Architecture

See the RAG pipeline in the Langchain-Chatchat architecture below (source).

Installation

Download Langchain-Chatchat

Download the Langchain-Chatchat with BigDL-LLM integrations from this link. Unzip the content into a directory, e.g.,C:\Users\arda\Downloads\Langchain-Chatchat-bigdl-llm.

Install Prerequisites

Visit the Install BigDL-LLM on Windows with Intel GPU Guide, and follow Install Prerequisites to install Visual Studio, GPU driver, oneAPI, and Conda.

Install Python Dependencies

1. Open Anaconda Prompt (miniconda3), and run the following commands to create a new python environment:

   conda create -n bigdl-langchain-chatchat python=3.11 libuv 
   conda activate bigdl-langchain-chatchat

   Note: When creating the conda environment we used python 3.11, which is different from the default recommended python version 3.9 in Install BigDL-LLM on Windows with Intel GPU

2. Install bigdl-llm

   pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu
   pip install --pre --upgrade torchaudio==2.1.0a0  -f https://developer.intel.com/ipex-whl-stable-xpu

3. Switch to the root directory of Langchain-Chatchat you've downloaded (refer to the download section), and install the dependencies with the commands below. Note: In the example commands we assume the root directory is C:\Users\arda\Downloads\Langchain-Chatchat-bigdl-llm. Remember to change it to your own path).

   cd C:\Users\arda\Downloads\Langchain-Chatchat-bigdl-llm
   pip install -r requirements_bigdl.txt 
   pip install -r requirements_api_bigdl.txt
   pip install -r requirements_webui.txt

Configuration

• In root directory of Langchain-Chatchat, run the following command to create a config:

  python copy_config_example.py

• Edit the file configs\model_config.py, change MODEL_ROOT_PATH to the absolute path where you put the downloaded models (LLMs, embedding models, ranking models, etc.)

Download Models

Download the models and place them in the path MODEL_ROOT_PATH (refer to details in Configuration section).

Currently, we support only the LLM/embedding models specified in the table below. You can download these models using the link provided in the table. Note: Ensure the model folder name matches the last segment of the model ID following "/", for example, for THUDM/chatglm3-6b, the model folder name should be chatglm3-6b.

Model	Category	download link
THUDM/chatglm3-6b	Chinese LLM	HF or ModelScope
meta-llama/Llama-2-7b-chat-hf	English LLM	HF
BAAI/bge-large-zh-v1.5	Chinese Embedding	HF
BAAI/bge-large-en-v1.5	English Embedding	HF

One-time Warm-up

When you run this applcation on Intel GPU for the first time, it is highly recommended to do a one-time warmup (for GPU kernels compilation).

In Anaconda Prompt (miniconda3), under the root directory of Langchain-Chatchat, with conda environment activated, run the following commands:

python warmup.py

Note: The warmup may take several minutes. You just have to run it one-time on after installation.

Start the Service

Open Anaconda Prompt (miniconda3) and run the following commands:

conda activate bigdl-langchain-chatchat
call "C:\Program Files (x86)\Intel\oneAPI\setvars.bat"
set SYCL_CACHE_PERSISTENT=1
set BIGDL_LLM_XMX_DISABLED=1
set no_proxy=localhost,127.0.0.1
python startup.py -a

You can find the Web UI's URL printted on the terminal logs, e.g. http://localhost:8501/.

Open a browser and navigate to the URL to use the Web UI.

Usage

To start chatting with LLMs, simply type your messages in the textbox at the bottom of the UI.

How to use RAG

Step 1: Create Knowledge Base

• Select Manage Knowledge Base from the menu on the left, then choose New Knowledge Base from the dropdown menu on the right side.

  

• Fill in the name of your new knowledge base (example: "test") and press the Create button. Adjust any other settings as needed.

  

• Upload knowledge files from your computer and allow some time for the upload to complete. Once finished, click on Add files to Knowledge Base button to build the vector store. Note: this process may take several minutes.

  

Step 2: Chat with RAG

You can now click Dialogue on the left-side menu to return to the chat UI. Then in Knowledge base settings menu, choose the Knowledge Base you just created, e.g, "test". Now you can start chatting.

For more information about how to use Langchain-Chatchat, refer to Official Quickstart guide in English, Chinese, or the Wiki.