GenAIExamples/ChatQnA

ChatQnA Application

Chatbots are the most widely adopted use case for leveraging the powerful chat and reasoning capabilities of large language models (LLMs). The retrieval augmented generation (RAG) architecture is quickly becoming the industry standard for chatbots development. It combines the benefits of a knowledge base (via a vector store) and generative models to reduce hallucinations, maintain up-to-date information, and leverage domain-specific knowledge.

RAG bridges the knowledge gap by dynamically fetching relevant information from external sources, ensuring that responses generated remain factual and current. The core of this architecture are vector databases, which are instrumental in enabling efficient and semantic retrieval of information. These databases store data as vectors, allowing RAG to swiftly access the most pertinent documents or data points based on semantic similarity.

ChatQnA architecture shows below:

ChatQnA is implemented on top of GenAIComps, the ChatQnA Flow Chart shows below:

---
config:
  flowchart:
    nodeSpacing: 100
    rankSpacing: 100
    curve: linear
  theme: base
  themeVariables:
    fontSize: 42px
---
flowchart LR
    %% Colors %%
    classDef blue fill:#ADD8E6,stroke:#ADD8E6,stroke-width:2px,fill-opacity:0.5
    classDef orange fill:#FBAA60,stroke:#ADD8E6,stroke-width:2px,fill-opacity:0.5
    classDef orchid fill:#C26DBC,stroke:#ADD8E6,stroke-width:2px,fill-opacity:0.5
    classDef invisible fill:transparent,stroke:transparent;
    style ChatQnA-MegaService stroke:#000000
    %% Subgraphs %%
    subgraph ChatQnA-MegaService["ChatQnA-MegaService"]
        direction LR
        EM([Embedding <br>]):::blue
        RET([Retrieval <br>]):::blue
        RER([Rerank <br>]):::blue
        LLM([LLM <br>]):::blue
    end
    subgraph User Interface
        direction TB
        a([User Input Query]):::orchid
        Ingest([Ingest data]):::orchid
        UI([UI server<br>]):::orchid
    end
    subgraph ChatQnA GateWay
        direction LR
        invisible1[ ]:::invisible
        GW([ChatQnA GateWay<br>]):::orange
    end
    subgraph .
        X([OPEA Micsrservice]):::blue
        Y{{Open Source Service}}
        Z([OPEA Gateway]):::orange
        Z1([UI]):::orchid
    end

    TEI_RER{{Reranking service<br>}}
    TEI_EM{{Embedding service <br>}}
    VDB{{Vector DB<br><br>}}
    R_RET{{Retriever service <br>}}
    DP([Data Preparation<br>]):::blue
    LLM_gen{{LLM Service <br>}}

    %% Data Preparation flow
    %% Ingest data flow
    direction LR
    Ingest[Ingest data] -->|a| UI
    UI -->|b| DP
    DP <-.->|c| TEI_EM

    %% Questions interaction
    direction LR
    a[User Input Query] -->|1| UI
    UI -->|2| GW
    GW <==>|3| ChatQnA-MegaService
    EM ==>|4| RET
    RET ==>|5| RER
    RER ==>|6| LLM


    %% Embedding service flow
    direction TB
    EM <-.->|3'| TEI_EM
    RET <-.->|4'| R_RET
    RER <-.->|5'| TEI_RER
    LLM <-.->|6'| LLM_gen

    direction TB
    %% Vector DB interaction
    R_RET <-.->|d|VDB
    DP <-.->|d|VDB

This ChatQnA use case performs RAG using LangChain, Redis VectorDB and Text Generation Inference on Intel Gaudi2 or Intel XEON Scalable Processors. The Intel Gaudi2 accelerator supports both training and inference for deep learning models in particular for LLMs. Visit Habana AI products for more details.

In the below, we provide a table that describes for each microservice component in the ChatQnA architecture, the default configuration of the open source project, hardware, port, and endpoint.

Gaudi default compose.yaml

MicroService	Open Source Project	HW	Port	Endpoint
Embedding	Langchain	Xeon	6000	/v1/embaddings
Retriever	Langchain, Redis	Xeon	7000	/v1/retrieval
Reranking	Langchain, TEI	Gaudi	8000	/v1/reranking
LLM	Langchain, TGI	Gaudi	9000	/v1/chat/completions
Dataprep	Redis, Langchain	Xeon	6007	/v1/dataprep

Deploy ChatQnA Service

The ChatQnA service can be effortlessly deployed on either Intel Gaudi2 or Intel XEON Scalable Processors.

Two types of ChatQnA pipeline are supported now: ChatQnA with/without Rerank. And the ChatQnA without Rerank pipeline (including Embedding, Retrieval, and LLM) is offered for Xeon customers who can not run rerank service on HPU yet require high performance and accuracy.

Prepare Docker Image

Currently we support two ways of deploying ChatQnA services with docker compose:

1. Using the docker image on docker hub:

   docker pull opea/chatqna:latest
   

   Two type of UI are supported now, choose one you like and pull the referred docker image.

   If you choose conversational UI, follow the instruction and modify the compose.yaml.

   docker pull opea/chatqna-ui:latest
   # or
   docker pull opea/chatqna-conversation-ui:latest
   

2. Using the docker images built from source: Guide

   Note: The opea/chatqna-without-rerank:latest docker image has not been published yet, users need to build this docker image from source.

Required Models

By default, the embedding, reranking and LLM models are set to a default value as listed below:

Service	Model
Embedding	BAAI/bge-base-en-v1.5
Reranking	BAAI/bge-reranker-base
LLM	Intel/neural-chat-7b-v3-3

Change the xxx_MODEL_ID in docker_compose/xxx/set_env.sh for your needs.

For customers with proxy issues, the models from ModelScope are also supported in ChatQnA. Refer to this readme for details.

Setup Environment Variable

To set up environment variables for deploying ChatQnA services, follow these steps:

1. Set the required environment variables:

   # Example: host_ip="192.168.1.1"
   export host_ip="External_Public_IP"
   # Example: no_proxy="localhost, 127.0.0.1, 192.168.1.1"
   export no_proxy="Your_No_Proxy"
   export HUGGINGFACEHUB_API_TOKEN="Your_Huggingface_API_Token"
   

2. If you are in a proxy environment, also set the proxy-related environment variables:

   export http_proxy="Your_HTTP_Proxy"
   export https_proxy="Your_HTTPs_Proxy"
   

3. Set up other environment variables:

   Notice that you can only choose one command below to set up envs according to your hardware. Other that the port numbers may be set incorrectly.

   # on Gaudi
   source ./docker_compose/intel/hpu/gaudi/set_env.sh
   # on Xeon
   source ./docker_compose/intel/cpu/xeon/set_env.sh
   # on Nvidia GPU
   source ./docker_compose/nvidia/gpu/set_env.sh
   

Deploy ChatQnA on Gaudi

Find the corresponding compose.yaml.

cd GenAIExamples/ChatQnA/docker_compose/intel/hpu/gaudi/
docker compose up -d

Notice: Currently only the Habana Driver 1.16.x is supported for Gaudi.

Refer to the Gaudi Guide to build docker images from source.

Deploy ChatQnA on Xeon

Find the corresponding compose.yaml.

cd GenAIExamples/ChatQnA/docker_compose/intel/cpu/xeon/
docker compose up -d

Refer to the Xeon Guide for more instructions on building docker images from source.

Deploy ChatQnA on NVIDIA GPU

cd GenAIExamples/ChatQnA/docker_compose/nvidia/gpu/
docker compose up -d

Refer to the NVIDIA GPU Guide for more instructions on building docker images from source.

Deploy ChatQnA into Kubernetes on Xeon & Gaudi with GMC

Refer to the Kubernetes Guide for instructions on deploying ChatQnA into Kubernetes on Xeon & Gaudi with GMC.

Deploy ChatQnA into Kubernetes on Xeon & Gaudi without GMC

Refer to the Kubernetes Guide for instructions on deploying ChatQnA into Kubernetes on Xeon & Gaudi without GMC.

Deploy ChatQnA into Kubernetes using Helm Chart

Install Helm (version >= 3.15) first. Refer to the Helm Installation Guide for more information.

Refer to the ChatQnA helm chart for instructions on deploying ChatQnA into Kubernetes on Xeon & Gaudi.

Deploy ChatQnA on AI PC

Refer to the AI PC Guide for instructions on deploying ChatQnA on AI PC.

Deploy ChatQnA on Red Hat OpenShift Container Platform (RHOCP)

Refer to the Intel Technology enabling for Openshift readme for instructions to deploy ChatQnA prototype on RHOCP with Red Hat OpenShift AI (RHOAI).

Consume ChatQnA Service

Before consuming ChatQnA Service, make sure the TGI/vLLM service is ready (which takes up to 2 minutes to start).

# TGI example
docker logs tgi-service | grep Connected

Consume ChatQnA service until you get the TGI response like below.

2024-09-03T02:47:53.402023Z  INFO text_generation_router::server: router/src/server.rs:2311: Connected

Two ways of consuming ChatQnA Service:

1. Use cURL command on terminal

   curl http://${host_ip}:8888/v1/chatqna \
       -H "Content-Type: application/json" \
       -d '{
           "messages": "What is the revenue of Nike in 2023?"
       }'
   

2. Access via frontend

   To access the frontend, open the following URL in your browser: http://{host_ip}:5173

   By default, the UI runs on port 5173 internally.

   If you choose conversational UI, use this URL: http://{host_ip}:5174

Troubleshooting

1. If you get errors like "Access Denied", validate micro service first. A simple example:

   http_proxy="" curl ${host_ip}:6006/embed -X POST  -d '{"inputs":"What is Deep Learning?"}' -H 'Content-Type: application/json'
   

2. (Docker only) If all microservices work well, check the port ${host_ip}:8888, the port may be allocated by other users, you can modify the compose.yaml.

3. (Docker only) If you get errors like "The container name is in use", change container name in compose.yaml.

Monitoring OPEA Service with Prometheus and Grafana dashboard

OPEA microservice deployment can easily be monitored through Grafana dashboards in conjunction with Prometheus data collection. Follow the README to setup Prometheus and Grafana servers and import dashboards to monitor the OPEA service.