Deploying AudioQnA on AMD EPYC™ Processors

This document provides a step-by-step guide for deploying the AudioQnA application on a single node, leveraging the GenAIComps microservices, optimized for AMD EPYC™ Processors. The process covers pulling Docker images, deploying containers using Docker Compose, and running services with the llm microservices.

Note: The default LLM is meta-llama/Meta-Llama-3-8B-Instruct. Before deploying the application, please make sure either you’ve requested and been granted the access to it on Huggingface or you’ve downloaded the model locally from ModelScope.

Table of Contents

  1. AudioQnA Quick Start Deployment

  2. AudioQnA Docker Compose Files

  3. Validate Microservices

  4. Conclusion

AudioQnA Quick Start Deployment

This section describes how to quickly deploy and test the AudioQnA service manually on an AMD EPYC™ processor. The basic steps are:

  1. Access the Code

  2. Install Docker

  3. Determine your host’s external IP address

  4. Configure the Deployment Environment

  5. Deploy the Services Using Docker Compose

  6. Check the Deployment Status

  7. Validate the Pipeline

  8. Cleanup the Deployment

Access the Code

Clone the GenAIExample repository and access the AudioQnA AMD EPYC™ platform Docker Compose files and supporting scripts:

git clone https://github.com/opea-project/GenAIExamples.git
cd GenAIExamples/AudioQnA/docker_compose/amd/cpu/epyc

Install Docker

Ensure Docker is installed on your system. If Docker is not already installed, use the provided script to set it up:

source ./install_docker.sh

This script installs Docker and its dependencies. After running it, verify the installation by checking the Docker version:

docker --version

If Docker is already installed, this step can be skipped.

Determine your host external IP address

Run the following command in your terminal to list network interfaces:

ifconfig

Look for the inet address associated with your active network interface (e.g., enp99s0). For example:

enp99s0: flags=4163<UP,BROADCAST,RUNNING,MULTICAST>  mtu 1500
    inet 10.101.16.119  netmask 255.255.255.0  broadcast 10.101.16.255

In this example, the (host_ip) would be (10.101.16.119).

# Replace with your host's external IP address
export host_ip="your_external_ip_address"

Configure the Deployment Environment

The model_cache directory, by default, stores models in the ./data directory. To change this, use the following command:

# Optional
export model_cache=/home/documentation/data_audioqna/data # Path to save cache models

To set up environment variables for deploying AudioQnA services, set up some parameters specific to the deployment environment and source the set_env.sh script in this directory:

export HF_TOKEN="Your_HuggingFace_API_Token"
export http_proxy="Your_HTTP_Proxy"           # http proxy if any
export https_proxy="Your_HTTPs_Proxy"         # https proxy if any
export no_proxy=localhost,127.0.0.1,$host_ip,whisper-service,speecht5-service,vllm-service,tgi-service,audioqna-epyc-backend-server,audioqna-epyc-ui-server  # additional no proxies if needed
export NGINX_PORT=${your_nginx_port}          # your usable port for nginx, 80 for example
source ./set_env.sh

Deploy the Services Using Docker Compose

To deploy the AudioQnA services, execute the docker compose up command with the appropriate arguments. For a default deployment, execute the command below. It uses the ‘compose.yaml’ file.

docker compose -f compose.yaml up -d

Note: developers should build docker image from source when:

  • Developing off the git main branch (as the container’s ports in the repo may be different > from the published docker image).

  • Unable to download the docker image.

  • Use a specific version of Docker image.

Please refer to the table below to build different microservices from source:

Microservice

Deployment Guide

vLLM

vLLM build guide

LLM

LLM build guide

WHISPER

Whisper build guide

SPEECHT5

SpeechT5 build guide

GPT-SOVITS

GPT-SOVITS build guide

MegaService

MegaService build guide

UI

Basic UI build guide

Check the Deployment Status

After running docker compose, check if all the containers launched via docker compose have started:

docker ps -a

For the default deployment, the following 5 containers should have started:

1c67e44c39d2   opea/audioqna-ui:latest   "docker-entrypoint.s…"   About a minute ago   Up About a minute             0.0.0.0:5173->5173/tcp, :::5173->5173/tcp   audioqna-epyc-ui-server
833a42677247   opea/audioqna:latest      "python audioqna.py"     About a minute ago   Up About a minute             0.0.0.0:3008->8888/tcp, :::3008->8888/tcp   audioqna-epyc-backend-server
5dc4eb9bf499   opea/speecht5:latest      "python speecht5_ser…"   About a minute ago   Up About a minute             0.0.0.0:7055->7055/tcp, :::7055->7055/tcp   speecht5-service
814e6efb1166   opea/vllm:latest          "python3 -m vllm.ent…"   About a minute ago   Up About a minute (healthy)   0.0.0.0:3006->80/tcp, :::3006->80/tcp       vllm-service
46f7a00f4612   opea/whisper:latest       "python whisper_serv…"   About a minute ago   Up About a minute             0.0.0.0:7066->7066/tcp, :::7066->7066/tcp   whisper-service

If any issues are encountered during deployment, refer to the Troubleshooting section.

Validate the Pipeline

Once the AudioQnA services are running, test the pipeline using the following command:

# Test the AudioQnA megaservice by recording a .wav file, encoding the file into the base64 format, and then sending the base64 string to the megaservice endpoint.
# The megaservice will return a spoken response as a base64 string. To listen to the response, decode the base64 string and save it as a .wav file.
wget https://github.com/intel/intel-extension-for-transformers/raw/refs/heads/main/intel_extension_for_transformers/neural_chat/assets/audio/sample_2.wav
base64_audio=$(base64 -w 0 sample_2.wav)

# if you are using speecht5 as the tts service, voice can be "default" or "male"
# if you are using gpt-sovits for the tts service, you can set the reference audio following https://github.com/opea-project/GenAIComps/blob/main/comps/third_parties/gpt-sovits/src/README.md

curl http://${host_ip}:3008/v1/audioqna \
  -X POST \
  -H "Content-Type: application/json" \
  -d "{\"audio\": \"${base64_audio}\", \"max_tokens\": 64, \"voice\": \"default\"}" \
  | sed 's/^"//;s/"$//' | base64 -d > output.wav

Note : Access the AudioQnA UI by web browser through this URL: http://${host_ip}:5173. Please confirm the 5173 port is opened in the firewall. To validate each microservice used in the pipeline refer to the Validate Microservices section.

Cleanup the Deployment

To stop the containers associated with the deployment, execute the following command:

docker compose -f compose.yaml down

AudioQnA Docker Compose Files

When deploying an AudioQnA pipeline on an AMD EPYC™ platform, users can select from various large language model serving frameworks or opt for either single-language English TTS. The table below highlights the available configurations included in the application. These configurations serve as templates and can be extended to incorporate additional components from GenAIComps.

File

Description

compose.yaml

Default compose file using vllm as serving framework and redis as vector database

compose_tgi.yaml

The LLM serving framework is TGI. All other configurations remain the same as the default

Validate MicroServices

  1. Whisper Service

    wget https://github.com/intel/intel-extension-for-transformers/raw/main/intel_extension_for_transformers/neural_chat/assets/audio/sample.wav
curl http://${host_ip}:${WHISPER_SERVER_PORT}/v1/audio/transcriptions \
  -H "Content-Type: multipart/form-data" \
  -F file="@./sample.wav" \
  -F model="openai/whisper-small"

  2. LLM backend Service

    During the initial startup, the service requires additional time to download, load, and warm up the model. Once this process is complete, the service will be ready, and the container (either vllm-service or tgi-service) will display a healthy status when viewed using docker ps. Prior to this, the status will appear as health: starting.

    Or try the command below to check whether the LLM serving is ready.

    # vLLM service
docker logs vllm-service 2>&1 | grep complete
# If the service is ready, you will get the response like below.
INFO:     Application startup complete.

    # TGI service
docker logs tgi-service | grep Connected
# If the service is ready, you will get the response like below.
2024-09-03T02:47:53.402023Z  INFO text_generation_router::server: router/src/server.rs:2311: Connected

    Then try the cURL command below to validate services.

    # either vLLM or TGI service
curl http://${host_ip}:${LLM_SERVER_PORT}/v1/chat/completions \
  -X POST \
  -d '{"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "user", "content": "What is Deep Learning?"}], "max_tokens":17}' \
  -H 'Content-Type: application/json'

  3. TTS Service

    # speecht5 service
curl http://${host_ip}:${SPEECHT5_SERVER_PORT}/v1/audio/speech -XPOST -d '{"input": "Who are you?"}' -H 'Content-Type: application/json' --output speech.mp3

# gpt-sovits service (optional)
curl http://${host_ip}:${GPT_SOVITS_SERVER_PORT}/v1/audio/speech -XPOST -d '{"input": "Who are you?"}' -H 'Content-Type: application/json' --output speech.mp3

Profile Microservices

To further analyze MicroService Performance, users could follow the instructions to profile MicroServices.

1. vLLM backend Service

Users could follow previous section to testing vLLM microservice or CodeGen MegaService. By default, vLLM profiling is not enabled. Users could start and stop profiling by following commands.

Start vLLM profiling
curl http://${host_ip}:${LLM_SERVER_PORT}/start_profile \
    -X POST \
    -d '{"model": "meta-llama/Meta-Llama-3-8B-Instruct"}' \
    -H 'Content-Type: application/json'

After vLLM profiling is started, users could start asking questions and get responses from vLLM MicroService

curl http://${host_ip}:${LLM_SERVER_PORT}/v1/chat/completions \
    -X POST \
    -d '{"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "user", "content": "What is Deep Learning?"}], "max_tokens":17}' \
    -H 'Content-Type: application/json'
Stop vLLM profiling

By following command, users could stop vLLM profiling and generate a *.pt.trace.json.gz file as profiling result under /mnt folder in vllm-service docker instance.

curl http://${host_ip}:${LLM_SERVER_PORT}/stop_profile \
     -X POST \
     -d '{"model": "meta-llama/Meta-Llama-3-8B-Instruct"}' \
     -H 'Content-Type: application/json'

After vllm profiling is stopped, users could use below command to get the *.pt.trace.json.gz file under /mnt folder.

docker cp  vllm-service:/mnt/ .
Check profiling result

Open a web browser and type “chrome://tracing” or “ui.perfetto.dev”, and then load the json.gz file.

Conclusion

This guide should enable developers to deploy the default configuration or any of the other compose yaml files for different configurations. It also highlights the configurable parameters that can be set before deployment.