[AudioQnA] Enable vLLM and set it as default LLM serving (opea-project#1657)

Signed-off-by: Wang, Kai Lawrence <[email protected]>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

Author: wangkl2

AudioQnA/audioqna.py

@@ -16,7 +16,7 @@
 SPEECHT5_SERVER_PORT = int(os.getenv("SPEECHT5_SERVER_PORT", 7055))
 LLM_SERVER_HOST_IP = os.getenv("LLM_SERVER_HOST_IP", "0.0.0.0")
 LLM_SERVER_PORT = int(os.getenv("LLM_SERVER_PORT", 3006))
-LLM_MODEL_ID = os.getenv("LLM_MODEL_ID", "Intel/neural-chat-7b-v3-3")
+LLM_MODEL_ID = os.getenv("LLM_MODEL_ID", "meta-llama/Meta-Llama-3-8B-Instruct")
 
 
 def align_inputs(self, inputs, cur_node, runtime_graph, llm_parameters_dict, **kwargs):

AudioQnA/audioqna_multilang.py

@@ -17,7 +17,7 @@
 GPT_SOVITS_SERVER_PORT = int(os.getenv("GPT_SOVITS_SERVER_PORT", 9088))
 LLM_SERVER_HOST_IP = os.getenv("LLM_SERVER_HOST_IP", "0.0.0.0")
 LLM_SERVER_PORT = int(os.getenv("LLM_SERVER_PORT", 8888))
-LLM_MODEL_ID = os.getenv("LLM_MODEL_ID", "Intel/neural-chat-7b-v3-3")
+LLM_MODEL_ID = os.getenv("LLM_MODEL_ID", "meta-llama/Meta-Llama-3-8B-Instruct")
 
 
 def align_inputs(self, inputs, cur_node, runtime_graph, llm_parameters_dict, **kwargs):

AudioQnA/docker_compose/intel/cpu/xeon/README.md

@@ -2,6 +2,10 @@
 
 This document outlines the deployment process for a AudioQnA application utilizing the [GenAIComps](https://github.com/opea-project/GenAIComps.git) microservice pipeline on Intel Xeon server.
 
+The default pipeline deploys with vLLM as the LLM serving component. It also provides options of using TGI backend for LLM microservice, please refer to [Start the MegaService](#-start-the-megaservice) section in this page.
+
+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](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) or you've downloaded the model locally from [ModelScope](https://www.modelscope.cn/models).
+
 ## 🚀 Build Docker images
 
 ### 1. Source Code install GenAIComps
@@ -17,9 +21,15 @@ cd GenAIComps
 docker build -t opea/whisper:latest --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -f comps/asr/src/integrations/dependency/whisper/Dockerfile .
 ```
 
-### 3. Build LLM Image
+### 3. Build vLLM Image
 
-Intel Xeon optimized image hosted in huggingface repo will be used for TGI service: ghcr.io/huggingface/text-generation-inference:2.4.0-intel-cpu (https://github.com/huggingface/text-generation-inference)
+```bash
+git clone https://github.com/vllm-project/vllm.git
+cd ./vllm/
+VLLM_VER="$(git describe --tags "$(git rev-list --tags --max-count=1)" )"
+git checkout ${VLLM_VER}
+docker build --no-cache --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -f Dockerfile.cpu -t opea/vllm:latest --shm-size=128g .
+```
 
 ### 4. Build TTS Image
 
@@ -43,9 +53,10 @@ docker build --no-cache -t opea/audioqna:latest --build-arg https_proxy=$https_p
 Then run the command `docker images`, you will have following images ready:
 
 1. `opea/whisper:latest`
-2. `opea/speecht5:latest`
-3. `opea/audioqna:latest`
-4. `opea/gpt-sovits:latest` (optional)
+2. `opea/vllm:latest`
+3. `opea/speecht5:latest`
+4. `opea/audioqna:latest`
+5. `opea/gpt-sovits:latest` (optional)
 
 ## 🚀 Set the environment variables
 
@@ -55,7 +66,7 @@ Before starting the services with `docker compose`, you have to recheck the foll
 export host_ip=<your External Public IP>    # export host_ip=$(hostname -I | awk '{print $1}')
 export HUGGINGFACEHUB_API_TOKEN=<your HF token>
 
-export LLM_MODEL_ID=Intel/neural-chat-7b-v3-3
+export LLM_MODEL_ID="meta-llama/Meta-Llama-3-8B-Instruct"
 
 export MEGA_SERVICE_HOST_IP=${host_ip}
 export WHISPER_SERVER_HOST_IP=${host_ip}
@@ -73,40 +84,90 @@ export BACKEND_SERVICE_ENDPOINT=http://${host_ip}:3008/v1/audioqna
 
 or use set_env.sh file to setup environment variables.
 
-Note: Please replace with host_ip with your external IP address, do not use localhost.
+Note:
+
+- Please replace with host_ip with your external IP address, do not use localhost.
+- 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"
+# Example: no_proxy="localhost, 127.0.0.1, 192.168.1.1"
+export no_proxy="Your_No_Proxy",${host_ip},whisper-service,speecht5-service,gpt-sovits-service,tgi-service,vllm-service,audioqna-xeon-backend-server,audioqna-xeon-ui-server
+```
 
 ## 🚀 Start the MegaService
 
 ```bash
 cd GenAIExamples/AudioQnA/docker_compose/intel/cpu/xeon/
+```
+
+If use vLLM as the LLM serving backend:
+
+```
 docker compose up -d
 
 # multilang tts (optional)
 docker compose -f compose_multilang.yaml up -d
 ```
 
+If use TGI as the LLM serving backend:
+
+```
+docker compose -f compose_tgi.yaml up -d
+```
+
 ## 🚀 Test MicroServices
 
-```bash
-# 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://${host_ip}:7066/v1/audio/transcriptions \
-  -H "Content-Type: multipart/form-data" \
-  -F file="@./sample.wav" \
-  -F model="openai/whisper-small"
-
-# tgi service
-curl http://${host_ip}:3006/generate \
-  -X POST \
-  -d '{"inputs":"What is Deep Learning?","parameters":{"max_new_tokens":17, "do_sample": true}}' \
-  -H 'Content-Type: application/json'
+1. Whisper Service
 
-# speecht5 service
-curl http://${host_ip}:7055/v1/audio/speech -XPOST -d '{"input": "Who are you?"}' -H 'Content-Type: application/json' --output speech.mp3
+   ```bash
+   wget https://github.com/intel/intel-extension-for-transformers/raw/main/intel_extension_for_transformers/neural_chat/assets/audio/sample.wav
+   curl http://${host_ip}:${WHISPER_SERVER_PORT}/v1/audio/transcriptions \
+     -H "Content-Type: multipart/form-data" \
+     -F file="@./sample.wav" \
+     -F model="openai/whisper-small"
+   ```
 
-# gpt-sovits service (optional)
-curl http://${host_ip}:9880/v1/audio/speech -XPOST -d '{"input": "Who are you?"}' -H 'Content-Type: application/json' --output speech.mp3
-```
+2. LLM backend Service
+
+   In the first startup, this service will take more time to download, load and warm up the model. After it's finished, the service will be ready and the container (`vllm-service` or `tgi-service`) status shown via `docker ps` will be `healthy`. Before that, the status will be `health: starting`.
+
+   Or try the command below to check whether the LLM serving is ready.
+
+   ```bash
+   # vLLM service
+   docker logs vllm-service 2>&1 | grep complete
+   # If the service is ready, you will get the response like below.
+   INFO:     Application startup complete.
+   ```
+
+   ```bash
+   # TGI service
+   docker logs tgi-service | grep Connected
+   # If the service is ready, you will get the response like below.
+   2024-09-03T02:47:53.402023Z  INFO text_generation_router::server: router/src/server.rs:2311: Connected
+   ```
+
+   Then try the `cURL` command below to validate services.
+
+   ```bash
+   # either vLLM or TGI service
+   curl http://${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
+   ```
 
 ## 🚀 Test MegaService
 

AudioQnA/docker_compose/intel/cpu/xeon/compose.yaml

@@ -6,7 +6,7 @@ services:
     image: ${REGISTRY:-opea}/whisper:${TAG:-latest}
     container_name: whisper-service
     ports:
-      - "7066:7066"
+      - ${WHISPER_SERVER_PORT:-7066}:7066
     ipc: host
     environment:
       no_proxy: ${no_proxy}
@@ -17,38 +17,41 @@ services:
     image: ${REGISTRY:-opea}/speecht5:${TAG:-latest}
     container_name: speecht5-service
     ports:
-      - "7055:7055"
+      - ${SPEECHT5_SERVER_PORT:-7055}:7055
     ipc: host
     environment:
       no_proxy: ${no_proxy}
       http_proxy: ${http_proxy}
       https_proxy: ${https_proxy}
     restart: unless-stopped
-  tgi-service:
-    image: ghcr.io/huggingface/text-generation-inference:2.4.0-intel-cpu
-    container_name: tgi-service
+  vllm-service:
+    image: ${REGISTRY:-opea}/vllm:${TAG:-latest}
+    container_name: vllm-service
     ports:
-      - "3006:80"
+      - ${LLM_SERVER_PORT:-3006}:80
     volumes:
-      - "${MODEL_CACHE:-./data}:/data"
-    shm_size: 1g
+      - "${MODEL_CACHE:-./data}:/root/.cache/huggingface/hub"
+    shm_size: 128g
     environment:
       no_proxy: ${no_proxy}
       http_proxy: ${http_proxy}
       https_proxy: ${https_proxy}
       HF_TOKEN: ${HUGGINGFACEHUB_API_TOKEN}
+      LLM_MODEL_ID: ${LLM_MODEL_ID}
+      VLLM_TORCH_PROFILER_DIR: "/mnt"
+      LLM_SERVER_PORT: ${LLM_SERVER_PORT}
     healthcheck:
-      test: ["CMD-SHELL", "curl -f http://$host_ip:3006/health || exit 1"]
+      test: ["CMD-SHELL", "curl -f http://$host_ip:${LLM_SERVER_PORT}/health || exit 1"]
       interval: 10s
       timeout: 10s
       retries: 100
-    command: --model-id ${LLM_MODEL_ID} --cuda-graphs 0
+    command: --model ${LLM_MODEL_ID} --host 0.0.0.0 --port 80
   audioqna-xeon-backend-server:
     image: ${REGISTRY:-opea}/audioqna:${TAG:-latest}
     container_name: audioqna-xeon-backend-server
     depends_on:
       - whisper-service
-      - tgi-service
+      - vllm-service
       - speecht5-service
     ports:
       - "3008:8888"

AudioQnA/docker_compose/intel/cpu/xeon/compose_multilang.yaml

@@ -6,7 +6,7 @@ services:
     image: ${REGISTRY:-opea}/whisper:${TAG:-latest}
     container_name: whisper-service
     ports:
-      - "7066:7066"
+      - ${WHISPER_SERVER_PORT:-7066}:7066
     ipc: host
     environment:
       no_proxy: ${no_proxy}
@@ -18,27 +18,35 @@ services:
     image: ${REGISTRY:-opea}/gpt-sovits:${TAG:-latest}
     container_name: gpt-sovits-service
     ports:
-      - "9880:9880"
+      - ${GPT_SOVITS_SERVER_PORT:-9880}:9880
     ipc: host
     environment:
       no_proxy: ${no_proxy}
       http_proxy: ${http_proxy}
       https_proxy: ${https_proxy}
     restart: unless-stopped
-  tgi-service:
-    image: ghcr.io/huggingface/text-generation-inference:2.4.0-intel-cpu
-    container_name: tgi-service
+  vllm-service:
+    image: ${REGISTRY:-opea}/vllm:${TAG:-latest}
+    container_name: vllm-service
     ports:
-      - "3006:80"
+      - ${LLM_SERVER_PORT:-3006}:80
     volumes:
-      - "${MODEL_CACHE:-./data}:/data"
-    shm_size: 1g
+      - "${MODEL_CACHE:-./data}:/root/.cache/huggingface/hub"
+    shm_size: 128g
     environment:
       no_proxy: ${no_proxy}
       http_proxy: ${http_proxy}
       https_proxy: ${https_proxy}
       HF_TOKEN: ${HUGGINGFACEHUB_API_TOKEN}
-    command: --model-id ${LLM_MODEL_ID} --cuda-graphs 0
+      LLM_MODEL_ID: ${LLM_MODEL_ID}
+      VLLM_TORCH_PROFILER_DIR: "/mnt"
+      LLM_SERVER_PORT: ${LLM_SERVER_PORT}
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://$host_ip:${LLM_SERVER_PORT}/health || exit 1"]
+      interval: 10s
+      timeout: 10s
+      retries: 100
+    command: --model ${LLM_MODEL_ID} --host 0.0.0.0 --port 80
   audioqna-xeon-backend-server:
     image: ${REGISTRY:-opea}/audioqna-multilang:${TAG:-latest}
     container_name: audioqna-xeon-backend-server

AudioQnA/docker_compose/intel/cpu/xeon/compose_tgi.yaml

@@ -0,0 +1,87 @@
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+services:
+  whisper-service:
+    image: ${REGISTRY:-opea}/whisper:${TAG:-latest}
+    container_name: whisper-service
+    ports:
+      - ${WHISPER_SERVER_PORT:-7066}:7066
+    ipc: host
+    environment:
+      no_proxy: ${no_proxy}
+      http_proxy: ${http_proxy}
+      https_proxy: ${https_proxy}
+    restart: unless-stopped
+  speecht5-service:
+    image: ${REGISTRY:-opea}/speecht5:${TAG:-latest}
+    container_name: speecht5-service
+    ports:
+      - ${SPEECHT5_SERVER_PORT:-7055}:7055
+    ipc: host
+    environment:
+      no_proxy: ${no_proxy}
+      http_proxy: ${http_proxy}
+      https_proxy: ${https_proxy}
+    restart: unless-stopped
+  tgi-service:
+    image: ghcr.io/huggingface/text-generation-inference:2.4.0-intel-cpu
+    container_name: tgi-service
+    ports:
+      - ${LLM_SERVER_PORT:-3006}:80
+    volumes:
+      - "${MODEL_CACHE:-./data}:/data"
+    shm_size: 1g
+    environment:
+      no_proxy: ${no_proxy}
+      http_proxy: ${http_proxy}
+      https_proxy: ${https_proxy}
+      HF_TOKEN: ${HUGGINGFACEHUB_API_TOKEN}
+      LLM_SERVER_PORT: ${LLM_SERVER_PORT}
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://$host_ip:${LLM_SERVER_PORT}/health || exit 1"]
+      interval: 10s
+      timeout: 10s
+      retries: 100
+    command: --model-id ${LLM_MODEL_ID} --cuda-graphs 0
+  audioqna-xeon-backend-server:
+    image: ${REGISTRY:-opea}/audioqna:${TAG:-latest}
+    container_name: audioqna-xeon-backend-server
+    depends_on:
+      - whisper-service
+      - tgi-service
+      - speecht5-service
+    ports:
+      - "3008:8888"
+    environment:
+      - no_proxy=${no_proxy}
+      - https_proxy=${https_proxy}
+      - http_proxy=${http_proxy}
+      - MEGA_SERVICE_HOST_IP=${MEGA_SERVICE_HOST_IP}
+      - WHISPER_SERVER_HOST_IP=${WHISPER_SERVER_HOST_IP}
+      - WHISPER_SERVER_PORT=${WHISPER_SERVER_PORT}
+      - LLM_SERVER_HOST_IP=${LLM_SERVER_HOST_IP}
+      - LLM_SERVER_PORT=${LLM_SERVER_PORT}
+      - LLM_MODEL_ID=${LLM_MODEL_ID}
+      - SPEECHT5_SERVER_HOST_IP=${SPEECHT5_SERVER_HOST_IP}
+      - SPEECHT5_SERVER_PORT=${SPEECHT5_SERVER_PORT}
+    ipc: host
+    restart: always
+  audioqna-xeon-ui-server:
+    image: ${REGISTRY:-opea}/audioqna-ui:${TAG:-latest}
+    container_name: audioqna-xeon-ui-server
+    depends_on:
+      - audioqna-xeon-backend-server
+    ports:
+      - "5173:5173"
+    environment:
+      - no_proxy=${no_proxy}
+      - https_proxy=${https_proxy}
+      - http_proxy=${http_proxy}
+      - CHAT_URL=${BACKEND_SERVICE_ENDPOINT}
+    ipc: host
+    restart: always
+
+networks:
+  default:
+    driver: bridge

AudioQnA/docker_compose/intel/cpu/xeon/set_env.sh

@@ -8,7 +8,7 @@ export host_ip=$(hostname -I | awk '{print $1}')
 export HUGGINGFACEHUB_API_TOKEN=${HUGGINGFACEHUB_API_TOKEN}
 # <token>
 
-export LLM_MODEL_ID=Intel/neural-chat-7b-v3-3
+export LLM_MODEL_ID="meta-llama/Meta-Llama-3-8B-Instruct"
 
 export MEGA_SERVICE_HOST_IP=${host_ip}
 export WHISPER_SERVER_HOST_IP=${host_ip}