Add VisualQnA docker for both Gaudi and Xeon using TGI serving (#547)

* Add VisualQnA docker for both Gaudi and Xeon Signed-off-by: lvliang-intel <liang1.lv@intel.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * update token length Signed-off-by: lvliang-intel <liang1.lv@intel.com> --------- Signed-off-by: lvliang-intel <liang1.lv@intel.com> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
opea-project · Aug 9, 2024 · 2390920 · 2390920
1 parent 02a1536
commit 2390920
Show file tree

Hide file tree

Showing 9 changed files with 595 additions and 39 deletions.
diff --git a/VisualQnA/README.md b/VisualQnA/README.md
@@ -18,61 +18,63 @@ This example guides you through how to deploy a [LLaVA](https://llava-vl.github.
 ![llava screenshot](./assets/img/llava_screenshot1.png)
 ![llava-screenshot](./assets/img/llava_screenshot2.png)
 
-## Start the LLaVA service
+# Deploy VisualQnA Service
 
-1. Build the Docker image needed for starting the service
+The VisualQnA service can be effortlessly deployed on either Intel Gaudi2 or Intel XEON Scalable Processors.
 
-```
-cd serving/
-docker build . --build-arg http_proxy=${http_proxy} --build-arg https_proxy=${http_proxy} -t intel/gen-ai-examples:llava-gaudi
-```
+Currently we support deploying VisualQnA services with docker compose.
 
-2. Start the LLaVA service on Intel Gaudi2
+## Setup Environment Variable
 
-```
-docker run -d -p 8085:8000 -v ./data:/root/.cache/huggingface/hub/ -e http_proxy=$http_proxy -e https_proxy=$http_proxy --runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --ipc=host intel/gen-ai-examples:llava-gaudi
-```
+To set up environment variables for deploying VisualQnA services, follow these steps:
 
-Here are some explanation about the above parameters:
+1. Set the required environment variables:
 
-- `-p 8085:8000`: This will map the 8000 port of the LLaVA service inside the container to the 8085 port on the host
-- `-v ./data:/root/.cache/huggingface/hub/`: This is to prevent from re-downloading model files
-- `http_proxy` and `https_proxy` are used if you have some proxy setting
-- `--runtime=habana ...` is required for running this service on Intel Gaudi2
+   ```bash
+   # 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"
+   ```
 
-Now you have a LLaVa service with the exposed port `8085` and you can check whether this service is up by:
+2. If you are in a proxy environment, also set the proxy-related environment variables:
 
-```
-curl localhost:8085/health -v
-```
+   ```bash
+   export http_proxy="Your_HTTP_Proxy"
+   export https_proxy="Your_HTTPs_Proxy"
+   ```
 
-If the reply has a `200 OK`, then the service is up.
+3. Set up other environment variables:
 
-## Start the Gradio app
+   > 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.
 
-Now you have two options to start the frontend UI by following commands:
+   ```bash
+   # on Gaudi
+   source ./docker/gaudi/set_env.sh
+   # on Xeon
+   source ./docker/xeon/set_env.sh
+   ```
 
-### English Interface (Default)
+## Deploy VisualQnA on Gaudi
 
-```
-cd ui/
-pip install -r requirements.txt
-http_proxy= python app.py --host 0.0.0.0 --port 7860 --worker-addr http://localhost:8085 --share
-```
+Refer to the [Gaudi Guide](./docker/gaudi/README.md) to build docker images from source.
 
-### Chinese Interface
+Find the corresponding [compose.yaml](./docker/gaudi/compose.yaml).
 
-```
-cd ui/
-pip install -r requirements.txt
-http_proxy= python app.py --host 0.0.0.0 --port 7860 --worker-addr http://localhost:8085 --lang CN --share
+```bash
+cd GenAIExamples/VisualQnA/docker/gaudi/
+docker compose up -d
 ```
 
-Here are some explanation about the above parameters:
+> Notice: Currently only the **Habana Driver 1.16.x** is supported for Gaudi.
 
-- `--host`: the host of the gradio app
-- `--port`: the port of the gradio app, by default 7860
-- `--worker-addr`: the LLaVA service IP address. If you setup the service on a different machine, please replace `localhost` to the IP address of your Gaudi2 host machine
-- `--lang`: Specify this parameter to use the Chinese interface. The default UI language is English and can be used without any additional parameter.
+## Deploy VisualQnA on Xeon
 
-SCRIPT USAGE NOTICE:  By downloading and using any script file included with the associated software package (such as files with .bat, .cmd, or .JS extensions, Docker files, or any other type of file that, when executed, automatically downloads and/or installs files onto your system) (the “Script File”), it is your obligation to review the Script File to understand what files (e.g.,  other software, AI models, AI Datasets) the Script File will download to your system (“Downloaded Files”). Furthermore, by downloading and using the Downloaded Files, even if they are installed through a silent install, you agree to any and all terms and conditions associated with such files, including but not limited to, license terms, notices, or disclaimers.
+Refer to the [Xeon Guide](./docker/xeon/README.md) for more instructions on building docker images from source.
+
+Find the corresponding [compose.yaml](./docker/xeon/compose.yaml).
+
+```bash
+cd GenAIExamples/VisualQnA/docker/xeon/
+docker compose up -d
+```
diff --git a/VisualQnA/docker/Dockerfile b/VisualQnA/docker/Dockerfile
@@ -0,0 +1,33 @@
+
+
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+FROM python:3.11-slim
+
+RUN apt-get update -y && apt-get install -y --no-install-recommends --fix-missing \
+    libgl1-mesa-glx \
+    libjemalloc-dev \
+    vim \
+    git
+
+RUN useradd -m -s /bin/bash user && \
+    mkdir -p /home/user && \
+    chown -R user /home/user/
+
+WORKDIR /home/user/
+RUN git clone https://github.com/opea-project/GenAIComps.git
+
+WORKDIR /home/user/GenAIComps
+RUN pip install --no-cache-dir --upgrade pip && \
+    pip install --no-cache-dir -r /home/user/GenAIComps/requirements.txt
+
+COPY ./visualqna.py /home/user/visualqna.py
+
+ENV PYTHONPATH=$PYTHONPATH:/home/user/GenAIComps
+
+USER user
+
+WORKDIR /home/user
+
+ENTRYPOINT ["python", "visualqna.py"]
diff --git a/VisualQnA/docker/gaudi/README.md b/VisualQnA/docker/gaudi/README.md
@@ -0,0 +1,139 @@
+# Build MegaService of VisualQnA on Gaudi
+
+This document outlines the deployment process for a VisualQnA application utilizing the [GenAIComps](https://github.com/opea-project/GenAIComps.git) microservice pipeline on Intel Gaudi server. The steps include Docker image creation, container deployment via Docker Compose, and service execution to integrate microservices such as llm. We will publish the Docker images to Docker Hub, it will simplify the deployment process for this service.
+
+## 🚀 Build Docker Images
+
+First of all, you need to build Docker Images locally. This step can be ignored after the Docker images published to Docker hub.
+
+### 1. Source Code install GenAIComps
+
+```bash
+git clone https://github.com/opea-project/GenAIComps.git
+cd GenAIComps
+```
+
+### 2. Build LLM Image
+
+```bash
+docker build --no-cache -t opea/lvm-tgi:latest --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -f comps/lvms/Dockerfile_tgi .
+```
+
+### 3. Build TGI Gaudi Image
+
+Since TGI Gaudi has not supported llava-next in main branch, we'll need to build it from a PR branch for now.
+
+```bash
+git clone https://github.com/yuanwu2017/tgi-gaudi.git
+cd tgi-gaudi/
+git checkout v2.0.4
+docker build -t opea/llava-tgi:latest .
+cd ../
+```
+
+### 4. Build MegaService Docker Image
+
+To construct the Mega Service, we utilize the [GenAIComps](https://github.com/opea-project/GenAIComps.git) microservice pipeline within the `visuralqna.py` Python script. Build the MegaService Docker image using the command below:
+
+```bash
+git clone https://github.com/opea-project/GenAIExamples.git
+cd GenAIExamples/VisualQnA/docker
+docker build --no-cache -t opea/visualqna:latest --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -f Dockerfile .
+cd ../../..
+```
+
+### 5. Build UI Docker Image
+
+Build frontend Docker image via below command:
+
+```bash
+cd GenAIExamples/VisualQnA/docker/ui/
+docker build --no-cache -t opea/visualqna-ui:latest --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -f ./docker/Dockerfile .
+cd ../../../..
+```
+
+Then run the command `docker images`, you will have the following 4 Docker Images:
+
+1. `opea/llava-tgi:latest`
+2. `opea/lvm-tgi:latest`
+3. `opea/visualqna:latest`
+4. `opea/visualqna-ui:latest`
+
+## 🚀 Start MicroServices and MegaService
+
+### Setup Environment Variables
+
+Since the `compose.yaml` will consume some environment variables, you need to setup them in advance as below.
+
+```bash
+export no_proxy=${your_no_proxy}
+export http_proxy=${your_http_proxy}
+export https_proxy=${your_http_proxy}
+export LVM_MODEL_ID="llava-hf/llava-v1.6-mistral-7b-hf"
+export LVM_ENDPOINT="http://${host_ip}:8399"
+export LVM_SERVICE_PORT=9399
+export MEGA_SERVICE_HOST_IP=${host_ip}
+export LVM_SERVICE_HOST_IP=${host_ip}
+export BACKEND_SERVICE_ENDPOINT="http://${host_ip}:8888/v1/visualqna"
+```
+
+Note: Please replace with `host_ip` with you external IP address, do **NOT** use localhost.
+
+### Start all the services Docker Containers
+
+```bash
+cd GenAIExamples/VisualQnA/docker/gaudi/
+```
+
+```bash
+docker compose -f compose.yaml up -d
+```
+
+> **_NOTE:_** Users need at least one Gaudi cards to run the VisualQnA successfully.
+
+### Validate MicroServices and MegaService
+
+Follow the instructions to validate MicroServices.
+
+1. LLM Microservice
+
+```bash
+http_proxy="" curl http://${host_ip}:9399/v1/lvm -XPOST -d '{"image": "iVBORw0KGgoAAAANSUhEUgAAAAoAAAAKCAYAAACNMs+9AAAAFUlEQVR42mP8/5+hnoEIwDiqkL4KAcT9GO0U4BxoAAAAAElFTkSuQmCC", "prompt":"What is this?"}' -H 'Content-Type: application/json'
+```
+
+2. MegaService
+
+```bash
+curl http://${host_ip}:8888/v1/visualqna -H "Content-Type: application/json" -d '{
+     "messages": [
+      {
+        "role": "user",
+        "content": [
+          {
+            "type": "text",
+            "text": "What'\''s in this image?"
+          },
+          {
+            "type": "image_url",
+            "image_url": {
+              "url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg"
+            }
+          }
+        ]
+      }
+    ],
+    "max_tokens": 300
+    }'
+```
+
+## 🚀 Launch the UI
+
+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 prefer to use a different host port to access the frontend, you can modify the port mapping in the `compose.yaml` file as shown below:
+
+```yaml
+  visualqna-gaudi-ui-server:
+    image: opea/visualqna-ui:latest
+    ...
+    ports:
+      - "80:5173"
+```
diff --git a/VisualQnA/docker/gaudi/compose.yaml b/VisualQnA/docker/gaudi/compose.yaml
@@ -0,0 +1,77 @@
+
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+version: "3.8"
+
+services:
+  llava-tgi-service:
+    image: opea/llava-tgi:latest
+    container_name: tgi-llava-gaudi-server
+    ports:
+      - "8399:80"
+    volumes:
+      - "./data:/data"
+    environment:
+      no_proxy: ${no_proxy}
+      http_proxy: ${http_proxy}
+      https_proxy: ${https_proxy}
+      HF_HUB_DISABLE_PROGRESS_BARS: 1
+      HF_HUB_ENABLE_HF_TRANSFER: 0
+      HABANA_VISIBLE_DEVICES: all
+      OMPI_MCA_btl_vader_single_copy_mechanism: none
+    runtime: habana
+    cap_add:
+      - SYS_NICE
+    ipc: host
+    command: --model-id ${LVM_MODEL_ID} --max-input-length 4096 --max-total-tokens 8192
+  lvm-tgi:
+    image: opea/lvm-tgi:latest
+    container_name: lvm-tgi-gaudi-server
+    depends_on:
+      - llava-tgi-service
+    ports:
+      - "9399:9399"
+    ipc: host
+    environment:
+      no_proxy: ${no_proxy}
+      http_proxy: ${http_proxy}
+      https_proxy: ${https_proxy}
+      LVM_ENDPOINT: ${LVM_ENDPOINT}
+      HF_HUB_DISABLE_PROGRESS_BARS: 1
+      HF_HUB_ENABLE_HF_TRANSFER: 0
+    restart: unless-stopped
+  visualqna-gaudi-backend-server:
+    image: opea/visualqna:latest
+    container_name: visualqna-gaudi-backend-server
+    depends_on:
+      - llava-tgi-service
+      - lvm-tgi
+    ports:
+      - "8888:8888"
+    environment:
+      - no_proxy=${no_proxy}
+      - https_proxy=${https_proxy}
+      - http_proxy=${http_proxy}
+      - MEGA_SERVICE_HOST_IP=${MEGA_SERVICE_HOST_IP}
+      - LVM_SERVICE_HOST_IP=${LVM_SERVICE_HOST_IP}
+    ipc: host
+    restart: always
+  visualqna-gaudi-ui-server:
+    image: opea/visualqna-ui:latest
+    container_name: visualqna-gaudi-ui-server
+    depends_on:
+      - visualqna-gaudi-backend-server
+    ports:
+      - "5173:5173"
+    environment:
+      - no_proxy=${no_proxy}
+      - https_proxy=${https_proxy}
+      - http_proxy=${http_proxy}
+      - CHAT_BASE_URL=${BACKEND_SERVICE_ENDPOINT}
+    ipc: host
+    restart: always
+
+networks:
+  default:
+    driver: bridge
diff --git a/VisualQnA/docker/gaudi/set_env.sh b/VisualQnA/docker/gaudi/set_env.sh
@@ -0,0 +1,11 @@
+#!/usr/bin/env bash
+
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+export LVM_MODEL_ID="llava-hf/llava-v1.6-mistral-7b-hf"
+export LVM_ENDPOINT="http://${host_ip}:8399"
+export LVM_SERVICE_PORT=9399
+export MEGA_SERVICE_HOST_IP=${host_ip}
+export LVM_SERVICE_HOST_IP=${host_ip}
+export BACKEND_SERVICE_ENDPOINT="http://${host_ip}:8888/v1/visualqna"