Merge pull request #285 from univerone/add-docker-build

[MRG]:whale: Update modelci docker image （CPU&GPU). docker-compose is done

Author: HuaizhengZhang

.dockerignore

@@ -5,7 +5,6 @@ build
 dist
 *.egg-info
 *.egg/
-*.pyc
 *.swp
 
 .tox
@@ -15,7 +14,8 @@ html/*
 __pycache__
 
 ### Front end ###
-frontend
+node_modules
+npm-debug.log
 
 ### Build cache ###
 */**/*.cache

.github/workflows/run_test.yml

@@ -11,7 +11,7 @@ on:
     - 'frontend/**'
 jobs:
   test:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-18.04
     services:
       mongodb:
         image: mongo

README.md

@@ -47,79 +47,151 @@ Several features are in beta testing and will be available in the next release s
 
 *If your want to join in our development team, please contact huaizhen001 @ e.ntu.edu.sg*
 
-## Installation
+## Demo
 
-### using pip
+The below figures illusrates the web interface of our system and overall workflow.
+| Web frontend |   Workflow     |
+|:------------:|:--------------:|
+| <img src="https://i.loli.net/2020/12/10/4FsfciXjtPO12BQ.png" alt="drawing" width="500"/> | <img src="https://i.loli.net/2020/12/10/8IaeW9mS2NjQEYB.png" alt="drawing" width="500"/>    |
+
+
+## Installation Guide
+
+### Prerequisites
+
+- A GNU/Linux environment(Ubuntu preferred)
+- [Docker](https://docs.docker.com/engine/install/)
+- [Docker Compose](https://docs.docker.com/compose/) (Optional, for Installation via Docker)
+- [TVM](https://github.com/apache/incubator-tvm) and `tvm` Python module(Optional)
+- [TensorRT](https://docs.nvidia.com/deeplearning/tensorrt/install-guide/index.html) and Python API(Optional)
+- Python >= 3.7
+
+### Installation using  pip
 
 ```shell script
-# upgrade installation packages first
-pip install -U setuptools requests
 # install modelci from GitHub
 pip install git+https://github.com/cap-ntu/ML-Model-CI.git@master
 ```
 
-### create conda workspace 
+Once you have installed, make sure the docker daemon is running, then you can start MLModelCI service on a leader server by:
+
+```bash
+modelci service init
+```
 
-**Note**
-- Conda and Docker are required to run this installation script.
-- To use TensorRT, you have to manually install TensorRT (`sudo` is required). See instruction 
-[here](https://docs.nvidia.com/deeplearning/tensorrt/install-guide/index.html). 
+![CLI start service](https://i.loli.net/2021/04/15/rLiMoxkqRO67Tyg.gif)
 
-```shell script
-bash scripts/install.sh
+
+
+Or stop the service by:
+
+```bash
+modelci service stop
 ```
 
-### Docker
+![CLI stop service](https://i.loli.net/2021/04/16/jo1ZnWsqrmxFvlU.gif)
+
+
+
+### Installation using Docker
 
-![](https://img.shields.io/docker/pulls/mlmodelci/mlmodelci.svg) ![](https://img.shields.io/docker/image-size/mlmodelci/mlmodelci)
+![](https://img.shields.io/docker/pulls/mlmodelci/mlmodelci.svg) 
+
+#### For CPU-only Machines
+
+![](https://img.shields.io/docker/v/mlmodelci/mlmodelci/cpu)![](https://img.shields.io/docker/image-size/mlmodelci/mlmodelci/cpu)
 
 ```shell script
-docker pull mlmodelci/mlmodelci
+docker pull mlmodelci/mlmodelci:cpu
 ```
 
-<!-- Please refer to [here](/integration/README.md) for more information. -->
+Start basic services by Docker Compose:
 
+```bash
+docker-compose -f ML-Model-CI/docker/docker-compose-cpu-modelhub.yml up -d
+```
 
-## Quick Start
+Stop the services by:
 
-The below figurs illusrates the 
-| Web frontend |   Workflow     |
-|:------------:|:--------------:|
-| <img src="https://i.loli.net/2020/12/10/4FsfciXjtPO12BQ.png" alt="drawing" width="500"/> | <img src="https://i.loli.net/2020/12/10/8IaeW9mS2NjQEYB.png" alt="drawing" width="500"/>    |
+```bash
+docker-compose -f ML-Model-CI/docker/docker-compose-cpu-modelhub.yml down
+```
+
+#### For CUDA10.2 Machine
 
-### 0. Start the ModelCI service
+![](https://img.shields.io/docker/v/mlmodelci/mlmodelci/cuda10.2-cudnn8)![](https://img.shields.io/docker/image-size/mlmodelci/mlmodelci/cuda10.2-cudnn8)
 
-Once you have installed, start MLModelCI service on a leader server by:
 ```shell script
-modelci service init
+docker pull mlmodelci/mlmodelci:cuda10.2-cudnn8
+```
+
+Start basic services by Docker Compose:
+
+```bash
+docker-compose -f ML-Model-CI/docker/docker-compose-gpu-modelhub.yml up -d
+```
+
+![docker-compose start service](https://i.loli.net/2021/04/15/65oYIBurfhPRK3U.gif)
+
+Stop the services by:
+
+```bash
+docker-compose -f ML-Model-CI/docker/docker-compose-gpu-modelhub.yml down
 ```
 
-**We provide three options for users to use MLModelCI: CLI, Web interface and import it as a python package**
+![docker-compose stop service](https://i.loli.net/2021/04/15/CyNzo4uhXkSrQRE.gif)
+
+<!-- Please refer to [here](/integration/README.md) for more information. -->
+
+
+## Usage
+
+**We provide three options for users to use MLModelCI: CLI, Running Programmatically and Web interface**
+
 ### 1. CLI
 
-```python
+```console
 # publish a model to the system
-modelci modelhub publish registration_example.yml
+modelci@modelci-PC:~$ modelci modelhub publish -f example/resnet50.yml
+{'data': {'id': ['60746e4bc3d5598e0e7a786d']}, 'status': True}
 ```
 
-Please refer to [WIKI](https://github.com/cap-ntu/ML-Model-CI/wiki) for more CLIs.
+Please refer to [WIKI](https://github.com/cap-ntu/ML-Model-CI/wiki) for more CLI options.
 
-### 2. Python Package
+### 2. Running Programmatically
 
 ```python
 # utilize the convert function
-from modelci.hub.converter import ONNXConverter
+from modelci.hub.converter import convert
 from modelci.types.bo import IOShape
 
 # the system can trigger the function automaticlly
 # users can call the function individually 
-ONNXConverter.from_torch_module(
-    '<path to torch model>', 
-    '<path to export onnx model>', 
-    inputs=[IOShape([-1, 3, 224, 224], float)],
-)
+convert(
+    '<torch model>',
+    src_framework='pytorch', 
+    dst_framework='onnx',
+    save_path='<path to export onnx model>', 
+    inputs=[IOShape([-1, 3, 224, 224], dtype=float)], 
+    outputs=[IOShape([-1, 1000], dtype=float)], 
+    opset=11)
 ```
 
+### 3. Web Interface
+
+If you have installed MLModelCI via pip, you should start the frontend service manually.
+
+```bash
+# Navigate to the frontend folder
+cd frontend
+# Install dependencies
+yarn install
+# Start the frontend
+yarn start
+```
+
+The frontend will start on <http://localhost:3333>
+
 ## Quickstart with Notebook
 
 - [Publish an image classification model](./example/notebook/image_classification_model_deployment.ipynb)  [![nbviewer](https://raw.githubusercontent.com/jupyter/design/master/logos/Badges/nbviewer_badge.svg)](https://nbviewer.jupyter.org/github/cap-ntu/ML-Model-CI/blob/master/example/notebook/image_classification_model_deployment.ipynb)