food101-model-deployment

Deploy effnet b2 model that was fine-tuned on food 101 on gradio

Adding a large file

In order to track large files it advised to install git lfs. The link below shows how to track a large file and commit it. These steps have been followed to add the PyTorch model artifact.

https://docs.github.com/en/repositories/working-with-files/managing-large-files/configuring-git-large-file-storage

install dependencies

method 1 - using a requirements.txt

When starting an ubuntu EC2 image one first has to perform some housekeeping:

sudo apt update -y

followed by:

supo apt upgrade -y

after this your machine should be up to date.

Now we can install pip3 that is required for the installation of python packages

sudo apt install python3-pip -y

The last thing to install is git for large files (not sure if this is needed)

sudo apt install git-lfs

Now we can install the necessary packages. First install install PyTorch for the cpu:

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

Then install gradio

pip3 install gradio

now clone this repository on your local machine or VM:

git clone https://github.com/longtongster/food101-model-deployment.git

When you use a cloud provider (e.g. AWS EC2) make sure that port 8080 is open to the world (or your IP).

Then cd into the repo and and enter:

python3 app.py

Now the app should be up and running at <EC2 public IP>:8080

method 2 - using a bootstrap script on AWS EC2

Using a bootstrap script we can automate all the above manual steps.

#!/bin/bash
sudo apt update -y
supo apt upgrade -y
sudo apt install python3-pip -y
sudo apt install git-lfs -y
pip3 install gradio
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
git clone https://github.com/longtongster/food101-model-deployment.git
cd food101-model-deployment
python3 app.py

The logs of the userdata installation can be checked in the following files: /var/log/cloud-init.log and /var/log/cloud-init-output.log

The gradio app is running on port 8080. Please make sure that it is open. Open a webbrowser and type

<EC2 public IP>:8080

method 3 - containerize the app

First start an AWS EC2 instance with a slightly shorter bootstrap script

#!/bin/bash
sudo apt update -y
supo apt upgrade -y
sudo apt install python3-pip -y
sudo apt install git-lfs -y
sudo apt install git-lfs -y
sudo apt-get install docker-compose -y
sudo usermod -aG docker ubuntu

Once the image is up and running docker and git will be available. Clone the repo

git clone https://github.com/longtongster/food101-model-deployment.git

create a dockerfile

vim Dockerfile

and insert the following in the file

FROM python
WORKDIR /usr/src/app
RUN pip install gradio
RUN pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
COPY ./food101-model-deployment ./
CMD [ "python", "./app.py" ]

build the image

sudo docker image build -t model-app .

start a container from this image

sudo docker container run -d -p 80:8080 model-app

Port 80 of the host is mapped to 8080 on the container. Open a webbrowser and enter the ip ardress of the instance

<EC2 public IP>