tensorflow-lambda

Example of running TensorFlow in Lambda

Requirements

• asdf
• curl
• docker

Setup

Clone the repository.

git clone https://github.com/RyoWakabayashi/tensorflow-lambda.git
cd tensorflow-lambda

Install asdf plugins.

asdf plugin-add aws-sam-cli \
  ; asdf plugin-add direnv \
  ; asdf plugin-add jq \
  ; asdf plugin-add nodejs \
  ; asdf plugin-add python \
  ; asdf plugin-add yarn

Install Languages and Tools.

asdf install

Download models

Download MobileNet v2 model files.

python download_models.py

Build

Move to the sam directory.

cd sam

Build sam.

Make sure Docker is launched.

sam build

Run locally

$ sam local start-api
...
2022-08-01 15:12:33  * Running on http://127.0.0.1:3000/ (Press CTRL+C to quit)

Calling the API from another terminal.

curl -XPOST \
  http://127.0.0.1:3000/predictions \
  -H "Content-Type: image/jpeg" \
  --data-binary @imgs/sample.jpg | jq

Deploy

Deploy with a guide the first time.

API URL is displayed after execution.

$ sam deploy --guided
...
-------------------------------------------------------------------------------------------------------
Outputs
-------------------------------------------------------------------------------------------------------
Key                 Api
Description         API Gateway endpoint URL for v1 stage for Tensorflow sample function
Value               https://some.execute-api.ap-northeast-1.amazonaws.com/v1/predictions
-------------------------------------------------------------------------------------------------------

If you save the configuration information, you do not need to use the guide for the second time or later.

sam deploy

Test

Change the URL to the value displayed after deployment.

curl -XPOST \
  https://some.execute-api.ap-northeast-1.amazonaws.com/v1/predictions \
  -H "Content-Type: image/jpeg" \
  --data-binary @imgs/sample.jpg | jq

You may view the log.

sam logs --stack-name some-stack-name

Preparing to edit this repository

Install commitlint.

yarn

Install pre-commit.

pip install --requirement requirements.txt
asdf reshim python
pre-commit install

Run pre-commit manually.

pre-commit run --all-files

Notes

No authentication is applied in this sample.

Please add authentication to the production.