This sample demonstrates how to automatically publish models to Firebase ML for each TensorFlow Lite file that is uploaded to Firebase Storage.
See file functions/index.js for the model publishing code.
The dependencies are listed in functions/package.json.
The function triggers on upload of any file to your Firebase project's default Cloud Storage bucket.
To deploy and test the sample:
- Create a Firebase project on the Firebase Console and visit the Storage tab.
- Clone this repo:
git clone https://github.com/firebase/functions-samples. - Open this sample's directory:
cd functions-samples/publish-model - Setup your project by running
firebase use --addand select the project you had created. - Install dependencies in the functions directory:
cd functions; npm install; cd - - Deploy your project using
firebase deploy - Go to your project's Cloud Console > IAM & admin > IAM, Find the App Engine default service account.
- Add the Service Account Token Creator role to that member. This will allow your app to create signed public URLs to the models.
- Add the Firebase ML Kit Admin role to that member. This will allow your app to manage Firebase ML models.
- Add the Firebase Admin SDK Administrator Service Agent role to that member. This will allow your app to manage models with the Firebase Admin SDK.
- Go to the Firebase Console Storage tab and upload a Tensorflow Lite model (*.tflite). After a short time a model with same file name will be published to Firebase ML.
- Go to the Firebase Console ML Kit (Custom) tab and see that a new model with file name has been created.