siloka-ml-gateway

Siloka-ml-gateway is a consumable REST API service for the siloka-backend and collaboratively developed by Bangkit Academy 2022 Cohort.

Contributor to this repostory:

• Dennis Al Baihaqi Walangadi
• Gita Permatasari Sujatmiko
• Radhiansya Zain Antriksa Putra

Tech Stacks

This project was built on top of:

• FastAPI
• Google Compute Engine
• Cloud Storage
• Cloud Build
• Container Registry
• Cloud Logging

Development

1. Install Poetry
2. Install dependencies

   poetry install
   

3. Create .env file and fill it with

   GOOGLE_APPLICATION_CREDENTIALS='local/path/to/service_account_key.json'
   

Running the Service without Container

$ poetry run uvicorn app.main:app --reload

Or

$ poetry shell
$ uvicorn app.main:app --reload

Building the Container

$ docker build -t siloka-ml-gateway .

Running the Container

$ docker run -d -p 80:80 siloka-ml-gateway

Load the ML model

1. Open main.py and checks

   GOOGLE_APP_CREDENTIALS = os.environ.get("GOOGLE_APPLICATION_CREDENTIALS")
   ML_STORAGE_BUCKET = "siloka-ml-resources"
   

2. Change ML_STORAGE_BUCKET value to match your bucket name where you stored your model.

3. Store your model, tokenizer, and intents in Google Cloud Storage Bucket.

4. Make sure your model, tokenizer, and intents are named with chatbot-model.h5, tokenizer.pickle, and intents.json. or modify this snippet in main.py according to your needs

   download_blob_from_bucket(
       ML_STORAGE_BUCKET, "chatbot-model.h5", "model/chatbot-model.h5"
   )
   download_blob_from_bucket(
       ML_STORAGE_BUCKET, "tokenizer.pickle", "model/tokenizer.pickle"
   )
   download_blob_from_bucket(ML_STORAGE_BUCKET, "intents.json", "model/intents.json")
   
   model = load_model("./model/chatbot-model.h5")
   tokenizer = load_tokenizer("./model/tokenizer.pickle")
   intents = load_file("./model/intents.json")