Projector test

Source: https://prjctr.com/course/machine-learning-in-production

Kaggle: https://www.kaggle.com/competitions/commonlitreadabilityprize/data

How to start

# train the model with CPU
docker-compose run train

# train the model with CUDA
docker-compose run train_cuda

# start a server
docker-compose up -d server

# run the test terminal to communicate with the server
docker-compose run client

Configuring the train model

Checkout the config.yaml for the Hugging Face model configuration. The current model prajjwal1/bert-tiny is running fast on CPU.

API definition

1. Getting a prediction based on a text

POST /predict
<excerpts>

2. Get metrics

GET /metrics