Train distilbert-base-uncased from Docker

This repository contains all the files needed to train a distilbert-base-uncased Model. To install the needed dependencies, you can use Docker.

Get files

Pull this repository to get all necessary Files.
git clone https://github.com/Reti97/mlops-docker destination-folder

Docker Setup

Get Container from Docker Hub

Get the dockerfile with the following command from the docker hub: docker pull reti97/new-python-app:latest

Run Training for 3 Epochs

Then you can train the Model with the best parameters found in Project 1: docker run -v Path/to/your/file/main.py:/code dockerimage python main.py --learning_rate 0.00011633938221625261 --adam_epsilon 6.73493036944769e-08 --warmup_steps 23

Build Dockerfile

You can also build the dockerfile yourself with:

docker build -t docker-name .

Training

The parameters are the following:

• learning_rate = 0.00011633938221625261
• adam_epsilon = 6.73493036944769e-08
• warmup_steps = 23

The rest of the parameters will be set with a default.