How to train when using docker

Hi, I’m new to rasa, I setup a docker container with rasa and duckling. The docker-compose.yml contains the following:

version: '3.0'
services:
  rasa:
    image: rasa/rasa:2.0.0rc2-full
    ports:
      - 5005:5005
    volumes:
      - ./:/app
    command:
      - run
      - --cors
      - "*"
      - --enable-api
  duckling:
    image: rasa/duckling
    ports:
      - "8000:8000"

I run docker-compose up when I want to run the servers and I run docker-compose run rasa train when I want to train a new model. However every time I run the training command tf has to be downloaded again.

How can I avoid doing this? Thanks a lot