compose.yaml

models:
  llama:
    model: ai/example-model
    context-size: 1024
    runtime-flags: "--no-prefill-assistant"

services:
# Option 1 : by host alias

  app:
    models:
      - llama  # container will run with `llama` alias in /etc/hosts

  another_app:
    models:
      # mimic https://github.com/compose-spec/compose-spec/blob/main/05-services.md#links to let user choose a network alias for Model
      - llama:model-runner 
      

# Option 2 : by env variable

  app_with_env:
    models:
      - llama # container will run with `LLAMA_URL` set

  another_app_with_env:
    models:
      llama: 
        variable: MODEL_URL