docker-compose.yaml

# Requires Docker Compose V2.
# See https://docs.docker.com/compose/compose-file/compose-file-v3
# and https://github.com/compose-spec/compose-spec/blob/master/spec.md
# for details concerning the `docker-compose.yaml` file syntax.

# Using `docker-compose.yaml` has many advantages over writing custom shell scripts.
# The settings are much easier to see and maintain than scattered shell scripts.
# Also, Compose is a native Docker component, simplifying project maintenance.

# Run `make env` to create a basic `.env` file with the UID and GID variables.
# Variables are in ${VARIABLE:-DEFAULT_VALUE} format to specify default values.
# Using the `.env` file to set variables to non-default values is strongly recommended.
# Note that the host shell has higher priority than `.env` for Docker Compose.
# https://docs.docker.com/compose/environment-variables/envvars-precedence

# Variables specified via Docker Compose have higher priority than those
# specified in the Dockerfile, which only function as default values.
# All default values in the Dockerfile are overridden by default values in Compose.

# Set the host environment variable `BUILDKIT_PROGRESS=plain` to see the full build log.
# https://github.com/docker/cli/blob/master/docs/reference/commandline/cli.md#environment-variables

networks: # Use the host network instead of creating a separate network.
  default: # This reduces load and conflicts with the host network.
    name: host # This may cause security issues in production, however.
    external: true

services:
  base: # Base service containing configurations common to all services.
    tty: true # Equivalent to `-t` flag in `docker run`.
    init: true # Equivalent to `--init` flag in `docker run`.
    stdin_open: true # equivalent to `-i` flag in `docker run`.
    working_dir: ${PROJECT_ROOT:-/opt/project}
    user: ${UID:-1000}:${GID:-1000} # Specify USR/GRP at runtime.
    # Use different image names for different users and projects.
    # Otherwise, images will be repeatedly removed and recreated.
    # The removed images will remain cached, however.
    image: ${IMAGE_NAME}
    network_mode: host # Use the same network as the host, may cause security issues.
    # `ipc: host` removes the shared memory cap but is a known security vulnerability.
    ipc: host # Equivalent to `--ipc=host` in `docker run`. **Disable this on WSL.**
    # shm_size: 1GB  # Explicit shared memory limit. No security issues this way.

    # Common environment variables for the container runtime. No effect on build.
    environment: # Equivalent to `--env`
      HISTSIZE: 50000 # Hard-coded large command history size.
      TZ: ${TZ:-UTC} # Timezone settings used during runtime.
    tmpfs:  # Create directory in RAM for fast data IO.
      - /opt/data
    # Default volume pairings of ${HOST_PATH}:${CONTAINER_PATH}.
    # Allows the container to access `HOST_PATH` as `CONTAINER_PATH`.
    # See https://docs.docker.com/storage/volumes for details.
    # Always use the ${HOME} variable to specify the host home directory.
    # See https://github.com/docker/compose/issues/6506 for details.
    volumes: # Equivalent to `-v` flag in `docker run`.
      # Current working directory `.` is connected to `PROJECT_ROOT`.
      # Mount `.` if the docker-compose.yaml file is at the project root.
      # Mount `..` if Cresset is a subdirectory in a different project, etc.
      - ${HOST_ROOT:-.}:${PROJECT_ROOT:-/opt/project}
      # Preserve VSCode extensions between containers.
      # Assumes default VSCode server directory.
      # May cause VSCode issues if multiple Cresset-based projects are on the
      # same machine writing to the `${HOME}/.vscode-server` directory.
      # If so, specify a different host directory for each project.
      - ${HOME}/.vscode-server:/home/${USR:-user}/.vscode-server

    build:
      context: . # Nearly all files are ignored due to `.dockerignore` settings.
      target: ${TARGET_STAGE:-train} # Specify the `Dockerfile` target build stage.
      args: # Common build-time environment variables.
        # Even if these variables are unnecessary during the build,
        # they can be ignored simply by not defining them in that stage.
        ADD_USER: ${ADD_USER:-include} # Whether to create a new sudo user in the image.
        PROJECT_ROOT: ${PROJECT_ROOT:-/opt/project}
        GID: ${GID:-1000}
        UID: ${UID:-1000}
        GRP: ${GRP:-user}
        USR: ${USR:-user}
        TZ: ${TZ:-UTC}
        TMUX_HIST_LIMIT: 50000
        # Change the `CONDA_URL` for different hardware architectures.
        # URLs from https://github.com/conda-forge/miniforge are recommended over
        # Miniconda URLs from https://docs.conda.io/en/latest/miniconda.html.
        # The `CONDA_MANAGER` may be either `mamba` (the default) or `conda`.
        # However, `mamba` may be unable to resolve conflicts that `conda` can.
        # In such cases, set `CONDA_MANAGER=conda` for conda-based installation.
        # Installing `mamba` via mini-forge is strongly recommended.
        CONDA_URL: ${CONDA_URL:-https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh}
        CONDA_MANAGER: ${CONDA_MANAGER:-mamba}

  gaudi:  # Gaudi-specific configurations.
    extends:
      service: base  # Inherit from the `base` service.
    runtime: habana
    cap_add: # Grant the container the CAP_SYS_NICE capability.
      - sys_nice
    environment:
      OMPI_MCA_btl_vader_single_copy_mechanism: none
      PT_HPU_ENABLE_LAZY_COLLECTIVES: true
      # Specify which HPU to use.
      HABANA_VISIBLE_DEVICES: ${HABANA_VISIBLE_DEVICES:-all}
      # Enable when using DeepSpeed.
     # PT_HPU_MAX_COMPOUND_OP_SIZE: 1024
     # PT_HPU_POOL_MEM_ACQUIRE_PERC: 100
    build:
      dockerfile: Dockerfile
      args:
        OS: ${OS:-ubuntu22.04}
        PYTORCH_VERSION: ${PYTORCH_VERSION:-2.3.1}
        SYNAPSE_VERSION: ${SYNAPSE_VERSION:-1.17.0}
        IMAGE_TAG: ${IMAGE_TAG:-latest}