ray_cluster.yaml

cluster_name: ray_cluster_gpu_aws_llm_finetuning

max_workers: 8

upscaling_speed: 1.0

docker:
    image: "rayproject/ray-ml:latest-gpu"
    container_name: "ray_container"
    pull_before_run: False
    run_options:   # Extra options to pass into "docker run"
        - --ulimit nofile=65536:65536

    head_image: "rayproject/ray-ml:latest-gpu"
    worker_image: "rayproject/ray-ml:latest-gpu"

idle_timeout_minutes: 5

provider:
    type: aws
    region: eu-west-2
    cache_stopped_nodes: True

auth:
    ssh_user: ubuntu

available_node_types:
    head:
        node_config:
            InstanceType: g4dn.xlarge
            ImageId: ami-0bacb0baeaa1dbf0b
            BlockDeviceMappings:
                - DeviceName: /dev/sda1
                  Ebs:
                      VolumeSize: 300
                      VolumeType: gp3
    worker:
        min_workers: 2
        max_workers: 8
        node_config:
            InstanceType: g4dn.xlarge
            ImageId: ami-0bacb0baeaa1dbf0b
            # Run workers on spot by default. Comment this out to use on-demand.
            InstanceMarketOptions:
                MarketType: spot
                SpotOptions:
                    MaxPrice: "1"  # Max Hourly Price
            BlockDeviceMappings:
                - DeviceName: /dev/sda1
                  Ebs:
                      VolumeSize: 300
                      VolumeType: gp3


file_mounts: {
    "~/src": "./src",
    "~/data": "./data",
    "~/infra": "./infra",
    "~/config": "./config",
}

file_mounts_sync_continuously: False

cluster_synced_files: []

rsync_exclude:
    - "**/.git"
    - "**/.git/**"

rsync_filter:
    - ".gitignore"

head_node_type: head

head_start_ray_commands:
    - ulimit -n 65536
    - conda activate py310 && ray stop
    - conda activate py310 && ray start --head --port=6379 --object-manager-port=8076 --autoscaling-config=~/ray_bootstrap_config.yaml --dashboard-host=0.0.0.0

worker_start_ray_commands:
    - ulimit -n 65536
    - conda activate py310 && ray stop
    - conda activate py310 && ray start --address=$RAY_HEAD_IP:6379 --object-manager-port=8076

initialization_commands: []

head_setup_commands:
    - conda activate py310 && pip install -U pip
    - conda activate py310 && pip install -r infra/requirements.txt

worker_setup_commands:
    - conda activate py310 && pip install -U pip
    - conda activate py310 && pip install -r infra/requirements.txt

setup_commands:
    - conda create -n py310 python=3.10 -y && conda activate py310