Tasks¶

A task allows you to run arbitrary commands on one or more nodes. They are best suited for jobs like training or batch processing.

Define a configuration¶

First, define a task configuration as a YAML file in your project folder. The filename must end with .dstack.yml (e.g. .dstack.yml or dev.dstack.yml are both acceptable).

type: task
# The name is optional, if not specified, generated randomly
name: axolotl-train

# Using the official Axolotl's Docker image
image: winglian/axolotl-cloud:main-20240429-py3.11-cu121-2.2.1

# Required environment variables
env:
  - HF_TOKEN
  - WANDB_API_KEY
# Commands of the task
commands:
  - accelerate launch -m axolotl.cli.train examples/fine-tuning/axolotl/config.yaml

resources:
  gpu:
    # 24GB or more vRAM
    memory: 24GB..
    # Two or more GPU
    count: 2..

Ports¶

A task can configure ports. In this case, if the task is running an application on a port, dstack apply will securely allow you to access this port from your local machine through port forwarding.

type: task
# The name is optional, if not specified, generated randomly
name: streamlit-hello

python: "3.10"

# Commands of the task
commands:
  - pip3 install streamlit
  - streamlit hello
# Expose the port to access the web app
ports: 
  - 8501

When running it, dstack apply forwards 8501 port to localhost:8501, enabling secure access to the running application.

Distributed tasks¶

By default, a task runs on a single node. However, you can run it on a cluster of nodes by specifying nodes.

type: task
# The name is optional, if not specified, generated randomly
name: train-distrib

# The size of the cluster
nodes: 2

python: "3.12"

# Commands to run on each node
commands:
  - git clone https://github.com/pytorch/examples.git
  - cd examples/distributed/ddp-tutorial-series
  - pip install -r requirements.txt
  - torchrun
    --nproc-per-node=$DSTACK_GPUS_PER_NODE
    --node-rank=$DSTACK_NODE_RANK
    --nnodes=$DSTACK_NODES_NUM
    --master-addr=$DSTACK_MASTER_NODE_IP
    --master-port=12345
    multinode.py 50 10

resources:
  gpu: 24GB
  # Uncomment if using multiple GPUs
  #shm_size: 24GB

Nodes can communicate using their private IP addresses. Use DSTACK_MASTER_NODE_IP, $DSTACK_NODE_RANK, and other System environment variables to discover IP addresses and other details.

Network interface

Distributed frameworks usually detect the correct network interface automatically, but sometimes you need to specify it explicitly.

For example, with PyTorch and the NCCL backend, you may need to add these commands to tell NCCL to use the private interface:

commands:
  - apt-get install -y iproute2
  - >
    if [[ $DSTACK_NODE_RANK == 0 ]]; then
      export NCCL_SOCKET_IFNAME=$(ip -4 -o addr show | fgrep $DSTACK_MASTER_NODE_IP | awk '{print $2}')
    else
      export NCCL_SOCKET_IFNAME=$(ip route get $DSTACK_MASTER_NODE_IP | sed -E 's/.*?dev (\S+) .*/\1/;t;d')
    fi
  # ... The rest of the commands

Fleets

Distributed tasks can only run on fleets with cluster placement. While dstack can provision such fleets automatically, it is recommended to create them via a fleet configuration to ensure the highest level of inter-node connectivity.

dstack is easy to use with accelerate, torchrun, Ray, Spark, and any other distributed frameworks.

Resources¶

When you specify a resource value like cpu or memory, you can either use an exact value (e.g. 24GB) or a range (e.g. 24GB.., or 24GB..80GB, or ..80GB).

type: task
# The name is optional, if not specified, generated randomly
name: train    

# Commands of the task
commands:
  - pip install -r fine-tuning/qlora/requirements.txt
  - python fine-tuning/qlora/train.py

resources:
  # 200GB or more RAM
  memory: 200GB..
  # 4 GPUs from 40GB to 80GB
  gpu: 40GB..80GB:4
  # Shared memory (required by multi-gpu)
  shm_size: 16GB
  # Disk size
  disk: 500GB

The gpu property allows specifying not only memory size but also GPU vendor, names and their quantity. Examples: nvidia (one NVIDIA GPU), A100 (one A100), A10G,A100 (either A10G or A100), A100:80GB (one A100 of 80GB), A100:2 (two A100), 24GB..40GB:2 (two GPUs between 24GB and 40GB), A100:40GB:2 (two A100 GPUs of 40GB).

Google Cloud TPU

To use TPUs, specify its architecture via the gpu property.

type: task
# The name is optional, if not specified, generated randomly
name: train    

python: "3.10"

# Commands of the task
commands:
  - pip install -r fine-tuning/qlora/requirements.txt
  - python fine-tuning/qlora/train.py

resources:
  gpu: v2-8

Currently, only 8 TPU cores can be specified, supporting single TPU device workloads. Multi-TPU support is coming soon.

Shared memory

If you are using parallel communicating processes (e.g., dataloaders in PyTorch), you may need to configure shm_size, e.g. set it to 16GB.

Python version¶

If you don't specify image, dstack uses its base Docker image pre-configured with python, pip, conda (Miniforge), and essential CUDA drivers. The python property determines which default Docker image is used.

type: task
# The name is optional, if not specified, generated randomly
name: train    

# If `image` is not specified, dstack uses its base image
python: "3.10"

# Commands of the task
commands:
  - pip install -r fine-tuning/qlora/requirements.txt
  - python fine-tuning/qlora/train.py

nvcc

By default, the base Docker image doesn’t include nvcc, which is required for building custom CUDA kernels. If you need nvcc, set the corresponding property to true.

type: task
# The name is optional, if not specified, generated randomly
name: train    

# If `image` is not specified, dstack uses its base image
python: "3.10"
# Ensure nvcc is installed (req. for Flash Attention) 
nvcc: true

commands:
  - pip install -r fine-tuning/qlora/requirements.txt
  - python fine-tuning/qlora/train.py

Docker¶

If you want, you can specify your own Docker image via image.

type: task
# The name is optional, if not specified, generated randomly
name: train    

# Any custom Docker image
image: dstackai/base:py3.13-0.6-cuda-12.1

# Commands of the task
commands:
  - pip install -r fine-tuning/qlora/requirements.txt
  - python fine-tuning/qlora/train.py

Private registry

Use the registry_auth property to provide credentials for a private Docker registry.

type: dev-environment
# The name is optional, if not specified, generated randomly
name: train

# Any private Docker image
image: dstackai/base:py3.13-0.6-cuda-12.1
# Credentials of the private Docker registry
registry_auth:
  username: peterschmidt85
  password: ghp_e49HcZ9oYwBzUbcSk2080gXZOU2hiT9AeSR5

# Commands of the task
commands:
  - pip install -r fine-tuning/qlora/requirements.txt
  - python fine-tuning/qlora/train.py

Privileged mode

All backends except runpod, vastai, and kubernetes support running containers in privileged mode. This mode enables features like using Docker and Docker Compose inside dstack runs.

Environment variables¶

type: task
# The name is optional, if not specified, generated randomly
name: train

python: "3.10"

# Environment variables
env:
  - HF_TOKEN
  - HF_HUB_ENABLE_HF_TRANSFER=1

# Commands of the task
commands:
  - pip install -r fine-tuning/qlora/requirements.txt
  - python fine-tuning/qlora/train.py

If you don't assign a value to an environment variable (see HF_TOKEN above), dstack will require the value to be passed via the CLI or set in the current process.

System environment variables

The following environment variables are available in any run by default:

Name	Description
`DSTACK_RUN_NAME`	The name of the run
`DSTACK_REPO_ID`	The ID of the repo
`DSTACK_GPUS_NUM`	The total number of GPUs in the run
`DSTACK_NODES_NUM`	The number of nodes in the run
`DSTACK_GPUS_PER_NODE`	The number of GPUs per node
`DSTACK_NODE_RANK`	The rank of the node
`DSTACK_MASTER_NODE_IP`	The internal IP address of the master node
`DSTACK_NODES_IPS`	The list of internal IP addresses of all nodes delimited by "\n"

Spot policy¶

By default, dstack uses on-demand instances. However, you can change that via the spot_policy property. It accepts spot, on-demand, and auto.

Reference

Tasks support many more configuration options, incl. backends, regions, max_price, and max_duration, among others.

Run a configuration¶

To run a task, pass the configuration to dstack apply:

$ HF_TOKEN=...
$ WANDB_API_KEY=...
$ dstack apply -f examples/.dstack.yml

 #  BACKEND  REGION    RESOURCES                    SPOT  PRICE
 1  runpod   CA-MTL-1  18xCPU, 100GB, A5000:24GB:2  yes   $0.22
 2  runpod   EU-SE-1   18xCPU, 100GB, A5000:24GB:2  yes   $0.22
 3  gcp      us-west4  27xCPU, 150GB, A5000:24GB:3  yes   $0.33

Submit the run axolotl-train? [y/n]: y

Launching `axolotl-train`...
---> 100%

{'loss': 1.4967, 'grad_norm': 1.2734375, 'learning_rate': 1.0000000000000002e-06, 'epoch': 0.0}
  0% 1/24680 [00:13<95:34:17, 13.94s/it]
  6% 73/1300 [00:48<13:57,  1.47it/s]

dstack apply automatically provisions instances, uploads the contents of the repo (incl. your local uncommitted changes), and runs the commands.

Retry policy¶

By default, if dstack can't find capacity, the task exits with an error, or the instance is interrupted, the run will fail.

If you'd like dstack to automatically retry, configure the retry property accordingly:

type: task
# The name is optional, if not specified, generated randomly
name: train    

python: "3.10"

# Commands of the task
commands:
  - pip install -r fine-tuning/qlora/requirements.txt
  - python fine-tuning/qlora/train.py

retry:
  # Retry on specific events
  on_events: [no-capacity, error, interruption]
  # Retry for up to 1 hour
  duration: 1h

Creation policy¶

By default, when you run dstack apply with a dev environment, task, or service, if no idle instances from the available fleets meet the requirements, dstack creates a new fleet using configured backends.

To ensure dstack apply doesn't create a new fleet but reuses an existing one, pass -R (or --reuse) to dstack apply.

$ dstack apply -R -f examples/.dstack.yml

Or, set creation_policy to reuse in the run configuration.

Idle duration¶

If a fleet is created automatically, it stays idle for 5 minutes by default and can be reused within that time. If the fleet is not reused within this period, it is automatically terminated. To change the default idle duration, set idle_duration in the run configuration (e.g., 0s, 1m, or off for unlimited).

Fleets

For greater control over fleet provisioning, it is recommended to create fleets explicitly.

Manage runs¶

List runs¶

The dstack ps command lists all running jobs and their statuses. Use --watch (or -w) to monitor the live status of runs.

Stop a run¶

A dev environment runs until you stop it or its lifetime exceeds max_duration. To gracefully stop a dev environment, use dstack stop. Pass --abort or -x to stop without waiting for a graceful shutdown.

Attach to a run¶

By default, dstack apply runs in attached mode – it establishes the SSH tunnel to the run, forwards ports, and shows real-time logs. If you detached from a run, you can reattach to it using dstack attach.

See run logs¶

To see the logs of a run without attaching, use dstack logs. Pass --diagnose/-d to dstack logs to see the diagnostics logs. It may be useful if a run fails. For more information on debugging failed runs, see the troubleshooting guide.

What's next?

Read about dev environments, services, and repos
Learn how to manage fleets
Check the Axolotl example