>>> hf jobs run --name hello-world python:3.12 python -c "print('Hello world')"
>>> hf jobs run --name gpu-check --flavor a10g-small pytorch/pytorch:2.6.0-cuda12.4-cudnn9-devel python -c "import torch; print(torch.cuda.get_device_name())"

>>> hf jobs uv run --name my-script my_script.py
>>> hf jobs uv run --name sft-training --flavor a10g-small "https://raw.githubusercontent.com/huggingface/trl/main/trl/scripts/sft.py"

# Directly run Python code
>>> from huggingface_hub import run_job
>>> run_job(
...     image="python:3.12",
...     command=["python", "-c", "print('Hello from the cloud!')"],
... )

# Use GPUs without any setup
>>> run_job(
...     image="pytorch/pytorch:2.6.0-cuda12.4-cudnn9-devel",
...     command=["python", "-c", "import torch; print(torch.cuda.get_device_name())"],
...     flavor="a10g-small",
... )

# Run in an organization account
>>> run_job(
...     image="python:3.12",
...     command=["python", "-c", "print('Running in an org account')"],
...     namespace="my-org-name",
... )

# Run from Hugging Face Spaces
>>> run_job(
...     image="hf.co/spaces/lhoestq/duckdb",
...     command=["duckdb", "-c", "select 'hello world'"],
... )

# Run a Python script with `uv` (experimental)
>>> from huggingface_hub import run_uv_job
>>> run_uv_job("my_script.py")

>>> from huggingface_hub import run_job
>>> job = run_job(
...     image="python:3.12",
...     command=["python", "-c", "print('Hello from the cloud!')"]
... )
>>> job.url
https://huggingface.co/jobs/lhoestq/687f911eaea852de79c4a50a
>>> job.id
687f911eaea852de79c4a50a

>>> from huggingface_hub import rerun_job, run_job
>>> job = run_job(image="python:3.12", command=["python", "train.py"], attempts=3)
>>> job.retry
2

# Start a new Job with the same spec, including its retry setting
>>> new_job = rerun_job(job_id=job.id)

# List your jobs (results are paginated and returned as an iterator)
>>> from huggingface_hub import list_jobs
>>> jobs = list_jobs()
>>> next(iter(jobs))
JobInfo(id='687f911eaea852de79c4a50a', created_at=datetime.datetime(2025, 7, 22, 13, 24, 46, 909000, tzinfo=datetime.timezone.utc), docker_image='python:3.12', space_id=None, command=['python', '-c', "print('Hello from the cloud!')"], arguments=[], environment={}, secrets={}, flavor='cpu-basic', status=JobStatus(stage='COMPLETED', message=None), owner=JobOwner(id='5e9ecfc04957053f60648a3e', name='lhoestq'), endpoint='https://huggingface.co', url='https://huggingface.co/jobs/lhoestq/687f911eaea852de79c4a50a')

# Materialize the iterator into a list
>>> all_jobs = [job for job in list_jobs()]

# List your running jobs
>>> running_jobs = list_jobs(status="RUNNING")

# Filter by one or more statuses and/or labels
>>> list_jobs(status=["RUNNING", "SCHEDULING"], labels={"env": "prod"})

# Inspect the status of a job
>>> from huggingface_hub import inspect_job
>>> inspect_job(job_id=job_id)
JobInfo(id='687f911eaea852de79c4a50a', created_at=datetime.datetime(2025, 7, 22, 13, 24, 46, 909000, tzinfo=datetime.timezone.utc), docker_image='python:3.12', space_id=None, command=['python', '-c', "print('Hello from the cloud!')"], arguments=[], environment={}, secrets={}, flavor='cpu-basic', status=JobStatus(stage='COMPLETED', message=None), owner=JobOwner(id='5e9ecfc04957053f60648a3e', name='lhoestq'), endpoint='https://huggingface.co', url='https://huggingface.co/jobs/lhoestq/687f911eaea852de79c4a50a')

# View logs from a job
>>> from huggingface_hub import fetch_job_logs
>>> for log in fetch_job_logs(job_id=job_id):
...     print(log)
Hello from the cloud!

# View resources usage metrics from a job
>>> from huggingface_hub import fetch_job_metrics
>>> for metrics in fetch_job_metrics(job_id=job_id):
...     print(metrics)
{
    "cpu_usage_pct": 0,
    "cpu_millicores": 2000,
    "memory_used_bytes": 929792,
    "memory_total_bytes": 17179869184,
    "rx_bps": 0,
    "tx_bps": 0,
    "gpus": {},
    "replica": "4dzsh"
}

# Cancel a job
>>> from huggingface_hub import cancel_job
>>> cancel_job(job_id=job_id)

>>> from huggingface_hub import run_job, wait_for_job
>>> job = run_job(image="python:3.12", command=["python", "-c", "print('hello')"])
>>> wait_for_job(job_id=job.id).status.stage
'COMPLETED'

# Run multiple jobs in parallel and wait for all of them to finish
>>> jobs = [run_job(image=image, command=command) for command in commands]
>>> finished_jobs = wait_for_job(job_id=[job.id for job in jobs], timeout=3600)

# Chain on success
hf jobs wait <job_id> && hf jobs run --detach python:3.12 python eval.py

# Wait for all currently running jobs
hf jobs ls -q | xargs hf jobs wait

# Use an A10G GPU to check PyTorch CUDA
>>> from huggingface_hub import run_job
>>> run_job(
...     image="pytorch/pytorch:2.6.0-cuda12.4-cudnn9-devel",
...     command=["python", "-c", "import torch; print(f'This code ran with the following GPU: {torch.cuda.get_device_name()}')"],
...     flavor="a10g-small",
... )

This code ran with the following GPU: NVIDIA A10G

>>> from huggingface_hub import run_uv_job
>>> run_uv_job(
...     "sft.py",
...     script_args=["--model_name_or_path", "Qwen/Qwen2-0.5B", ...],
...     dependencies=["trl"],
...     env={"HF_TOKEN": ...},
...     flavor="a10g-small",
... )

>>> hf jobs hardware

>>> from huggingface_hub import list_jobs_hardware
>>> list_jobs_hardware()

>>> from huggingface_hub import run_job, Volume
>>> job = run_job(
...     image="duckdb/duckdb",
...     command=["duckdb", "-c", "SELECT * FROM '/data/**/*.parquet' LIMIT 5"],
...     volumes=[Volume(type="dataset", source="HuggingFaceFW/fineweb", mount_path="/data")],
... )

>>> from huggingface_hub import run_uv_job, Volume
>>> script = "my_sft.py"
>>> script_args = ["--output_dir", "/training-outputs/training-v3-final", ...]
>>> checkpoints_bucket = Volume(type="bucket", source="username/my-bucket", mount_path="/training-outputs")
>>> run_uv_job(script, script_args=script_args, volumes=[checkpoints_bucket])

>>> from huggingface_hub import run_uv_job, sync_job_volume

# Upload ./training-data once...
>>> volume = sync_job_volume("./training-data", "/data")

# ...then run as many Jobs as you want against it
>>> run_uv_job("train.py", script_args=["--learning-rate", "0.01"], volumes=[volume])
>>> run_uv_job("train.py", script_args=["--learning-rate", "0.05"], volumes=[volume])

>>> from huggingface_hub import run_uv_job, sync_bucket, sync_job_volume

>>> outputs = sync_job_volume("./outputs", "/outputs", read_only=False)
>>> job = run_uv_job("process.py", volumes=[outputs])

# ...once the Job completes, pull back the data:
>>> sync_bucket(f"hf://buckets/{outputs.source}/{outputs.path}", "./outputs")

>>> hf jobs uv run -v ./pdfs:/input -v ./md-out:/output:rw ocr.py

>>> from huggingface_hub import run_job, update_job_expose
>>> job = run_job(
...     image="python:3.12",
...     command=["python", "-m", "http.server", "8000"],
...     expose_public=[8000],
... )
>>> job.expose_public
[8000]

# Keep the port exposed, but require an HF token from now on
>>> job = update_job_expose(job_id=job.id, expose=[8000])

# Close all exposed ports
>>> job = update_job_expose(job_id=job.id)

>>> from huggingface_hub import run_job
>>> job = run_job(
...     image="python:3.12",
...     command=["sleep", "infinity"],
...     ssh=True,
... )
>>> job.status.ssh_url
'ssh://68498e23210b3a4f4e6e2a23@ssh.hf.jobs'

>>> hf jobs ssh 68498e23210b3a4f4e6e2a23

>>> from huggingface_hub import run_job
>>> server = run_job(
...     image="python:3.12",
...     command=["python", "-m", "http.server", "8000"],
...     network_group="train",
...     network_aliases=["master"],
... )
>>> client = run_job(
...     image="python:3.12",
...     command=["sh", "-c", 'curl --retry 10 --retry-connrefused "http://${HF_NETWORK_GROUP_PREFIX}master:8000/"'],
...     network_group="train",
... )
>>> server.network
JobNetwork(group='train', aliases=['master'])

>>> from huggingface_hub import run_job
>>> job = run_job(
...     image="pytorch/pytorch:2.6.0-cuda12.4-cudnn9-devel",
...     command=["python", "train_model.py"],
...     flavor="a10g-large",
...     timeout=7200,  # 2 hours in seconds
... )

>>> # Using different time units
>>> job = run_job(
...     image="pytorch/pytorch:2.6.0-cuda12.4-cudnn9-devel",
...     command=["python", "train_model.py"],
...     flavor="a10g-large",
...     timeout="2h",  # 2 hours
... )

>>> # Other examples:
>>> # timeout="30m"    # 30 minutes
>>> # timeout="1.5h"   # 1.5 hours
>>> # timeout="1d"     # 1 day
>>> # timeout="3600s"  # 3600 seconds

>>> from huggingface_hub import run_uv_job
>>> job = run_uv_job(
...     "training_script.py",
...     flavor="a10g-large",
...     timeout="90m",  # 90 minutes
... )

>>> from huggingface_hub import inspect_job, fetch_job_logs
>>> # Check job status
>>> job_info = inspect_job(job_id=job.id)
>>> if job_info.status.stage == "ERROR":
...     print(f"Job failed: {job_info.status.message}")
...     # Check logs for more details
...     for log in fetch_job_logs(job_id=job.id):
...         print(log)

# Pass environment variables
>>> from huggingface_hub import run_job
>>> run_job(
...     image="python:3.12",
...     command=["python", "-c", "import os; print(os.environ['FOO'], os.environ['BAR'])"],
...     env={"FOO": "foo", "BAR": "bar"},
... )

# Pass secrets - they will be encrypted server side
>>> from huggingface_hub import run_job
>>> run_job(
...     image="python:3.12",
...     command=["python", "-c", "import os; print(os.environ['MY_SECRET'])"],
...     secrets={"MY_SECRET": "psswrd"},
... )

# Access job environment information
>>> from huggingface_hub import run_job
>>> run_job(
...     image="python:3.12",
...     command=["python", "-c", """
... import os
... print(f"Job ID: {os.environ.get('JOB_ID')}")
... print(f"Accelerator: {os.environ.get('ACCELERATOR', 'none')}")
... print(f"CPU cores: {os.environ.get('CPU_CORES')}")
... print(f"Memory: {os.environ.get('MEMORY')}")
... """],
... )

>>> from huggingface_hub import run_job
>>> run_job(name="daily-report", image="python:3.12", command=["python", "report.py"])

>>> hf jobs run --name daily-report python:3.12 python report.py
>>> hf jobs labels <job_id> --name daily-report

>>> hf jobs ls -a --name daily-report

# Pass extra metadata with Labels
>>> from huggingface_hub import run_job
>>> run_job(
...     image="python:3.12",
...     command=["python", "-c", "import os; print(os.environ['MY_SECRET'])"],
...     labels={"my-label": "my-value", "foo": "bar"},
... )

>>> from huggingface_hub import update_job_labels
>>> update_job_labels(job_id=job_id, labels={"env": "prod", "team": "ml"})

>>> from huggingface_hub import update_scheduled_job_labels
>>> update_scheduled_job_labels(scheduled_job_id=scheduled_job_id, labels={"env": "staging"})

>>> hf jobs labels <job_id> --label env=prod --label team=ml
>>> hf jobs scheduled labels <scheduled_job_id> --label env=staging

>>> hf jobs labels <job_id> --clear

# Run a UV script (creates temporary repo)
>>> from huggingface_hub import run_uv_job
>>> run_uv_job("my_script.py")

# Run with GPU
>>> run_uv_job("ml_training.py", flavor="gpu-t4-small")

# Run with dependencies
>>> run_uv_job("inference.py", dependencies=["transformers", "torch"])

# Run a script directly from a URL
>>> run_uv_job("https://huggingface.co/datasets/username/scripts/resolve/main/example.py")

# Run a command
>>> run_uv_job("python", script_args=["-c", "import lighteval"], dependencies=["lighteval"])

# /// script
# requires-python = ">=3.11"
# dependencies = ["vllm", "datasets"]
#
# [tool.hf-jobs]
# image   = "vllm/vllm-openai:unlimited-ocr"
# flavor  = "l4x1"
# python  = "/usr/bin/python3"
# secrets = ["HF_TOKEN"]
# ///

hf jobs uv run \
 --flavor a10g-large \
 --image vllm/vllm-openai:latest \
...

# Schedule a job that runs every hour
>>> from huggingface_hub import create_scheduled_job
>>> create_scheduled_job(
...     image="python:3.12",
...     command=["python",  "-c", "print('This runs every hour!')"],
...     schedule="@hourly"
... )

# Use the CRON syntax
>>> create_scheduled_job(
...     image="python:3.12",
...     command=["python",  "-c", "print('This runs every 5 minutes!')"],
...     schedule="*/5 * * * *"
... )

# Schedule with GPU
>>> create_scheduled_job(
...     image="pytorch/pytorch:2.6.0-cuda12.4-cudnn9-devel",
...     command=["python",  "-c", 'import torch; print(f"This code ran with the following GPU: {torch.cuda.get_device_name()}")'],
...     schedule="@hourly",
...     flavor="a10g-small",
... )

# Schedule a UV script
>>> from huggingface_hub import create_scheduled_uv_job
>>> create_scheduled_uv_job("my_script.py", schedule="@hourly")

# List your active scheduled jobs
>>> from huggingface_hub import list_scheduled_jobs
>>> list_scheduled_jobs()

# Only list scheduled jobs with the given labels
>>> list_scheduled_jobs(labels={"env": "prod"})

# Inspect the status of a job
>>> from huggingface_hub import inspect_scheduled_job
>>> inspect_scheduled_job(scheduled_job_id=scheduled_job_id)

# Suspend (pause) a scheduled job
>>> from huggingface_hub import suspend_scheduled_job
>>> suspend_scheduled_job(scheduled_job_id=scheduled_job_id)

# Resume a scheduled job
>>> from huggingface_hub import resume_scheduled_job
>>> resume_scheduled_job(scheduled_job_id=scheduled_job_id)

# Change future run times without recreating the scheduled job
>>> from huggingface_hub import update_scheduled_job_schedule
>>> updated = update_scheduled_job_schedule(scheduled_job_id=scheduled_job_id, schedule="0 9 * * 1")
>>> updated.schedule
'0 9 * * 1'

# Trigger a scheduled job to run right now (does not change the schedule)
>>> from huggingface_hub import trigger_scheduled_job
>>> job = trigger_scheduled_job(scheduled_job_id=scheduled_job_id)
>>> job.url

# Delete a scheduled job
>>> from huggingface_hub import delete_scheduled_job
>>> delete_scheduled_job(scheduled_job_id=scheduled_job_id)

from huggingface_hub import create_webhook

# Example: Creating a webhook that triggers a Job
webhook = create_webhook(
    job_id=job_id,
    watched=[{"type": "user", "name": "your-username"}, {"type": "org", "name": "your-org-name"}],
    domains=["repo", "discussion"],
    secret="your-secret"
)