The azure-native:machinelearningservices:Job resource, part of the Pulumi Azure Native provider, defines Azure Machine Learning jobs: training runs, hyperparameter sweeps, and multi-step pipelines. This guide focuses on three capabilities: command jobs with distributed training, hyperparameter sweep configuration, and pipeline workflow orchestration.
Jobs execute within Azure ML workspaces and reference compute clusters, registered environments, and uploaded code assets. The examples are intentionally small. Combine them with your own workspace infrastructure, compute targets, and data assets.
Run a command job with distributed training
Teams training models start with command jobs that execute scripts on compute clusters, specifying code, runtime, and resources.
import * as pulumi from "@pulumi/pulumi";
import * as azure_native from "@pulumi/azure-native";
const job = new azure_native.machinelearningservices.Job("job", {
id: "string",
jobBaseProperties: {
codeId: "string",
command: "string",
computeId: "string",
description: "string",
displayName: "string",
distribution: {
distributionType: "TensorFlow",
parameterServerCount: 1,
workerCount: 1,
},
environmentId: "string",
environmentVariables: {
string: "string",
},
experimentName: "string",
identity: {
identityType: "AMLToken",
},
inputs: {
string: {
description: "string",
jobInputType: "literal",
value: "string",
},
},
jobType: "Command",
limits: {
jobLimitsType: "Command",
timeout: "PT5M",
},
outputs: {
string: {
description: "string",
jobOutputType: "uri_file",
mode: azure_native.machinelearningservices.OutputDeliveryMode.ReadWriteMount,
uri: "string",
},
},
properties: {
string: "string",
},
resources: {
instanceCount: 1,
instanceType: "string",
properties: {
string: {
"e6b6493e-7d5e-4db3-be1e-306ec641327e": null,
},
},
},
services: {
string: {
endpoint: "string",
jobServiceType: "string",
port: 1,
properties: {
string: "string",
},
},
},
tags: {
string: "string",
},
},
resourceGroupName: "test-rg",
workspaceName: "my-aml-workspace",
});
import pulumi
import pulumi_azure_native as azure_native
job = azure_native.machinelearningservices.Job("job",
id="string",
job_base_properties={
"code_id": "string",
"command": "string",
"compute_id": "string",
"description": "string",
"display_name": "string",
"distribution": {
"distribution_type": "TensorFlow",
"parameter_server_count": 1,
"worker_count": 1,
},
"environment_id": "string",
"environment_variables": {
"string": "string",
},
"experiment_name": "string",
"identity": {
"identity_type": "AMLToken",
},
"inputs": {
"string": {
"description": "string",
"job_input_type": "literal",
"value": "string",
},
},
"job_type": "Command",
"limits": {
"job_limits_type": "Command",
"timeout": "PT5M",
},
"outputs": {
"string": {
"description": "string",
"job_output_type": "uri_file",
"mode": azure_native.machinelearningservices.OutputDeliveryMode.READ_WRITE_MOUNT,
"uri": "string",
},
},
"properties": {
"string": "string",
},
"resources": {
"instance_count": 1,
"instance_type": "string",
"properties": {
"string": {
"e6b6493e-7d5e-4db3-be1e-306ec641327e": None,
},
},
},
"services": {
"string": {
"endpoint": "string",
"job_service_type": "string",
"port": 1,
"properties": {
"string": "string",
},
},
},
"tags": {
"string": "string",
},
},
resource_group_name="test-rg",
workspace_name="my-aml-workspace")
package main
import (
machinelearningservices "github.com/pulumi/pulumi-azure-native-sdk/machinelearningservices/v3"
"github.com/pulumi/pulumi/sdk/v3/go/pulumi"
)
func main() {
pulumi.Run(func(ctx *pulumi.Context) error {
_, err := machinelearningservices.NewJob(ctx, "job", &machinelearningservices.JobArgs{
Id: pulumi.String("string"),
JobBaseProperties: &machinelearningservices.CommandJobArgs{
CodeId: pulumi.String("string"),
Command: pulumi.String("string"),
ComputeId: pulumi.String("string"),
Description: pulumi.String("string"),
DisplayName: pulumi.String("string"),
Distribution: machinelearningservices.TensorFlow{
DistributionType: "TensorFlow",
ParameterServerCount: 1,
WorkerCount: 1,
},
EnvironmentId: pulumi.String("string"),
EnvironmentVariables: pulumi.StringMap{
"string": pulumi.String("string"),
},
ExperimentName: pulumi.String("string"),
Identity: machinelearningservices.AmlToken{
IdentityType: "AMLToken",
},
Inputs: pulumi.Map{
"string": machinelearningservices.LiteralJobInput{
Description: "string",
JobInputType: "literal",
Value: "string",
},
},
JobType: pulumi.String("Command"),
Limits: &machinelearningservices.CommandJobLimitsArgs{
JobLimitsType: pulumi.String("Command"),
Timeout: pulumi.String("PT5M"),
},
Outputs: pulumi.Map{
"string": machinelearningservices.UriFileJobOutput{
Description: "string",
JobOutputType: "uri_file",
Mode: machinelearningservices.OutputDeliveryModeReadWriteMount,
Uri: "string",
},
},
Properties: pulumi.StringMap{
"string": pulumi.String("string"),
},
Resources: &machinelearningservices.JobResourceConfigurationArgs{
InstanceCount: pulumi.Int(1),
InstanceType: pulumi.String("string"),
Properties: pulumi.Map{
"string": pulumi.Any(map[string]interface{}{
"e6b6493e-7d5e-4db3-be1e-306ec641327e": nil,
}),
},
},
Services: machinelearningservices.JobServiceMap{
"string": &machinelearningservices.JobServiceArgs{
Endpoint: pulumi.String("string"),
JobServiceType: pulumi.String("string"),
Port: pulumi.Int(1),
Properties: pulumi.StringMap{
"string": pulumi.String("string"),
},
},
},
Tags: pulumi.StringMap{
"string": pulumi.String("string"),
},
},
ResourceGroupName: pulumi.String("test-rg"),
WorkspaceName: pulumi.String("my-aml-workspace"),
})
if err != nil {
return err
}
return nil
})
}
using System.Collections.Generic;
using System.Linq;
using Pulumi;
using AzureNative = Pulumi.AzureNative;
return await Deployment.RunAsync(() =>
{
var job = new AzureNative.MachineLearningServices.Job("job", new()
{
Id = "string",
JobBaseProperties = new AzureNative.MachineLearningServices.Inputs.CommandJobArgs
{
CodeId = "string",
Command = "string",
ComputeId = "string",
Description = "string",
DisplayName = "string",
Distribution = new AzureNative.MachineLearningServices.Inputs.TensorFlowArgs
{
DistributionType = "TensorFlow",
ParameterServerCount = 1,
WorkerCount = 1,
},
EnvironmentId = "string",
EnvironmentVariables =
{
{ "string", "string" },
},
ExperimentName = "string",
Identity = new AzureNative.MachineLearningServices.Inputs.AmlTokenArgs
{
IdentityType = "AMLToken",
},
Inputs =
{
{ "string", new AzureNative.MachineLearningServices.Inputs.LiteralJobInputArgs
{
Description = "string",
JobInputType = "literal",
Value = "string",
} },
},
JobType = "Command",
Limits = new AzureNative.MachineLearningServices.Inputs.CommandJobLimitsArgs
{
JobLimitsType = "Command",
Timeout = "PT5M",
},
Outputs =
{
{ "string", new AzureNative.MachineLearningServices.Inputs.UriFileJobOutputArgs
{
Description = "string",
JobOutputType = "uri_file",
Mode = AzureNative.MachineLearningServices.OutputDeliveryMode.ReadWriteMount,
Uri = "string",
} },
},
Properties =
{
{ "string", "string" },
},
Resources = new AzureNative.MachineLearningServices.Inputs.JobResourceConfigurationArgs
{
InstanceCount = 1,
InstanceType = "string",
Properties =
{
{ "string", new Dictionary<string, object?>
{
["e6b6493e-7d5e-4db3-be1e-306ec641327e"] = null,
} },
},
},
Services =
{
{ "string", new AzureNative.MachineLearningServices.Inputs.JobServiceArgs
{
Endpoint = "string",
JobServiceType = "string",
Port = 1,
Properties =
{
{ "string", "string" },
},
} },
},
Tags =
{
{ "string", "string" },
},
},
ResourceGroupName = "test-rg",
WorkspaceName = "my-aml-workspace",
});
});
package generated_program;
import com.pulumi.Context;
import com.pulumi.Pulumi;
import com.pulumi.core.Output;
import com.pulumi.azurenative.machinelearningservices.Job;
import com.pulumi.azurenative.machinelearningservices.JobArgs;
import java.util.List;
import java.util.ArrayList;
import java.util.Map;
import java.io.File;
import java.nio.file.Files;
import java.nio.file.Paths;
public class App {
public static void main(String[] args) {
Pulumi.run(App::stack);
}
public static void stack(Context ctx) {
var job = new Job("job", JobArgs.builder()
.id("string")
.jobBaseProperties(CommandJobArgs.builder()
.codeId("string")
.command("string")
.computeId("string")
.description("string")
.displayName("string")
.distribution(TensorFlowArgs.builder()
.distributionType("TensorFlow")
.parameterServerCount(1)
.workerCount(1)
.build())
.environmentId("string")
.environmentVariables(Map.of("string", "string"))
.experimentName("string")
.identity(AmlTokenArgs.builder()
.identityType("AMLToken")
.build())
.inputs(Map.of("string", LiteralJobInputArgs.builder()
.description("string")
.jobInputType("literal")
.value("string")
.build()))
.jobType("Command")
.limits(Map.ofEntries(
Map.entry("jobLimitsType", "Command"),
Map.entry("timeout", "PT5M")
))
.outputs(Map.of("string", UriFileJobOutputArgs.builder()
.description("string")
.jobOutputType("uri_file")
.mode("ReadWriteMount")
.uri("string")
.build()))
.properties(Map.of("string", "string"))
.resources(JobResourceConfigurationArgs.builder()
.instanceCount(1)
.instanceType("string")
.properties(Map.of("string", Map.of("e6b6493e-7d5e-4db3-be1e-306ec641327e", null)))
.build())
.services(Map.of("string", JobServiceArgs.builder()
.endpoint("string")
.jobServiceType("string")
.port(1)
.properties(Map.of("string", "string"))
.build()))
.tags(Map.of("string", "string"))
.build())
.resourceGroupName("test-rg")
.workspaceName("my-aml-workspace")
.build());
}
}
resources:
job:
type: azure-native:machinelearningservices:Job
properties:
id: string
jobBaseProperties:
codeId: string
command: string
computeId: string
description: string
displayName: string
distribution:
distributionType: TensorFlow
parameterServerCount: 1
workerCount: 1
environmentId: string
environmentVariables:
string: string
experimentName: string
identity:
identityType: AMLToken
inputs:
string:
description: string
jobInputType: literal
value: string
jobType: Command
limits:
jobLimitsType: Command
timeout: PT5M
outputs:
string:
description: string
jobOutputType: uri_file
mode: ReadWriteMount
uri: string
properties:
string: string
resources:
instanceCount: 1
instanceType: string
properties:
string:
e6b6493e-7d5e-4db3-be1e-306ec641327e: null
services:
string:
endpoint: string
jobServiceType: string
port: 1
properties:
string: string
tags:
string: string
resourceGroupName: test-rg
workspaceName: my-aml-workspace
The command property defines the script to execute, while codeId points to your uploaded code asset. The distribution property configures TensorFlow distributed training across multiple workers and parameter servers. The resources block specifies instance type and count for the compute cluster. Environment variables pass configuration to your training script at runtime.
Optimize hyperparameters with sweep jobs
Model tuning requires testing multiple hyperparameter combinations. Sweep jobs automate this by running trials across a search space with early termination.
import * as pulumi from "@pulumi/pulumi";
import * as azure_native from "@pulumi/azure-native";
const job = new azure_native.machinelearningservices.Job("job", {
id: "string",
jobBaseProperties: {
computeId: "string",
description: "string",
displayName: "string",
earlyTermination: {
delayEvaluation: 1,
evaluationInterval: 1,
policyType: "MedianStopping",
},
experimentName: "string",
jobType: "Sweep",
limits: {
jobLimitsType: "Sweep",
maxConcurrentTrials: 1,
maxTotalTrials: 1,
trialTimeout: "PT1S",
},
objective: {
goal: azure_native.machinelearningservices.Goal.Minimize,
primaryMetric: "string",
},
properties: {
string: "string",
},
samplingAlgorithm: {
samplingAlgorithmType: "Grid",
},
searchSpace: {
string: {},
},
services: {
string: {
endpoint: "string",
jobServiceType: "string",
port: 1,
properties: {
string: "string",
},
},
},
tags: {
string: "string",
},
trial: {
codeId: "string",
command: "string",
distribution: {
distributionType: "Mpi",
processCountPerInstance: 1,
},
environmentId: "string",
environmentVariables: {
string: "string",
},
resources: {
instanceCount: 1,
instanceType: "string",
properties: {
string: {
"e6b6493e-7d5e-4db3-be1e-306ec641327e": null,
},
},
},
},
},
resourceGroupName: "test-rg",
workspaceName: "my-aml-workspace",
});
import pulumi
import pulumi_azure_native as azure_native
job = azure_native.machinelearningservices.Job("job",
id="string",
job_base_properties={
"compute_id": "string",
"description": "string",
"display_name": "string",
"early_termination": {
"delay_evaluation": 1,
"evaluation_interval": 1,
"policy_type": "MedianStopping",
},
"experiment_name": "string",
"job_type": "Sweep",
"limits": {
"job_limits_type": "Sweep",
"max_concurrent_trials": 1,
"max_total_trials": 1,
"trial_timeout": "PT1S",
},
"objective": {
"goal": azure_native.machinelearningservices.Goal.MINIMIZE,
"primary_metric": "string",
},
"properties": {
"string": "string",
},
"sampling_algorithm": {
"sampling_algorithm_type": "Grid",
},
"search_space": {
"string": {},
},
"services": {
"string": {
"endpoint": "string",
"job_service_type": "string",
"port": 1,
"properties": {
"string": "string",
},
},
},
"tags": {
"string": "string",
},
"trial": {
"code_id": "string",
"command": "string",
"distribution": {
"distribution_type": "Mpi",
"process_count_per_instance": 1,
},
"environment_id": "string",
"environment_variables": {
"string": "string",
},
"resources": {
"instance_count": 1,
"instance_type": "string",
"properties": {
"string": {
"e6b6493e-7d5e-4db3-be1e-306ec641327e": None,
},
},
},
},
},
resource_group_name="test-rg",
workspace_name="my-aml-workspace")
package main
import (
machinelearningservices "github.com/pulumi/pulumi-azure-native-sdk/machinelearningservices/v3"
"github.com/pulumi/pulumi/sdk/v3/go/pulumi"
)
func main() {
pulumi.Run(func(ctx *pulumi.Context) error {
_, err := machinelearningservices.NewJob(ctx, "job", &machinelearningservices.JobArgs{
Id: pulumi.String("string"),
JobBaseProperties: &machinelearningservices.SweepJobArgs{
ComputeId: pulumi.String("string"),
Description: pulumi.String("string"),
DisplayName: pulumi.String("string"),
EarlyTermination: machinelearningservices.MedianStoppingPolicy{
DelayEvaluation: 1,
EvaluationInterval: 1,
PolicyType: "MedianStopping",
},
ExperimentName: pulumi.String("string"),
JobType: pulumi.String("Sweep"),
Limits: &machinelearningservices.SweepJobLimitsArgs{
JobLimitsType: pulumi.String("Sweep"),
MaxConcurrentTrials: pulumi.Int(1),
MaxTotalTrials: pulumi.Int(1),
TrialTimeout: pulumi.String("PT1S"),
},
Objective: &machinelearningservices.ObjectiveArgs{
Goal: pulumi.String(machinelearningservices.GoalMinimize),
PrimaryMetric: pulumi.String("string"),
},
Properties: pulumi.StringMap{
"string": pulumi.String("string"),
},
SamplingAlgorithm: machinelearningservices.GridSamplingAlgorithm{
SamplingAlgorithmType: "Grid",
},
SearchSpace: pulumi.Any(map[string]interface{}{
"string": map[string]interface{}{},
}),
Services: machinelearningservices.JobServiceMap{
"string": &machinelearningservices.JobServiceArgs{
Endpoint: pulumi.String("string"),
JobServiceType: pulumi.String("string"),
Port: pulumi.Int(1),
Properties: pulumi.StringMap{
"string": pulumi.String("string"),
},
},
},
Tags: pulumi.StringMap{
"string": pulumi.String("string"),
},
Trial: &machinelearningservices.TrialComponentArgs{
CodeId: pulumi.String("string"),
Command: pulumi.String("string"),
Distribution: machinelearningservices.Mpi{
DistributionType: "Mpi",
ProcessCountPerInstance: 1,
},
EnvironmentId: pulumi.String("string"),
EnvironmentVariables: pulumi.StringMap{
"string": pulumi.String("string"),
},
Resources: &machinelearningservices.JobResourceConfigurationArgs{
InstanceCount: pulumi.Int(1),
InstanceType: pulumi.String("string"),
Properties: pulumi.Map{
"string": pulumi.Any(map[string]interface{}{
"e6b6493e-7d5e-4db3-be1e-306ec641327e": nil,
}),
},
},
},
},
ResourceGroupName: pulumi.String("test-rg"),
WorkspaceName: pulumi.String("my-aml-workspace"),
})
if err != nil {
return err
}
return nil
})
}
using System.Collections.Generic;
using System.Linq;
using Pulumi;
using AzureNative = Pulumi.AzureNative;
return await Deployment.RunAsync(() =>
{
var job = new AzureNative.MachineLearningServices.Job("job", new()
{
Id = "string",
JobBaseProperties = new AzureNative.MachineLearningServices.Inputs.SweepJobArgs
{
ComputeId = "string",
Description = "string",
DisplayName = "string",
EarlyTermination = new AzureNative.MachineLearningServices.Inputs.MedianStoppingPolicyArgs
{
DelayEvaluation = 1,
EvaluationInterval = 1,
PolicyType = "MedianStopping",
},
ExperimentName = "string",
JobType = "Sweep",
Limits = new AzureNative.MachineLearningServices.Inputs.SweepJobLimitsArgs
{
JobLimitsType = "Sweep",
MaxConcurrentTrials = 1,
MaxTotalTrials = 1,
TrialTimeout = "PT1S",
},
Objective = new AzureNative.MachineLearningServices.Inputs.ObjectiveArgs
{
Goal = AzureNative.MachineLearningServices.Goal.Minimize,
PrimaryMetric = "string",
},
Properties =
{
{ "string", "string" },
},
SamplingAlgorithm = new AzureNative.MachineLearningServices.Inputs.GridSamplingAlgorithmArgs
{
SamplingAlgorithmType = "Grid",
},
SearchSpace = new Dictionary<string, object?>
{
["string"] = new Dictionary<string, object?>
{
},
},
Services =
{
{ "string", new AzureNative.MachineLearningServices.Inputs.JobServiceArgs
{
Endpoint = "string",
JobServiceType = "string",
Port = 1,
Properties =
{
{ "string", "string" },
},
} },
},
Tags =
{
{ "string", "string" },
},
Trial = new AzureNative.MachineLearningServices.Inputs.TrialComponentArgs
{
CodeId = "string",
Command = "string",
Distribution = new AzureNative.MachineLearningServices.Inputs.MpiArgs
{
DistributionType = "Mpi",
ProcessCountPerInstance = 1,
},
EnvironmentId = "string",
EnvironmentVariables =
{
{ "string", "string" },
},
Resources = new AzureNative.MachineLearningServices.Inputs.JobResourceConfigurationArgs
{
InstanceCount = 1,
InstanceType = "string",
Properties =
{
{ "string", new Dictionary<string, object?>
{
["e6b6493e-7d5e-4db3-be1e-306ec641327e"] = null,
} },
},
},
},
},
ResourceGroupName = "test-rg",
WorkspaceName = "my-aml-workspace",
});
});
package generated_program;
import com.pulumi.Context;
import com.pulumi.Pulumi;
import com.pulumi.core.Output;
import com.pulumi.azurenative.machinelearningservices.Job;
import com.pulumi.azurenative.machinelearningservices.JobArgs;
import java.util.List;
import java.util.ArrayList;
import java.util.Map;
import java.io.File;
import java.nio.file.Files;
import java.nio.file.Paths;
public class App {
public static void main(String[] args) {
Pulumi.run(App::stack);
}
public static void stack(Context ctx) {
var job = new Job("job", JobArgs.builder()
.id("string")
.jobBaseProperties(SweepJobArgs.builder()
.computeId("string")
.description("string")
.displayName("string")
.earlyTermination(MedianStoppingPolicyArgs.builder()
.delayEvaluation(1)
.evaluationInterval(1)
.policyType("MedianStopping")
.build())
.experimentName("string")
.jobType("Sweep")
.limits(Map.ofEntries(
Map.entry("jobLimitsType", "Sweep"),
Map.entry("maxConcurrentTrials", 1),
Map.entry("maxTotalTrials", 1),
Map.entry("trialTimeout", "PT1S")
))
.objective(ObjectiveArgs.builder()
.goal("Minimize")
.primaryMetric("string")
.build())
.properties(Map.of("string", "string"))
.samplingAlgorithm(GridSamplingAlgorithmArgs.builder()
.samplingAlgorithmType("Grid")
.build())
.searchSpace(Map.of("string", Map.ofEntries(
)))
.services(Map.of("string", JobServiceArgs.builder()
.endpoint("string")
.jobServiceType("string")
.port(1)
.properties(Map.of("string", "string"))
.build()))
.tags(Map.of("string", "string"))
.trial(TrialComponentArgs.builder()
.codeId("string")
.command("string")
.distribution(MpiArgs.builder()
.distributionType("Mpi")
.processCountPerInstance(1)
.build())
.environmentId("string")
.environmentVariables(Map.of("string", "string"))
.resources(JobResourceConfigurationArgs.builder()
.instanceCount(1)
.instanceType("string")
.properties(Map.of("string", Map.of("e6b6493e-7d5e-4db3-be1e-306ec641327e", null)))
.build())
.build())
.build())
.resourceGroupName("test-rg")
.workspaceName("my-aml-workspace")
.build());
}
}
resources:
job:
type: azure-native:machinelearningservices:Job
properties:
id: string
jobBaseProperties:
computeId: string
description: string
displayName: string
earlyTermination:
delayEvaluation: 1
evaluationInterval: 1
policyType: MedianStopping
experimentName: string
jobType: Sweep
limits:
jobLimitsType: Sweep
maxConcurrentTrials: 1
maxTotalTrials: 1
trialTimeout: PT1S
objective:
goal: Minimize
primaryMetric: string
properties:
string: string
samplingAlgorithm:
samplingAlgorithmType: Grid
searchSpace:
string: {}
services:
string:
endpoint: string
jobServiceType: string
port: 1
properties:
string: string
tags:
string: string
trial:
codeId: string
command: string
distribution:
distributionType: Mpi
processCountPerInstance: 1
environmentId: string
environmentVariables:
string: string
resources:
instanceCount: 1
instanceType: string
properties:
string:
e6b6493e-7d5e-4db3-be1e-306ec641327e: null
resourceGroupName: test-rg
workspaceName: my-aml-workspace
The objective property defines what metric to optimize and whether to minimize or maximize it. The samplingAlgorithm determines how to explore the searchSpace (grid, random, or Bayesian). The earlyTermination policy stops underperforming trials early using median stopping rules, saving compute costs. The trial component defines the training script that runs for each hyperparameter combination, while limits controls maximum trials and concurrency.
Orchestrate multi-step workflows with pipeline jobs
Complex ML workflows chain multiple steps like data preparation, training, and evaluation. Pipeline jobs coordinate these steps and manage data flow.
import * as pulumi from "@pulumi/pulumi";
import * as azure_native from "@pulumi/azure-native";
const job = new azure_native.machinelearningservices.Job("job", {
id: "string",
jobBaseProperties: {
computeId: "string",
description: "string",
displayName: "string",
experimentName: "string",
inputs: {
string: {
description: "string",
jobInputType: "literal",
value: "string",
},
},
jobType: "Pipeline",
outputs: {
string: {
description: "string",
jobOutputType: "uri_file",
mode: azure_native.machinelearningservices.OutputDeliveryMode.Upload,
uri: "string",
},
},
properties: {
string: "string",
},
services: {
string: {
endpoint: "string",
jobServiceType: "string",
port: 1,
properties: {
string: "string",
},
},
},
settings: {},
tags: {
string: "string",
},
},
resourceGroupName: "test-rg",
workspaceName: "my-aml-workspace",
});
import pulumi
import pulumi_azure_native as azure_native
job = azure_native.machinelearningservices.Job("job",
id="string",
job_base_properties={
"compute_id": "string",
"description": "string",
"display_name": "string",
"experiment_name": "string",
"inputs": {
"string": {
"description": "string",
"job_input_type": "literal",
"value": "string",
},
},
"job_type": "Pipeline",
"outputs": {
"string": {
"description": "string",
"job_output_type": "uri_file",
"mode": azure_native.machinelearningservices.OutputDeliveryMode.UPLOAD,
"uri": "string",
},
},
"properties": {
"string": "string",
},
"services": {
"string": {
"endpoint": "string",
"job_service_type": "string",
"port": 1,
"properties": {
"string": "string",
},
},
},
"settings": {},
"tags": {
"string": "string",
},
},
resource_group_name="test-rg",
workspace_name="my-aml-workspace")
package main
import (
machinelearningservices "github.com/pulumi/pulumi-azure-native-sdk/machinelearningservices/v3"
"github.com/pulumi/pulumi/sdk/v3/go/pulumi"
)
func main() {
pulumi.Run(func(ctx *pulumi.Context) error {
_, err := machinelearningservices.NewJob(ctx, "job", &machinelearningservices.JobArgs{
Id: pulumi.String("string"),
JobBaseProperties: &machinelearningservices.PipelineJobArgs{
ComputeId: pulumi.String("string"),
Description: pulumi.String("string"),
DisplayName: pulumi.String("string"),
ExperimentName: pulumi.String("string"),
Inputs: pulumi.Map{
"string": machinelearningservices.LiteralJobInput{
Description: "string",
JobInputType: "literal",
Value: "string",
},
},
JobType: pulumi.String("Pipeline"),
Outputs: pulumi.Map{
"string": machinelearningservices.UriFileJobOutput{
Description: "string",
JobOutputType: "uri_file",
Mode: machinelearningservices.OutputDeliveryModeUpload,
Uri: "string",
},
},
Properties: pulumi.StringMap{
"string": pulumi.String("string"),
},
Services: machinelearningservices.JobServiceMap{
"string": &machinelearningservices.JobServiceArgs{
Endpoint: pulumi.String("string"),
JobServiceType: pulumi.String("string"),
Port: pulumi.Int(1),
Properties: pulumi.StringMap{
"string": pulumi.String("string"),
},
},
},
Settings: pulumi.Any(map[string]interface{}{}),
Tags: pulumi.StringMap{
"string": pulumi.String("string"),
},
},
ResourceGroupName: pulumi.String("test-rg"),
WorkspaceName: pulumi.String("my-aml-workspace"),
})
if err != nil {
return err
}
return nil
})
}
using System.Collections.Generic;
using System.Linq;
using Pulumi;
using AzureNative = Pulumi.AzureNative;
return await Deployment.RunAsync(() =>
{
var job = new AzureNative.MachineLearningServices.Job("job", new()
{
Id = "string",
JobBaseProperties = new AzureNative.MachineLearningServices.Inputs.PipelineJobArgs
{
ComputeId = "string",
Description = "string",
DisplayName = "string",
ExperimentName = "string",
Inputs =
{
{ "string", new AzureNative.MachineLearningServices.Inputs.LiteralJobInputArgs
{
Description = "string",
JobInputType = "literal",
Value = "string",
} },
},
JobType = "Pipeline",
Outputs =
{
{ "string", new AzureNative.MachineLearningServices.Inputs.UriFileJobOutputArgs
{
Description = "string",
JobOutputType = "uri_file",
Mode = AzureNative.MachineLearningServices.OutputDeliveryMode.Upload,
Uri = "string",
} },
},
Properties =
{
{ "string", "string" },
},
Services =
{
{ "string", new AzureNative.MachineLearningServices.Inputs.JobServiceArgs
{
Endpoint = "string",
JobServiceType = "string",
Port = 1,
Properties =
{
{ "string", "string" },
},
} },
},
Settings = null,
Tags =
{
{ "string", "string" },
},
},
ResourceGroupName = "test-rg",
WorkspaceName = "my-aml-workspace",
});
});
package generated_program;
import com.pulumi.Context;
import com.pulumi.Pulumi;
import com.pulumi.core.Output;
import com.pulumi.azurenative.machinelearningservices.Job;
import com.pulumi.azurenative.machinelearningservices.JobArgs;
import java.util.List;
import java.util.ArrayList;
import java.util.Map;
import java.io.File;
import java.nio.file.Files;
import java.nio.file.Paths;
public class App {
public static void main(String[] args) {
Pulumi.run(App::stack);
}
public static void stack(Context ctx) {
var job = new Job("job", JobArgs.builder()
.id("string")
.jobBaseProperties(PipelineJobArgs.builder()
.computeId("string")
.description("string")
.displayName("string")
.experimentName("string")
.inputs(Map.of("string", LiteralJobInputArgs.builder()
.description("string")
.jobInputType("literal")
.value("string")
.build()))
.jobType("Pipeline")
.outputs(Map.of("string", UriFileJobOutputArgs.builder()
.description("string")
.jobOutputType("uri_file")
.mode("Upload")
.uri("string")
.build()))
.properties(Map.of("string", "string"))
.services(Map.of("string", JobServiceArgs.builder()
.endpoint("string")
.jobServiceType("string")
.port(1)
.properties(Map.of("string", "string"))
.build()))
.settings(Map.ofEntries(
))
.tags(Map.of("string", "string"))
.build())
.resourceGroupName("test-rg")
.workspaceName("my-aml-workspace")
.build());
}
}
resources:
job:
type: azure-native:machinelearningservices:Job
properties:
id: string
jobBaseProperties:
computeId: string
description: string
displayName: string
experimentName: string
inputs:
string:
description: string
jobInputType: literal
value: string
jobType: Pipeline
outputs:
string:
description: string
jobOutputType: uri_file
mode: Upload
uri: string
properties:
string: string
services:
string:
endpoint: string
jobServiceType: string
port: 1
properties:
string: string
settings: {}
tags:
string: string
resourceGroupName: test-rg
workspaceName: my-aml-workspace
The inputs and outputs properties define data contracts between pipeline steps. The settings property would contain the pipeline component definitions (shown as empty in this example). Pipeline jobs enable reusable workflow definitions where each step can reference different compute targets and environments.
Beyond these examples
These snippets focus on specific job-level features: command jobs with distributed training, hyperparameter sweep with early termination, and pipeline orchestration. They’re intentionally minimal rather than full ML workflows.
The examples reference pre-existing infrastructure such as Azure Machine Learning workspaces, compute clusters (referenced by computeId), and registered environments and code assets. They focus on job configuration rather than provisioning the workspace and compute infrastructure.
To keep things focused, common job patterns are omitted, including:
- AutoML jobs for automated model selection
- Queue settings and priority configuration
- Notification and monitoring hooks
- Component-based pipeline definitions (jobs property)
- Resource limits and timeout controls
- Identity and credential management beyond AMLToken
These omissions are intentional: the goal is to illustrate how each job type is wired, not provide drop-in ML training modules. See the Azure Machine Learning Job resource reference for all available configuration options.
Let's configure Azure Machine Learning Jobs
Get started with Pulumi Cloud, then follow our quick setup guide to deploy this infrastructure.
Try Pulumi Cloud for FREEFrequently Asked Questions
Job Configuration & Immutability
What properties can't I change after creating a job?
The
id, resourceGroupName, and workspaceName properties are immutable. Plan these values carefully before creating the job, as modifications require recreating the resource.What job types are available in Azure Machine Learning?
Four job types are supported:
AutoML (automated machine learning), Command (single command execution), Pipeline (workflow orchestration), and Sweep (hyperparameter tuning). Specify the type using the jobType property in jobBaseProperties.How do I specify timeouts for jobs?
Use ISO 8601 duration format for timeout values. For example,
PT5M for 5 minutes or PT1S for 1 second. Command jobs use timeout in the limits property, while Sweep jobs use trialTimeout.Distributed Training & Outputs
How do I configure distributed training for a job?
Set the
distribution property with a distribution type. For TensorFlow, specify parameterServerCount and workerCount. For MPI, specify processCountPerInstance. The distribution type determines which configuration options are available.What's the difference between ReadWriteMount and Upload output modes?
Output modes control how job outputs are delivered.
ReadWriteMount mounts the output location for read/write access during job execution, while Upload uploads outputs after job completion. Specify the mode in the output configuration.API Versions & Import
Which API version should I use for Azure ML jobs?
The default API version is 2025-09-01. Version 2.x of the Azure Native provider used 2023-04-01. To access other versions (2021-03-01-preview through 2025-07-01-preview), use the CLI command:
pulumi package add azure-native machinelearningservices [ApiVersion]How do I import an existing Azure ML job into Pulumi?
Use the import command with the job’s resource identifier:
pulumi import azure-native:machinelearningservices:Job <name> /subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.MachineLearningServices/workspaces/{workspaceName}/jobs/{id}