azure-ai-ml-py
Azure Machine Learning SDK v2 for Python. Use for ML workspaces, jobs, models, datasets, compute, and pipelines. Triggers: "azure-ai-ml", "MLClient", "workspace", "model registry", "training jobs", "datasets".
- License 许可
- MIT
- 所属仓库
- microsoft/skills
- 最近更新
- 2026-11-14
- Stars
- ★ 3.2k
- Forks
- 368
01 前言 Front Matter
- name
- azure-ai-ml-py
- license
- MIT
- metadata
- { "author": "Microsoft", "package": "azure-ai-ml", "version": "1.0.0" }
- description
- Azure Machine Learning SDK v2 for Python. Use for ML workspaces, jobs, models, datasets, compute, and pipelines. Triggers: "azure-ai-ml", "MLClient", "workspace", "model registry", "training jobs", "datasets".
02 资源文件 2
- 参考文档references/capabilities.md1.6 KB
- 参考文档references/non-hero-scenarios.md2.4 KB
03 SKILL.md 正文
Azure Machine Learning SDK v2 for Python
Client library for managing Azure ML resources: workspaces, jobs, models, data, and compute.
Installation
pip install azure-ai-ml
Environment Variables
AZURE_SUBSCRIPTION_ID=<your-subscription-id> # Required for all auth methods
AZURE_RESOURCE_GROUP=<your-resource-group> # Required for all auth methods
AZURE_ML_WORKSPACE_NAME=<your-workspace-name> # Required for all auth methods
AZURE_TOKEN_CREDENTIALS=prod # Required only if DefaultAzureCredential is used in production
Authentication & Lifecycle
? Two rules apply to every code sample below:
- Prefer
DefaultAzureCredential. It works locally (Azure CLI / VS Code / Developer CLI) and in Azure (managed identity, workload identity) with no code change. Avoid connection strings, account/API keys — they bypass Entra audit and rotation.
- Local dev:
DefaultAzureCredentialworks as-is.- Production: set
AZURE_TOKEN_CREDENTIALS=prod(orAZURE_TOKEN_CREDENTIALS=<specific_credential>) to constrain the credential chain to production-safe credentials.- Wrap every client in a context manager so HTTP transports, sockets, and token caches are released deterministically:
- Sync:
with <Client>(...) as client:- Async:
async with <Client>(...) as client:andasync with DefaultAzureCredential() as credential:(fromazure.identity.aio)Snippets may abbreviate this setup, but production code should always follow both rules.
from azure.ai.ml import MLClient
from azure.identity import DefaultAzureCredential, ManagedIdentityCredential
import os
# Local dev: DefaultAzureCredential. Production: set AZURE_TOKEN_CREDENTIALS=prod or AZURE_TOKEN_CREDENTIALS=<specific_credential>
credential = DefaultAzureCredential(require_envvar=True)
# Or use a specific credential directly in production:
# See https://learn.microsoft.com/python/api/overview/azure/identity-readme?view=azure-python#credential-classes
# credential = ManagedIdentityCredential()
with MLClient(
credential=credential,
subscription_id=os.environ["AZURE_SUBSCRIPTION_ID"],
resource_group_name=os.environ["AZURE_RESOURCE_GROUP"],
workspace_name=os.environ["AZURE_ML_WORKSPACE_NAME"]
) as ml_client:
for ws in ml_client.workspaces.list():
print(ws.name)
From Config File
from azure.ai.ml import MLClient
from azure.identity import DefaultAzureCredential
# Uses config.json in current directory or parent
with MLClient.from_config(
credential=DefaultAzureCredential()
) as ml_client:
for ws in ml_client.workspaces.list():
print(ws.name)
Long-lived
ml_client: Subsequent examples in this skill assumeml_clientwas created via the pattern above and is alive for the lifetime of your script. In production, wrap your top-level workflow in a singlewith MLClient(...) as ml_client:block so the underlying HTTP transport closes cleanly on exit.
Workspace Management
Create Workspace
from azure.ai.ml.entities import Workspace
ws = Workspace(
name="my-workspace",
location="eastus",
display_name="My Workspace",
description="ML workspace for experiments",
tags={"purpose": "demo"}
)
ml_client.workspaces.begin_create(ws).result()
List Workspaces
for ws in ml_client.workspaces.list():
print(f"{ws.name}: {ws.location}")
Data Assets
Register Data
from azure.ai.ml.entities import Data
from azure.ai.ml.constants import AssetTypes
# Register a file
my_data = Data(
name="my-dataset",
version="1",
path="azureml://datastores/workspaceblobstore/paths/data/train.csv",
type=AssetTypes.URI_FILE,
description="Training data"
)
ml_client.data.create_or_update(my_data)
Register Folder
my_data = Data(
name="my-folder-dataset",
version="1",
path="azureml://datastores/workspaceblobstore/paths/data/",
type=AssetTypes.URI_FOLDER
)
ml_client.data.create_or_update(my_data)
Model Registry
Register Model
from azure.ai.ml.entities import Model
from azure.ai.ml.constants import AssetTypes
model = Model(
name="my-model",
version="1",
path="./model/",
type=AssetTypes.CUSTOM_MODEL,
description="My trained model"
)
ml_client.models.create_or_update(model)
List Models
for model in ml_client.models.list(name="my-model"):
print(f"{model.name} v{model.version}")
Compute
Create Compute Cluster
from azure.ai.ml.entities import AmlCompute
cluster = AmlCompute(
name="cpu-cluster",
type="amlcompute",
size="Standard_DS3_v2",
min_instances=0,
max_instances=4,
idle_time_before_scale_down=120
)
ml_client.compute.begin_create_or_update(cluster).result()
List Compute
for compute in ml_client.compute.list():
print(f"{compute.name}: {compute.type}")
Jobs
Command Job
from azure.ai.ml import command, Input
job = command(
code="./src",
command="python train.py --data ${{inputs.data}} --lr ${{inputs.learning_rate}}",
inputs={
"data": Input(type="uri_folder", path="azureml:my-dataset:1"),
"learning_rate": 0.01
},
environment="AzureML-sklearn-1.0-ubuntu20.04-py38-cpu@latest",
compute="cpu-cluster",
display_name="training-job"
)
returned_job = ml_client.jobs.create_or_update(job)
print(f"Job URL: {returned_job.studio_url}")
Monitor Job
ml_client.jobs.stream(returned_job.name)
Pipelines
from azure.ai.ml import dsl, Input, Output
from azure.ai.ml.entities import Pipeline
@dsl.pipeline(
compute="cpu-cluster",
description="Training pipeline"
)
def training_pipeline(data_input):
prep_step = prep_component(data=data_input)
train_step = train_component(
data=prep_step.outputs.output_data,
learning_rate=0.01
)
return {"model": train_step.outputs.model}
pipeline = training_pipeline(
data_input=Input(type="uri_folder", path="azureml:my-dataset:1")
)
pipeline_job = ml_client.jobs.create_or_update(pipeline)
Environments
Create Custom Environment
from azure.ai.ml.entities import Environment
env = Environment(
name="my-env",
version="1",
image="mcr.microsoft.com/azureml/openmpi4.1.0-ubuntu20.04",
conda_file="./environment.yml"
)
ml_client.environments.create_or_update(env)
Datastores
List Datastores
for ds in ml_client.datastores.list():
print(f"{ds.name}: {ds.type}")
Get Default Datastore
default_ds = ml_client.datastores.get_default()
print(f"Default: {default_ds.name}")
MLClient Operations
| Property | Operations |
|---|---|
workspaces |
create, get, list, delete |
jobs |
create_or_update, get, list, stream, cancel |
models |
create_or_update, get, list, archive |
data |
create_or_update, get, list |
compute |
begin_create_or_update, get, list, delete |
environments |
create_or_update, get, list |
datastores |
create_or_update, get, list, get_default |
components |
create_or_update, get, list |
Best Practices
- Pick sync OR async and stay consistent. Do not mix
azure.ai.mlsync clients withazure.ai.mlasync clients in the same call path. Choose one mode per module. - Always use context managers for clients and async credentials. Wrap every client in
with MLClient(...) as client:(sync) orasync with MLClient(...) as client:(async). For asyncDefaultAzureCredentialfromazure.identity.aio, also useasync with credential:so tokens and transports are cleaned up. - Use versioning for data, models, and environments
- Configure idle scale-down to reduce compute costs
- Use environments for reproducible training
- Stream job logs to monitor progress
- Register models after successful training jobs
- Use pipelines for multi-step workflows
- Tag resources for organization and cost tracking
Reference Files
| File | Contents |
|---|---|
| references/capabilities.md | Additional non-hero capabilities, operation-group coverage, and production checklists. |
| references/non-hero-scenarios.md | Dedicated non-hero examples for secondary/advanced scenarios. |
04 同仓库更多 Skills
azure-advisor-cost-optimization
Analyze Azure Advisor Cost/IdleResources findings and prevalidate deletion safety, including Service Groups. WHEN: "find unused Azure resources", "what resources am I paying for but no longer using", "idle-resource savings", "historical idle-resource savings", "are these idle resources safe to delete". Generic Cost, other subcategories, and cross-category requests use azure-advisor.
skill-creator
Guide for creating effective skills for AI coding agents working with Azure SDKs and Microsoft Foundry services. Use when creating new skills or updating existing skills.
podcast-generation
Generate AI-powered podcast-style audio narratives using Azure OpenAI's GPT Realtime Mini model via WebSocket. Use when building text-to-speech features, audio narrative generation, podcast creation from content, or integrating with Azure OpenAI Realtime API for real audio output. Covers full-stack implementation from React frontend to Python FastAPI backend with WebSocket streaming.
microsoft-docs
Understand Microsoft technologies by querying official documentation. Use whenever the user asks how something works, wants tutorials, needs configuration options, limits, quotas, or best practices for any Microsoft technology (Azure, .NET, M365, Windows, Power Platform, etc.)—even if they don't mention "docs." If the question is about understanding a concept rather than writing code, this is the right skill.
mcp-builder
Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP), Node/TypeScript (MCP SDK), or C#/.NET (Microsoft MCP SDK).
kql
KQL language expertise for writing correct, efficient Kusto Query Language queries. Covers syntax gotchas, join patterns, dynamic types, datetime pitfalls, regex patterns, serialization, memory management, result-size discipline, and advanced functions (geo, vector, graph). USE THIS SKILL whenever writing, debugging, or reviewing KQL queries — even simple ones — because the gotchas section prevents the most common errors that waste tool calls and cause expensive retry cascades. Trigger on: KQL, Kusto, ADX, Azure Data Explorer, Fabric Real-Time Intelligence, EventHouse, Log Analytics, log analysis, data exploration, time series, anomaly detection, summarize, where clause, join, extend, project, let statement, parse operator, extract function, any mention of pipe-forward query syntax.