Python SDK 0.28.0

The W&B Python SDK, accessible at wandb, enables you to train and fine-tune models, and manage models from experimentation to production.

After performing your training and fine-tuning operations with this SDK, you can use the Public API to query and analyze the data that was logged, and the Reports and Workspaces API to generate a web-publishable report summarizing your work.

To authenticate your machine with W&B, you must first generate an API key at User Settings.

Install and import packages

Install the W&B Python SDK using pip:

pip install wandb

Import W&B Python SDK

The following code snippet demonstrates how to import the W&B Python SDK and initialize a run. Replace <team_entity> with your team entity name.

import wandb

# Specify your team entity
entity = "<team_entity>"

# Project that the run is recorded to
project = "my-awesome-project"

with wandb.init(entity=entity, project=project) as run:
   run.log({"accuracy": 0.9, "loss": 0.1})

Python SDK extras

Install optional Python extras to extend the functionality of the W&B Python SDK. Specify the name of the extra you want to install within square brackets after wandb. The syntax is:

pip install wandb[extra]

For example, to install W&B with Google Cloud Storage support, run:

pip install wandb[gcp]

Install more than one optional dependency by separating them with commas:

pip install wandb[gcp,aws,media]

The following table lists Python SDK extras and their suggested use cases.

Extra	Packages included	Install if you
`gcp`	google-cloud-storage	Use `gs://` artifact references.
`aws`	boto3, botocore	Use `s3://` artifact references.
`azure`	azure-identity, azure-storage-blob	Use Azure Blob Storage artifact references.
`media`	numpy, moviepy, imageio, pillow, bokeh, soundfile, plotly, rdkit	Log images, video, audio, or plots from raw data (numpy arrays, tensors).
`sweeps`	sweeps	Run local sweep controller (`wandb.controller()`).
`workspaces`	wandb-workspaces	Programmatically manage workspaces.

The following tabs provide more details about each extra, including installation instructions, dependencies, and code examples.

Use the gcp extra if you add reference artifacts that start with gs:// URIs.Installation:

pip install wandb[gcp]

Dependencies:

google-cloud-storage

Example:

import wandb

artifact = wandb.Artifact("my-artifact", type="dataset")
artifact.add_reference("gs://bucket/path/to/file")

with wandb.init() as run:
    run.log_artifact(artifact)

Use the aws extra if you add reference artifacts that start with s3:// URIs.Installation:

pip install wandb[aws]

Dependencies:

boto3
botocore>=1.5.76

Example:

import wandb

artifact = wandb.Artifact("my-artifact", type="dataset")
artifact.add_reference("s3://bucket/path/to/file")

with wandb.init() as run:
    run.log_artifact(artifact)

Use the azure extra if you add artifact references to Azure Blob Storage (URLs ending in .blob.core.windows.net).Installation:

pip install wandb[azure]

Dependencies:

azure-identity
azure-storage-blob

Example:

import wandb

artifact = wandb.Artifact("my-artifact", type="dataset")
artifact.add_reference("https://<account>.blob.core.windows.net/container/blob")

with wandb.init() as run:
    run.log_artifact(artifact)

Use the media extra if you log images, video, audio, or plots from raw data (numpy arrays, tensors).Installation:

pip install wandb[media]

Dependencies:

numpy
moviepy>=1.0.0
imageio>=2.28.1
pillow
bokeh
soundfile
plotly>=5.18.0
rdkit

The following table lists the dependencies included in the media extra and their use cases:

Dependency	Use Case
pillow	Logging images with `wandb.Image()` with numpy arrays or tensors (wandb/sdk/data_types/image.py)
moviepy, imageio	Logging videos with `wandb.Video()` with numpy arrays (wandb/sdk/data_types/video.py)
soundfile	Logging audio with `wandb.Audio()` with raw numpy data (wandb/sdk/data_types/audio.py)
plotly	Logging interactive charts with `wandb.Plotly()` (wandb/sdk/data_types/plotly.py)
bokeh	Logging Bokeh plots with `wandb.Bokeh()` (wandb/sdk/data_types/bokeh.py)
rdkit	Logging molecular structures with `wandb.Molecule.from_rdkit()` (wandb/sdk/data_types/molecule.py)

Use the sweeps extra if you run a local sweep controller with wandb.controller().Installation:

pip install wandb[sweeps]

Dependencies:

sweeps>=0.2.0

The cloud-based sweep agent (wandb agent) does NOT require this extra.

Use the workspaces extra if you programmatically create or edit W&B workspaces with wandb.apis.workspaces.Installation:

pip install wandb[workspaces]

Dependencies:

wandb-workspaces

W&B Models

W&B Weave

Serverless Inference

Serverless Training

Install and import packages

Import W&B Python SDK

Python SDK extras

​Sign up and create an API key

​Install and import packages

​Import W&B Python SDK

​Python SDK extras

Sign up and create an API key

Install and import packages

Import W&B Python SDK

Python SDK extras