Weights and Biases

What's Changed

Notable Changes

This version drops support for protobuf v4, and requires protobuf v5 or newer.

This version drops compatibility with server versions older than 0.70.0.

Added

• New filters parameter to Api().project().sweeps() matching the runs filter format (@kmikowicz-wandb in https://github.com/wandb/wandb/pull/12059)
• Added Run.stop() to the public API (wandb.Api().run(...).stop()) to programmatically request that an active run stop gracefully, like the "Stop run" button in the W&B App UI (@dmitryduev in https://github.com/wandb/wandb/pull/12161)
• Added wandb.Api().organization() to fetch an Organization by name, or the current default organization for the user (@tonyyli-wandb in https://github.com/wandb/wandb/pull/12196)

Changed

• Remove temporary Unix socket files and directories on shutdown (@geoffhardy in https://github.com/wandb/wandb/pull/12058)
• wandb beta sync now skips online runs by default like wandb sync (@timoffex in https://github.com/wandb/wandb/pull/12087)
• wandb sync now routes to wandb beta sync for supported parameter sets (@timoffex in https://github.com/wandb/wandb/pull/12093)
• Restore original behavior with --legacy
• Dropped support for protobuf v4 (@jacobromero in https://github.com/wandb/wandb/pull/12115)
• wandb.sandbox now defaults serverless sandboxes to a 12-hour max lifetime (max_lifetime_seconds=43200); override per sandbox with max_lifetime_seconds or via SandboxDefaults (@nicholaspun-wandb in https://github.com/wandb/wandb/pull/12136)
• wandb.Api().runs() now raises a RunNotFoundError when unable to load data for a run, such as when a run is deleted prior to fully loading run data (@jacobromero in https://app.graphite.com/github/pr/wandb/wandb/12176)
• wandb.save now has an option glob=False that disables glob expansion (*, ?, [...]) for file paths; updated docs for improved explanation (@geoffhardy in https://github.com/wandb/wandb/pull/12192)

Removed

• Removed legacy fallback implementations for downloading artifact files on older EOL W&B Server releases. The following will no longer work on EOL servers: Artifact.files() and Artifact.download() on any artifact, as well as Artifact.get_entry() / Artifact.get_path() file downloads on non-reference artifacts. (@tonyyli-wandb in https://github.com/wandb/wandb/pull/12109)
• To continue using these operations, upgrade your W&B Server to v0.70.0 or newer.
• Removed legacy fallback implementations for fetching an artifact by name on older EOL W&B Server releases. The following will no longer work on EOL servers: wandb.Api().artifact(...) and other methods that fetch artifact(s) by their path. (@tonyyli-wandb in https://github.com/wandb/wandb/pull/12112)
• To continue using these operations, upgrade your W&B Server to v0.70.0 or newer.
• Removed the GitPython dependency. Git metadata is collected by invoking the git executable directly; the GIT_PYTHON_GIT_EXECUTABLE environment variable is still honored for locating it (@dmitryduev in https://github.com/wandb/wandb/pull/11983)

Fixed

• Saving a linked registry artifact (for example, when adding an alias) no longer fails when the caller lacks write access to the source project (@ibindlish in https://github.com/wandb/wandb/pull/12075)
• np.float16/np.float32 NaN values logged with Run.log() are now recorded as NaN instead of being silently dropped, matching np.float64 and native float (@dmitryduev in https://github.com/wandb/wandb/pull/12116)
• Run.upload_file() (via wandb.Api().run(...)) now registers the uploaded file with the run on self-hosted servers. Previously the file's bytes were uploaded but never committed, so the file did not appear on the run on deployments without object-store notifications (@dmitryduev in https://github.com/wandb/wandb/pull/12121)
• Sweep agents will now allow the in-progress run to complete before

<p align="center">

<img src="./assets/logo.svg" width="600" alt="Weights & Biases" />

</p>

<p align="center">

<a href="https://pypi.python.org/pypi/wandb"><img src="https://img.shields.io/pypi/v/wandb" /></a>

<a href="https://anaconda.org/conda-forge/wandb"><img src="https://img.shields.io/conda/vn/conda-forge/wandb" /></a>

<a href="https://pypi.python.org/pypi/wandb"><img src="https://img.shields.io/pypi/pyversions/wandb" /></a>

<a href="https://circleci.com/gh/wandb/wandb"><img src="https://img.shields.io/circleci/build/github/wandb/wandb/main" /></a>

<a href="https://codecov.io/gh/wandb/wandb"><img src="https://img.shields.io/codecov/c/gh/wandb/wandb" /></a>

</p>

<p align='center'>

<a href="https://colab.research.google.com/github/wandb/examples/blob/master/colabs/intro/Intro_to_Weights_%26_Biases.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" /></a>

</p>

Use W&B to build better models faster. Track and visualize all the pieces of your machine learning pipeline, from datasets to production machine learning models. Get started with W&B today, sign up for a W&B account!

<br>

Building an LLM app? Track, debug, evaluate, and monitor LLM apps with Weave, our new suite of tools for GenAI.

&nbsp;

Documentation

See the W&B Developer Guide and API Reference Guide for a full technical description of the W&B platform.

&nbsp;

Quickstart

Install W&B to track, visualize, and manage machine learning experiments of any size.

Install the wandb library

pip install wandb

Sign up and create an API key

Sign up for a W&B account. Create a new API key at wandb.ai/settings and store it securely. Optionally, use the wandb login CLI to configure your API key on your machine. You can skip this step -- W&B will prompt you to create an API key the first time you use it.

Note: API keys can only be viewed once when created. Store your API key in a secure location like a password manager or environment variable.

Create a machine learning training experiment

In your Python script or notebook, initialize a W&B run with wandb.init().

Specify hyperparameters and log metrics and other information to W&B.

import wandb

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

# Dictionary with hyperparameters
config = {"epochs": 1337, "lr": 3e-4}

# The `with` syntax marks the run as finished upon exiting the `with` block,
# and it marks the run "failed" if there's an exception.
#
# In a notebook, it may be more convenient to write `run = wandb.init()`
# and manually call `run.finish()` instead of using a `with` block.
with wandb.init(project=project, config=config) as run:
    # Training code here

    # Log values to W&B with run.log()
    run.log({"accuracy": 0.9, "loss": 0.1})

Visit wandb.ai/home to view recorded metrics such as accuracy and loss and how they changed during each training step. Each run object appears in the Runs column with generated names.

&nbsp;

Integrations

W&B integrates with popular ML frameworks and libraries making it fast and easy to set up experiment tracking and data versioning inside existing projects.

For developers adding W&B to a new framework, follow the W&B Developer Guide.

&nbsp;

W&B Hosting Options

Weights & Biases is available in the cloud or installed on your private infrastructure. Set up a W&B Server in a production environment in one of three ways:

1. Multi-tenant Cloud: Fully managed platform deployed in W&B’s Google Cloud Platform (GCP) account in GCP’s North America regions.
2. Dedicated Cloud: Single-tenant, fully managed platform deployed in W&B’s AWS, GCP, or Azure cloud accounts. Each Dedicated Cloud instance has its own isolated network, compute and storage from other W&B Dedicated Cloud instances.
3. Self-Managed: Deploy W&B Server on your AWS, GCP, or Azure cloud account or within your on-premises infrastructure.

See the Hosting documentation in the W&B Developer Guide for more information.

&nbsp;

Python Version Support

We are committed to supporting our minimum required Python version for _at least_ six months after its official end-of-life (EOL) date, as defined by the Python Software Foundation. You can find a list of Python EOL dates here.

When we discontinue support for a Python version, we will increment the library’s minor version number to reflect this change.

&nbsp;

Contribution guidelines

Weights & Biases ❤️ open source, and we welcome contributions from the community! See the Contribution Guide and the docs/ directory for more information on the development workflow and the internals of the wandb library. For wandb bugs and feature requests, visit GitHub Issues or contact support@wandb.com.

&nbsp;

W&B Community

Be a part of the growing W&B Community and interact with the W&B team in our Discord. Stay connected with the latest AI updates and tutorials with W&B Fully Connected.

&nbsp;

License

MIT License