wandb

Форк
0

README.md

wandb-core: A New Backend for the W&B SDK

PyPI version PyPI - License

Introduction

Good News, Everyone! We've developed a new and improved backend for the W&B SDK that is more performant, versatile, and robust.

Getting Started

To start using wandb-core, simply install the package in your environment. The wandb library will automatically detect and utilize it:

pip install -U wandb wandb-core

Note: ensure you have wandb>=0.16.1.

Platform Compatibility

wandb-core comes with wheels pre-built for the following platforms:

  • Linux:x86_64, aarch64
  • macOS: x86_64, arm64
  • Windows amd64

For other platforms, build wandb-core from the source as outlined in our contributing guide. If you're interested in support for additional platforms, please inform us by opening a GitHub issue. Your feedback helps us prioritize new platform support.

W&B Server Compatibility

wandb-core is compatible with our production and dedicated cloud deployments as well as W&B Server versions >=0.40.0.

Switching Back to the Old SDK Backend

To revert to the old SDK backend, simply uninstall wandb-core from your environment:

pip uninstall wandb-core

Contributing

Your contributions are welcome! Check our contributing guide for instructions on setting up your development environment and contributing to the project.

Feedback and Bug Reporting

We're eager to hear your thoughts on wandb-core. Your feedback, especially bug reports, is invaluable. If you encounter any issues, please raise a GitHub issue and mention your use of wandb-core.

Feature Support Status

Below is an overview of the feature support status in the wandb-core version 0.17.0b9.

Status legend:

  • ✅: Available: The feature is relatively stable and ready for use.
  • 🚧: In Development: The feature is available but may be unstable or incomplete.
  • ❌: Not Available: The feature is not yet available.
CategoryFeatureStatus
Experiments
init1
log
log_artifact
log_code
config
summary
define_metric🚧2
tags
notes
name
alert
save
restore
mark_preempting
resume
reinit
Media
Grouping
anonymous mode
offline mode
disabled mode
multiprocessing
TensorBoard sync🚧1
console logging3
system metrics4
system info
auto code saving
Settings🚧5
Login
default entity
team entity
service account🚧
CLI
sync16
<other commands>🚧7
Artifacts
Sweeps
Launch
Sweeps on Launch🚧8

Footnotes

  1. sync_tensorboard only uploads tfevent files to W&B, but metrics will not be logged in native W&B charts. 2 3

  2. define_metric only supports default summary.

  3. Only raw console logging is supported.

  4. Supported system metrics: CPU, Memory, Disk, Network, NVIDIA GPU, AMD GPU, Apple GPU.

  5. Unsupported settings: (anonymous, _flow_control*, _stats_open_metrics_endpoints, ...)

  6. The command is namespaced under wandb beta group.

  7. The rest of the CLI works, but uses the old backend under the hood for some commands.

  8. Sweeps on Launch requires verification.

Использование cookies

Мы используем файлы cookie в соответствии с Политикой конфиденциальности и Политикой использования cookies.

Нажимая кнопку «Принимаю», Вы даете АО «СберТех» согласие на обработку Ваших персональных данных в целях совершенствования нашего веб-сайта и Сервиса GitVerse, а также повышения удобства их использования.

Запретить использование cookies Вы можете самостоятельно в настройках Вашего браузера.