Weights and Biases

A CLI and library for interacting with the Weights and Biases API. Sign up for an account at wandb.ai

Features

• Keep a history of your weights and models from every training run
• Store all configuration parameters used in a training run
• Associate version control with your training runs
• Search and visualize training runs in a project
• Sync canonical models in your preferred format

Usage

CLI:

cd myproject
# Initialize a directory
wandb init
# Push files to W&B
wandb push bucket model.json weights.h5
# Sync training logs and push files when they change
./my_training.py | wandb bucket model.json weights.h5
# Manage configuration
wandb config set epochs=30

Client:

import wandb
conf = wandb.sync(["weights.h5", "model.json"], config={'existing': 'config'})

if conf.turbo:
    print("TURBO MODE!!!")

Detailed usage can be found in our documentation.