Sweeps
The tuning engine that automates hyperparameter search to find your best model faster.
Define only the search space and target metric, and grid, random, or Bayesian search automatically runs countless combinations and visualizes the results. Distributed parallel execution across machines and agents finds optimal hyperparameters efficiently.
What Is Sweeps
Hyperparameters — learning rate, batch size, regularization — heavily influence model performance, yet manual search is slow and unsystematic. W&B Sweeps automatically generates, runs, and compares combinations once you declare a search strategy and objective, finding the best configuration systematically.
Key Capabilities
Search strategies
Supports Grid, Random, and Bayesian optimization
Distributed parallel execution
Search concurrently across multiple machines and agents
Early termination
Cut costs by stopping unpromising runs early with Hyperband and similar methods
Importance analysis
Insights from parameter importance and correlation visualizations
What It's Used For
Maximizing model performance
Derive optimal hyperparameters through systematic search
Reducing search cost
Save GPU hours with early termination and distribution
Automating tuning
Turn repetitive manual tuning into a pipeline