HpBandSter

Overview

HpBandSter is a distributed implementation of the Hyperband algorithm for hyperparameter optimization. It uses bandit-based early stopping to efficiently allocate resources to promising configurations, scaling across multiple workers.

Best for

Best for
Researchers and engineers running distributed hyperparameter optimization with limited compute budgets

Use cases

• Tuning hyperparameters for deep learning models
• Running distributed hyperparameter search on a cluster
• Accelerating model selection with early stopping

Notes

HpBandSter is a distributed implementation of the Hyperband algorithm for hyperparameter optimization. It uses bandit-based early stopping to efficiently allocate resources to promising configurations, scaling across multiple workers.

630 stars on GitHub. Last updated 2022-10-16. Licensed BSD-3-Clause.

Use cases

• Tuning hyperparameters for deep learning models
• Running distributed hyperparameter search on a cluster
• Accelerating model selection with early stopping

Pros

• Efficient resource allocation via Hyperband’s adaptive early stopping
• Distributed execution for scaling to large search spaces
• Open source with a straightforward Python interface

Cons

• Limited to Hyperband strategy, not a general-purpose tuner
• Requires manual setup of distributed workers and shared storage
• Less actively maintained compared to newer alternatives

Indexed from awesome-llmops and enriched against its public facts.