FLAML

Overview

FLAML is a lightweight Python library that finds accurate machine learning models efficiently and economically. It frees users from selecting learners and hyperparameters for each learner. FLAML is powered by a novel cost-effective hyperparameter optimization and learner selection method that is capable of handling large search spaces with heterogeneous evaluation cost while achieving strong performance.

Key Features

• Cost-Effective Tuning -- Novel search strategy that prioritizes low-cost trials, achieving strong results with minimal compute budget.
• Easy to Use -- Simple API requiring minimal code to get started. Works with scikit-learn, XGBoost, LightGBM, and more.
• Fast and Lightweight -- Designed to be fast with low overhead, suitable for both small and large-scale tasks.
• Flexible -- Supports custom learners, metrics, and search spaces. Extensible to various ML tasks including classification, regression, and time series forecasting.
• LLM Tuning -- Includes support for tuning large language model inference parameters for cost-effective generation.

Impact

FLAML has been adopted by thousands of users in both industry and academia. It has received over 4,000 GitHub stars and is integrated into the Microsoft ecosystem. The library demonstrates that AutoML can be both fast and accurate, lowering the barrier to effective machine learning for practitioners of all skill levels.

Key Publications

• FLAML: A Fast and Lightweight AutoML Library

  MLSys 2021

• Frugal Optimization for Cost-related Hyperparameters

  AAAI 2021

• EcoOptiGen: Cost-Effective Hyperparameter Optimization for LLM Generation Inference

  AutoML 2023