LightGBM documentation

LightGBM documentation#

2025-12-15

1 min read time

Applies to Linux

Use LightGBM on ROCm for GPU-accelerated gradient boosting on AMD Instinct GPUs, enabling scalable performance for financial modeling tasks with large, imbalanced, and categorical datasets.

LightGBM uses histogram‑based splitting and leaf‑wise tree growth to achieve rapid convergence on datasets with many categorical features (such as employment type or verification status) without requiring one‑hot encoding. It maintains a low memory footprint and handles sparsity effectively, making it well‑suited to credit risk and scoring workloads. GPU acceleration can provide 2x to 5x performance gains through optimized histogram computation; however, care is needed to avoid overfitting without parameters such as min_data_in_leaf.

ROCm enablement accelerates LightGBM on AMD Instinct GPUs with optimized kernels and efficient memory handling, delivering meaningful performance gains over CPU‑only baselines, especially for finance pipelines rich in categorical data.

LightGBM is part of the AMD ROCm Finance toolkit.

The ROCm-Finance LightGBM source code is hosted on GitHub at ROCm/LightGBM.

ROCm-Finance LightGBM documentation is organized into the following categories: