RAPIDS RAFT (Reusable Accelerated Functions and Tools) is a library that provides fundamental, widely-used algorithms and primitives for machine learning and information retrieval. These algorithms are CUDA-accelerated, forming building blocks for high-performance applications. RAFT's primitives-based approach accelerates algorithm construction, reduces maintenance burdens, and centralizes core computations for future optimizations.

Features

• Data Formats: Supports sparse & dense data, conversions, and data generation.
• Dense Operations: Includes linear algebra, matrix and vector operations, slicing, norms, factorization, least squares, SVD, and eigenvalue problems.
• Sparse Operations: Offers linear algebra, eigenvalue problems, slicing, norms, reductions, factorization, symmetrization, components, and labeling.
• Solvers: Provides combinatorial optimization and iterative solvers.
• Statistics: Features sampling, moments and summary statistics, and metrics.
• Tools & Utilities: Common utilities for developing CUDA applications and multi-node multi-GPU infrastructure.

Benefits

• High Performance: CUDA-accelerated algorithms ensure high performance for machine learning and information retrieval tasks.
• Efficiency: Primitives-based approach reduces algorithm construction time and maintenance burden.
• Optimization: Centralized core computations allow future optimizations to benefit all algorithms using them.