Expand description
cuVS: Rust bindings for Vector Search on the GPU
This crate provides Rust bindings for cuVS, allowing you to run approximate nearest neighbors search on the GPU.
Modules§
- Brute Force KNN
- CAGRA is a graph-based nearest neighbors implementation with state-of-the art query performance for both small- and large-batch sized search.
- The IVF-Flat method is an ANN algorithm. It uses an inverted file index (IVF) with unmodified (that is, flat) vectors. This algorithm provides simple knobs to reduce the overall search space and to trade-off accuracy for speed.
- Inverted File Product Quantization
Structs§
- ManagedTensor is a wrapper around a dlpack DLManagedTensor object. This lets you pass matrices in device or host memory into cuvs.
- Resources are objects that are shared between function calls, and includes things like CUDA streams, cuBLAS handles and other resources that are expensive to create.