diff --git a/README.md b/README.md index 25ce059630..af2219fdd1 100755 --- a/README.md +++ b/README.md @@ -1,7 +1,7 @@ #
 RAFT: Reusable Accelerated Functions and Tools for Vector Search and More
> [!IMPORTANT] -> The vector search and clustering algorithms in RAFT are being migrated to a new library dedicated to vector search called [cuVS](https://github.com/rapidsai/cuvs). We will continue to support the vector search algorithms in RAFT during this move, but will no longer update them after the RAPIDS 24.06 (June) release. We plan to complete the migration by RAPIDS 24.08 (August) release. +> The vector search and clustering algorithms in RAFT are being migrated to a new library dedicated to vector search called [cuVS](https://github.com/rapidsai/cuvs). We will continue to support the vector search algorithms in RAFT during this move, but will no longer update them after the RAPIDS 24.06 (June) release. We plan to complete the migration by RAPIDS 24.10 (October) release and will be removing them altogether in the 24.12 (December) release. ![RAFT tech stack](img/raft-tech-stack-vss.png) @@ -36,7 +36,7 @@ ## What is RAFT? -RAFT contains fundamental widely-used algorithms and primitives for machine learning and information retrieval. The algorithms are CUDA-accelerated and form building blocks for more easily writing high performance applications. +RAFT contains fundamental widely-used algorithms and primitives for machine learning and data mining. The algorithms are CUDA-accelerated and form building blocks for more easily writing high performance applications. By taking a primitives-based approach to algorithm development, RAFT - accelerates algorithm construction time @@ -47,12 +47,10 @@ While not exhaustive, the following general categories help summarize the accele ##### | Category | Accelerated Functions in RAFT | |-----------------------|-----------------------------------------------------------------------------------------------------------------------------------| -| **Nearest Neighbors** | vector search, neighborhood graph construction, epsilon neighborhoods, pairwise distances | -| **Basic Clustering** | spectral clustering, hierarchical clustering, k-means | -| **Solvers** | combinatorial optimization, iterative solvers | | **Data Formats** | sparse & dense, conversions, data generation | | **Dense Operations** | linear algebra, matrix and vector operations, reductions, slicing, norms, factorization, least squares, svd & eigenvalue problems | | **Sparse Operations** | linear algebra, eigenvalue problems, slicing, norms, reductions, factorization, symmetrization, components & labeling | +| **Solvers** | combinatorial optimization, iterative solvers | | **Statistics** | sampling, moments and summary statistics, metrics, model evaluation | | **Tools & Utilities** | common tools and utilities for developing CUDA applications, multi-node multi-gpu infrastructure | @@ -67,42 +65,6 @@ In addition being a C++ library, RAFT also provides 2 Python libraries: ![RAFT is a C++ header-only template library with optional shared library and lightweight Python wrappers](img/arch.png) -## Use cases - -### Vector Similarity Search - -RAFT contains state-of-the-art implementations of approximate nearest neighbors search (ANNS) algorithms on the GPU, such as: - -* [Brute force](https://docs.rapids.ai/api/raft/nightly/pylibraft_api/neighbors/#brute-force). Performs a brute force nearest neighbors search without an index. -* [IVF-Flat](https://docs.rapids.ai/api/raft/nightly/pylibraft_api/neighbors/#ivf-flat) and [IVF-PQ](https://docs.rapids.ai/api/raft/nightly/pylibraft_api/neighbors/#ivf-pq). Use an inverted file index structure to map contents to their locations. IVF-PQ additionally uses product quantization to reduce the memory usage of vectors. These methods were originally popularized by the [FAISS](https://github.com/facebookresearch/faiss) library. -* [CAGRA](https://docs.rapids.ai/api/raft/nightly/pylibraft_api/neighbors/#cagra) (Cuda Anns GRAph-based). Uses a fast ANNS graph construction and search implementation optimized for the GPU. CAGRA outperforms state-of-the art CPU methods (i.e. HNSW) for large batch queries, single queries, and graph construction time. - -Projects that use the RAFT ANNS algorithms for accelerating vector search include: [Milvus](https://milvus.io/), [Redis](https://redis.io/), and [Faiss](https://github.com/facebookresearch/faiss). - -Please see the example [Jupyter notebook](https://github.com/rapidsai/raft/blob/HEAD/notebooks/VectorSearch_QuestionRetrieval.ipynb) to get started RAFT for vector search in Python. - - - -### Information Retrieval - -RAFT contains a catalog of reusable primitives for composing algorithms that require fast neighborhood computations, such as - -1. Computing distances between vectors and computing kernel gramm matrices -2. Performing ball radius queries for constructing epsilon neighborhoods -3. Clustering points to partition a space for smaller and faster searches -4. Constructing neighborhood "connectivities" graphs from dense vectors - -### Machine Learning - -RAFT's primitives are used in several RAPIDS libraries, including [cuML](https://github.com/rapidsai/cuml), [cuGraph](https://github.com/rapidsai/cugraph), and [cuOpt](https://github.com/rapidsai/cuopt) to build many end-to-end machine learning algorithms that span a large spectrum of different applications, including -- data generation -- model evaluation -- classification and regression -- clustering -- manifold learning -- dimensionality reduction. - -RAFT is also used by the popular collaborative filtering library [implicit](https://github.com/benfred/implicit) for recommender systems. ## Is RAFT right for me? @@ -327,70 +289,3 @@ When citing RAFT generally, please consider referencing this Github project. year={2022} } ``` -If citing the sparse pairwise distances API, please consider using the following bibtex: -```bibtex -@article{nolet2021semiring, - title={Semiring primitives for sparse neighborhood methods on the gpu}, - author={Nolet, Corey J and Gala, Divye and Raff, Edward and Eaton, Joe and Rees, Brad and Zedlewski, John and Oates, Tim}, - journal={arXiv preprint arXiv:2104.06357}, - year={2021} -} -``` - -If citing the single-linkage agglomerative clustering APIs, please consider the following bibtex: -```bibtex -@misc{nolet2023cuslink, - title={cuSLINK: Single-linkage Agglomerative Clustering on the GPU}, - author={Corey J. Nolet and Divye Gala and Alex Fender and Mahesh Doijade and Joe Eaton and Edward Raff and John Zedlewski and Brad Rees and Tim Oates}, - year={2023}, - eprint={2306.16354}, - archivePrefix={arXiv}, - primaryClass={cs.LG} -} -``` - -If citing CAGRA, please consider the following bibtex: -```bibtex -@misc{ootomo2023cagra, - title={CAGRA: Highly Parallel Graph Construction and Approximate Nearest Neighbor Search for GPUs}, - author={Hiroyuki Ootomo and Akira Naruse and Corey Nolet and Ray Wang and Tamas Feher and Yong Wang}, - year={2024}, - series = {ICDE '24} -} -``` - -If citing the k-selection routines, please consider the following bibtex: - -```bibtex -@proceedings{10.1145/3581784, - title = {Parallel Top-K Algorithms on GPU: A Comprehensive Study and New Methods}, - author={Jingrong Zhang, Akira Naruse, Xipeng Li, and Yong Wang}, - year = {2023}, - isbn = {9798400701092}, - publisher = {Association for Computing Machinery}, - address = {New York, NY, USA}, - location = {Denver, CO, USA}, - series = {SC '23} -} -``` - -If citing the nearest neighbors descent API, please consider the following bibtex: -```bibtex -@inproceedings{10.1145/3459637.3482344, - author = {Wang, Hui and Zhao, Wan-Lei and Zeng, Xiangxiang and Yang, Jianye}, - title = {Fast K-NN Graph Construction by GPU Based NN-Descent}, - year = {2021}, - isbn = {9781450384469}, - publisher = {Association for Computing Machinery}, - address = {New York, NY, USA}, - url = {https://doi.org/10.1145/3459637.3482344}, - doi = {10.1145/3459637.3482344}, - abstract = {NN-Descent is a classic k-NN graph construction approach. It is still widely employed in machine learning, computer vision, and information retrieval tasks due to its efficiency and genericness. However, the current design only works well on CPU. In this paper, NN-Descent has been redesigned to adapt to the GPU architecture. A new graph update strategy called selective update is proposed. It reduces the data exchange between GPU cores and GPU global memory significantly, which is the processing bottleneck under GPU computation architecture. This redesign leads to full exploitation of the parallelism of the GPU hardware. In the meantime, the genericness, as well as the simplicity of NN-Descent, are well-preserved. Moreover, a procedure that allows to k-NN graph to be merged efficiently on GPU is proposed. It makes the construction of high-quality k-NN graphs for out-of-GPU-memory datasets tractable. Our approach is 100-250\texttimes{} faster than the single-thread NN-Descent and is 2.5-5\texttimes{} faster than the existing GPU-based approaches as we tested on million as well as billion scale datasets.}, - booktitle = {Proceedings of the 30th ACM International Conference on Information \& Knowledge Management}, - pages = {1929–1938}, - numpages = {10}, - keywords = {high-dimensional, nn-descent, gpu, k-nearest neighbor graph}, - location = {Virtual Event, Queensland, Australia}, - series = {CIKM '21} -} -``` diff --git a/cpp/include/raft/cluster/kmeans.cuh b/cpp/include/raft/cluster/kmeans.cuh index eb28cc1626..38318e8ec8 100644 --- a/cpp/include/raft/cluster/kmeans.cuh +++ b/cpp/include/raft/cluster/kmeans.cuh @@ -86,13 +86,14 @@ using KeyValueIndexOp = detail::KeyValueIndexOp; * @param[out] n_iter Number of iterations run. */ template -void fit(raft::resources const& handle, - const KMeansParams& params, - raft::device_matrix_view X, - std::optional> sample_weight, - raft::device_matrix_view centroids, - raft::host_scalar_view inertia, - raft::host_scalar_view n_iter) +[[deprecated("Use cuVS instead")]] void fit( + raft::resources const& handle, + const KMeansParams& params, + raft::device_matrix_view X, + std::optional> sample_weight, + raft::device_matrix_view centroids, + raft::host_scalar_view inertia, + raft::host_scalar_view n_iter) { detail::kmeans_fit(handle, params, X, sample_weight, centroids, inertia, n_iter); } @@ -150,14 +151,15 @@ void fit(raft::resources const& handle, * their closest cluster center. */ template -void predict(raft::resources const& handle, - const KMeansParams& params, - raft::device_matrix_view X, - std::optional> sample_weight, - raft::device_matrix_view centroids, - raft::device_vector_view labels, - bool normalize_weight, - raft::host_scalar_view inertia) +[[deprecated("Use cuVS instead")]] void predict( + raft::resources const& handle, + const KMeansParams& params, + raft::device_matrix_view X, + std::optional> sample_weight, + raft::device_matrix_view centroids, + raft::device_vector_view labels, + bool normalize_weight, + raft::host_scalar_view inertia) { detail::kmeans_predict( handle, params, X, sample_weight, centroids, labels, normalize_weight, inertia); @@ -213,14 +215,15 @@ void predict(raft::resources const& handle, * @param[out] n_iter Number of iterations run. */ template -void fit_predict(raft::resources const& handle, - const KMeansParams& params, - raft::device_matrix_view X, - std::optional> sample_weight, - std::optional> centroids, - raft::device_vector_view labels, - raft::host_scalar_view inertia, - raft::host_scalar_view n_iter) +[[deprecated("Use cuVS instead")]] void fit_predict( + raft::resources const& handle, + const KMeansParams& params, + raft::device_matrix_view X, + std::optional> sample_weight, + std::optional> centroids, + raft::device_vector_view labels, + raft::host_scalar_view inertia, + raft::host_scalar_view n_iter) { detail::kmeans_fit_predict( handle, params, X, sample_weight, centroids, labels, inertia, n_iter); @@ -252,13 +255,13 @@ void transform(raft::resources const& handle, } template -void transform(raft::resources const& handle, - const KMeansParams& params, - const DataT* X, - const DataT* centroids, - IndexT n_samples, - IndexT n_features, - DataT* X_new) +[[deprecated("Use cuVS instead")]] void transform(raft::resources const& handle, + const KMeansParams& params, + const DataT* X, + const DataT* centroids, + IndexT n_samples, + IndexT n_features, + DataT* X_new) { detail::kmeans_transform( handle, params, X, centroids, n_samples, n_features, X_new); diff --git a/cpp/include/raft/cluster/kmeans_balanced.cuh b/cpp/include/raft/cluster/kmeans_balanced.cuh index a1a182608b..7479047fce 100644 --- a/cpp/include/raft/cluster/kmeans_balanced.cuh +++ b/cpp/include/raft/cluster/kmeans_balanced.cuh @@ -73,11 +73,11 @@ namespace raft::cluster::kmeans_balanced { * datatype. If DataT == MathT, this must be the identity. */ template -void fit(const raft::resources& handle, - kmeans_balanced_params const& params, - raft::device_matrix_view X, - raft::device_matrix_view centroids, - MappingOpT mapping_op = raft::identity_op()) +[[deprecated("Use cuVS instead")]] void fit(const raft::resources& handle, + kmeans_balanced_params const& params, + raft::device_matrix_view X, + raft::device_matrix_view centroids, + MappingOpT mapping_op = raft::identity_op()) { RAFT_EXPECTS(X.extent(1) == centroids.extent(1), "Number of features in dataset and centroids are different"); @@ -131,12 +131,13 @@ template -void predict(const raft::resources& handle, - kmeans_balanced_params const& params, - raft::device_matrix_view X, - raft::device_matrix_view centroids, - raft::device_vector_view labels, - MappingOpT mapping_op = raft::identity_op()) +[[deprecated("Use cuVS instead")]] void predict( + const raft::resources& handle, + kmeans_balanced_params const& params, + raft::device_matrix_view X, + raft::device_matrix_view centroids, + raft::device_vector_view labels, + MappingOpT mapping_op = raft::identity_op()) { RAFT_EXPECTS(X.extent(0) == labels.extent(0), "Number of rows in dataset and labels are different"); @@ -196,12 +197,13 @@ template -void fit_predict(const raft::resources& handle, - kmeans_balanced_params const& params, - raft::device_matrix_view X, - raft::device_matrix_view centroids, - raft::device_vector_view labels, - MappingOpT mapping_op = raft::identity_op()) +[[deprecated("Use cuVS instead")]] void fit_predict( + const raft::resources& handle, + kmeans_balanced_params const& params, + raft::device_matrix_view X, + raft::device_matrix_view centroids, + raft::device_vector_view labels, + MappingOpT mapping_op = raft::identity_op()) { auto centroids_const = raft::make_device_matrix_view( centroids.data_handle(), centroids.extent(0), centroids.extent(1)); @@ -255,14 +257,15 @@ template -void build_clusters(const raft::resources& handle, - const kmeans_balanced_params& params, - raft::device_matrix_view X, - raft::device_matrix_view centroids, - raft::device_vector_view labels, - raft::device_vector_view cluster_sizes, - MappingOpT mapping_op = raft::identity_op(), - std::optional> X_norm = std::nullopt) +[[deprecated("Use cuVS instead")]] void build_clusters( + const raft::resources& handle, + const kmeans_balanced_params& params, + raft::device_matrix_view X, + raft::device_matrix_view centroids, + raft::device_vector_view labels, + raft::device_vector_view cluster_sizes, + MappingOpT mapping_op = raft::identity_op(), + std::optional> X_norm = std::nullopt) { RAFT_EXPECTS(X.extent(0) == labels.extent(0), "Number of rows in dataset and labels are different"); @@ -334,13 +337,14 @@ template -void calc_centers_and_sizes(const raft::resources& handle, - raft::device_matrix_view X, - raft::device_vector_view labels, - raft::device_matrix_view centroids, - raft::device_vector_view cluster_sizes, - bool reset_counters = true, - MappingOpT mapping_op = raft::identity_op()) +[[deprecated("Use cuVS instead")]] void calc_centers_and_sizes( + const raft::resources& handle, + raft::device_matrix_view X, + raft::device_vector_view labels, + raft::device_matrix_view centroids, + raft::device_vector_view cluster_sizes, + bool reset_counters = true, + MappingOpT mapping_op = raft::identity_op()) { RAFT_EXPECTS(X.extent(0) == labels.extent(0), "Number of rows in dataset and labels are different"); diff --git a/cpp/include/raft/cluster/single_linkage.cuh b/cpp/include/raft/cluster/single_linkage.cuh index d9eba6edc5..067445c542 100644 --- a/cpp/include/raft/cluster/single_linkage.cuh +++ b/cpp/include/raft/cluster/single_linkage.cuh @@ -1,5 +1,5 @@ /* - * Copyright (c) 2021-2023, NVIDIA CORPORATION. + * Copyright (c) 2021-2024, NVIDIA CORPORATION. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -50,14 +50,14 @@ namespace raft::cluster { template -void single_linkage(raft::resources const& handle, - const value_t* X, - size_t m, - size_t n, - raft::distance::DistanceType metric, - linkage_output* out, - int c, - size_t n_clusters) +[[deprecated("Use cuVS instead")]] void single_linkage(raft::resources const& handle, + const value_t* X, + size_t m, + size_t n, + raft::distance::DistanceType metric, + linkage_output* out, + int c, + size_t n_clusters) { detail::single_linkage( handle, X, m, n, metric, out, c, n_clusters); @@ -87,13 +87,14 @@ constexpr int DEFAULT_CONST_C = 15; control of k. The algorithm will set `k = log(n) + c` */ template -void single_linkage(raft::resources const& handle, - raft::device_matrix_view X, - raft::device_matrix_view dendrogram, - raft::device_vector_view labels, - raft::distance::DistanceType metric, - size_t n_clusters, - std::optional c = std::make_optional(DEFAULT_CONST_C)) +[[deprecated("Use cuVS instead")]] void single_linkage( + raft::resources const& handle, + raft::device_matrix_view X, + raft::device_matrix_view dendrogram, + raft::device_vector_view labels, + raft::distance::DistanceType metric, + size_t n_clusters, + std::optional c = std::make_optional(DEFAULT_CONST_C)) { linkage_output out_arrs; out_arrs.children = dendrogram.data_handle(); diff --git a/cpp/include/raft/distance/distance-ext.cuh b/cpp/include/raft/distance/distance-ext.cuh index 2d41e029fe..dcbfbfdbc3 100644 --- a/cpp/include/raft/distance/distance-ext.cuh +++ b/cpp/include/raft/distance/distance-ext.cuh @@ -35,42 +35,43 @@ template -void distance(raft::resources const& handle, - const DataT* x, - const DataT* y, - OutT* dist, - IdxT m, - IdxT n, - IdxT k, - void* workspace, - size_t worksize, - FinalLambda fin_op, - bool isRowMajor = true, - DataT metric_arg = 2.0f) RAFT_EXPLICIT; +[[deprecated("Use cuVS instead")]] void distance(raft::resources const& handle, + const DataT* x, + const DataT* y, + OutT* dist, + IdxT m, + IdxT n, + IdxT k, + void* workspace, + size_t worksize, + FinalLambda fin_op, + bool isRowMajor = true, + DataT metric_arg = 2.0f) RAFT_EXPLICIT; template -void distance(raft::resources const& handle, - const DataT* x, - const DataT* y, - OutT* dist, - IdxT m, - IdxT n, - IdxT k, - void* workspace, - size_t worksize, - bool isRowMajor = true, - DataT metric_arg = 2.0f) RAFT_EXPLICIT; +[[deprecated("Use cuVS instead")]] void distance(raft::resources const& handle, + const DataT* x, + const DataT* y, + OutT* dist, + IdxT m, + IdxT n, + IdxT k, + void* workspace, + size_t worksize, + bool isRowMajor = true, + DataT metric_arg = 2.0f) RAFT_EXPLICIT; template -size_t getWorkspaceSize(const DataT* x, const DataT* y, IdxT m, IdxT n, IdxT k) RAFT_EXPLICIT; +[[deprecated("Use cuVS instead")]] size_t getWorkspaceSize( + const DataT* x, const DataT* y, IdxT m, IdxT n, IdxT k) RAFT_EXPLICIT; template -void distance(raft::resources const& handle, - const DataT* x, - const DataT* y, - OutT* dist, - IdxT m, - IdxT n, - IdxT k, - bool isRowMajor = true, - DataT metric_arg = 2.0f) RAFT_EXPLICIT; +[[deprecated("Use cuVS instead")]] void distance(raft::resources const& handle, + const DataT* x, + const DataT* y, + OutT* dist, + IdxT m, + IdxT n, + IdxT k, + bool isRowMajor = true, + DataT metric_arg = 2.0f) RAFT_EXPLICIT; template -void pairwise_distance(raft::resources const& handle, - const Type* x, - const Type* y, - Type* dist, - IdxT m, - IdxT n, - IdxT k, - rmm::device_uvector& workspace, - raft::distance::DistanceType metric, - bool isRowMajor = true, - Type metric_arg = 2.0f) RAFT_EXPLICIT; +[[deprecated("Use cuVS instead")]] void pairwise_distance(raft::resources const& handle, + const Type* x, + const Type* y, + Type* dist, + IdxT m, + IdxT n, + IdxT k, + rmm::device_uvector& workspace, + raft::distance::DistanceType metric, + bool isRowMajor = true, + Type metric_arg = 2.0f) RAFT_EXPLICIT; template -void pairwise_distance(raft::resources const& handle, - const Type* x, - const Type* y, - Type* dist, - IdxT m, - IdxT n, - IdxT k, - raft::distance::DistanceType metric, - bool isRowMajor = true, - Type metric_arg = 2.0f) RAFT_EXPLICIT; +[[deprecated("Use cuVS instead")]] void pairwise_distance(raft::resources const& handle, + const Type* x, + const Type* y, + Type* dist, + IdxT m, + IdxT n, + IdxT k, + raft::distance::DistanceType metric, + bool isRowMajor = true, + Type metric_arg = 2.0f) RAFT_EXPLICIT; template -void distance(raft::resources const& handle, - raft::device_matrix_view const x, - raft::device_matrix_view const y, - raft::device_matrix_view dist, - DataT metric_arg = 2.0f) RAFT_EXPLICIT; +[[deprecated("Use cuVS instead")]] void distance( + raft::resources const& handle, + raft::device_matrix_view const x, + raft::device_matrix_view const y, + raft::device_matrix_view dist, + DataT metric_arg = 2.0f) RAFT_EXPLICIT; template -void pairwise_distance(raft::resources const& handle, - device_matrix_view const x, - device_matrix_view const y, - device_matrix_view dist, - raft::distance::DistanceType metric, - Type metric_arg = 2.0f) RAFT_EXPLICIT; +[[deprecated("Use cuVS instead")]] void pairwise_distance( + raft::resources const& handle, + device_matrix_view const x, + device_matrix_view const y, + device_matrix_view dist, + raft::distance::DistanceType metric, + Type metric_arg = 2.0f) RAFT_EXPLICIT; }; // namespace distance }; // namespace raft diff --git a/cpp/include/raft/neighbors/ball_cover.cuh b/cpp/include/raft/neighbors/ball_cover.cuh index 20c88f3318..09938020b9 100644 --- a/cpp/include/raft/neighbors/ball_cover.cuh +++ b/cpp/include/raft/neighbors/ball_cover.cuh @@ -13,6 +13,7 @@ * See the License for the specific language governing permissions and * limitations under the License. */ + #pragma once #ifndef RAFT_EXPLICIT_INSTANTIATE_ONLY #include "ball_cover-inl.cuh" diff --git a/cpp/include/raft/neighbors/brute_force_types.hpp b/cpp/include/raft/neighbors/brute_force_types.hpp index a8f073edc6..4511f8d8ba 100644 --- a/cpp/include/raft/neighbors/brute_force_types.hpp +++ b/cpp/include/raft/neighbors/brute_force_types.hpp @@ -94,12 +94,14 @@ struct index : ann::index { * the dataset. If the dataset is in host memory, it will be copied to the device and the * index will own the device memory. */ + template - index(raft::resources const& res, - mdspan, row_major, data_accessor> dataset, - std::optional>&& norms, - raft::distance::DistanceType metric, - T metric_arg = 0.0) + [[deprecated("Use cuVS instead")]] index( + raft::resources const& res, + mdspan, row_major, data_accessor> dataset, + std::optional>&& norms, + raft::distance::DistanceType metric, + T metric_arg = 0.0) : ann::index(), metric_(metric), dataset_(make_device_matrix(res, 0, 0)), @@ -116,11 +118,12 @@ struct index : ann::index { * This class stores a non-owning reference to the dataset and norms here. * Having precomputed norms gives us a performance advantage at query time. */ - index(raft::resources const& res, - raft::device_matrix_view dataset_view, - std::optional> norms_view, - raft::distance::DistanceType metric, - T metric_arg = 0.0) + [[deprecated("Use cuVS instead")]] index( + raft::resources const& res, + raft::device_matrix_view dataset_view, + std::optional> norms_view, + raft::distance::DistanceType metric, + T metric_arg = 0.0) : ann::index(), metric_(metric), dataset_(make_device_matrix(res, 0, 0)), @@ -131,10 +134,11 @@ struct index : ann::index { } template - index(raft::resources const& res, - index_params const& params, - mdspan, row_major, data_accessor> dataset, - std::optional>&& norms = std::nullopt) + [[deprecated("Use cuVS instead")]] index( + raft::resources const& res, + index_params const& params, + mdspan, row_major, data_accessor> dataset, + std::optional>&& norms = std::nullopt) : ann::index(), metric_(params.metric), dataset_(make_device_matrix(res, 0, 0)), diff --git a/cpp/include/raft/neighbors/cagra_types.hpp b/cpp/include/raft/neighbors/cagra_types.hpp index 97c9c0d098..bc7c380db1 100644 --- a/cpp/include/raft/neighbors/cagra_types.hpp +++ b/cpp/include/raft/neighbors/cagra_types.hpp @@ -201,8 +201,9 @@ struct index : ann::index { ~index() = default; /** Construct an empty index. */ - index(raft::resources const& res, - raft::distance::DistanceType metric = raft::distance::DistanceType::L2Expanded) + [[deprecated("Use cuVS instead")]] index( + raft::resources const& res, + raft::distance::DistanceType metric = raft::distance::DistanceType::L2Expanded) : ann::index(), metric_(metric), graph_(make_device_matrix(res, 0, 0)), @@ -265,10 +266,11 @@ struct index : ann::index { * */ template - index(raft::resources const& res, - raft::distance::DistanceType metric, - mdspan, row_major, data_accessor> dataset, - mdspan, row_major, graph_accessor> knn_graph) + [[deprecated("Use cuVS instead")]] index( + raft::resources const& res, + raft::distance::DistanceType metric, + mdspan, row_major, data_accessor> dataset, + mdspan, row_major, graph_accessor> knn_graph) : ann::index(), metric_(metric), graph_(make_device_matrix(res, 0, 0)), diff --git a/cpp/include/raft/neighbors/hnsw_types.hpp b/cpp/include/raft/neighbors/hnsw_types.hpp index f90de6f01b..f78571f491 100644 --- a/cpp/include/raft/neighbors/hnsw_types.hpp +++ b/cpp/include/raft/neighbors/hnsw_types.hpp @@ -38,7 +38,6 @@ struct search_params : ann::search_params { int num_threads = 0; // number of host threads to use for concurrent searches. Value of 0 // automatically maximizes parallelism }; - template struct index : ann::index { public: @@ -51,7 +50,10 @@ struct index : ann::index { * @param[in] dim dimensions of the training dataset * @param[in] metric distance metric to search. Supported metrics ("L2Expanded", "InnerProduct") */ - index(int dim, raft::distance::DistanceType metric) : dim_{dim}, metric_{metric} {} + [[deprecated("Use cuVS instead")]] index(int dim, raft::distance::DistanceType metric) + : dim_{dim}, metric_{metric} + { + } /** @brief Get underlying index diff --git a/cpp/include/raft/neighbors/ivf_flat_types.hpp b/cpp/include/raft/neighbors/ivf_flat_types.hpp index 7605bd82a3..2cafceb512 100644 --- a/cpp/include/raft/neighbors/ivf_flat_types.hpp +++ b/cpp/include/raft/neighbors/ivf_flat_types.hpp @@ -261,12 +261,12 @@ struct index : ann::index { ~index() = default; /** Construct an empty index. It needs to be trained and then populated. */ - index(raft::resources const& res, - raft::distance::DistanceType metric, - uint32_t n_lists, - bool adaptive_centers, - bool conservative_memory_allocation, - uint32_t dim) + [[deprecated("Use cuVS instead")]] index(raft::resources const& res, + raft::distance::DistanceType metric, + uint32_t n_lists, + bool adaptive_centers, + bool conservative_memory_allocation, + uint32_t dim) : ann::index(), veclen_(calculate_veclen(dim)), metric_(metric), @@ -285,7 +285,9 @@ struct index : ann::index { } /** Construct an empty index. It needs to be trained and then populated. */ - index(raft::resources const& res, const index_params& params, uint32_t dim) + [[deprecated("Use cuVS instead")]] index(raft::resources const& res, + const index_params& params, + uint32_t dim) : index(res, params.metric, params.n_lists, diff --git a/cpp/include/raft/neighbors/ivf_pq_types.hpp b/cpp/include/raft/neighbors/ivf_pq_types.hpp index 3ee350c6fb..d5906d621c 100644 --- a/cpp/include/raft/neighbors/ivf_pq_types.hpp +++ b/cpp/include/raft/neighbors/ivf_pq_types.hpp @@ -361,14 +361,14 @@ struct index : ann::index { ~index() = default; /** Construct an empty index. It needs to be trained and then populated. */ - index(raft::resources const& handle, - raft::distance::DistanceType metric, - codebook_gen codebook_kind, - uint32_t n_lists, - uint32_t dim, - uint32_t pq_bits = 8, - uint32_t pq_dim = 0, - bool conservative_memory_allocation = false) + [[deprecated("Use cuVS instead")]] index(raft::resources const& handle, + raft::distance::DistanceType metric, + codebook_gen codebook_kind, + uint32_t n_lists, + uint32_t dim, + uint32_t pq_bits = 8, + uint32_t pq_dim = 0, + bool conservative_memory_allocation = false) : ann::index(), metric_(metric), codebook_kind_(codebook_kind), @@ -391,7 +391,9 @@ struct index : ann::index { } /** Construct an empty index. It needs to be trained and then populated. */ - index(raft::resources const& handle, const index_params& params, uint32_t dim) + [[deprecated("Use cuVS instead")]] index(raft::resources const& handle, + const index_params& params, + uint32_t dim) : index(handle, params.metric, params.codebook_kind, diff --git a/cpp/include/raft/neighbors/nn_descent_types.hpp b/cpp/include/raft/neighbors/nn_descent_types.hpp index eb01a423be..9decf47f39 100644 --- a/cpp/include/raft/neighbors/nn_descent_types.hpp +++ b/cpp/include/raft/neighbors/nn_descent_types.hpp @@ -101,7 +101,10 @@ struct index : ann::index { * @param n_cols number of cols in knn-graph * @param return_distances whether to allocate and get distances information */ - index(raft::resources const& res, int64_t n_rows, int64_t n_cols, bool return_distances = false) + [[deprecated("Use cuVS instead")]] index(raft::resources const& res, + int64_t n_rows, + int64_t n_cols, + bool return_distances = false) : ann::index(), res_{res}, metric_{raft::distance::DistanceType::L2Expanded}, @@ -128,11 +131,12 @@ struct index : ann::index { * storing knn-graph distances * @param return_distances whether to allocate and get distances information */ - index(raft::resources const& res, - raft::host_matrix_view graph_view, - std::optional> distances_view = - std::nullopt, - bool return_distances = false) + [[deprecated("Use cuVS instead")]] index( + raft::resources const& res, + raft::host_matrix_view graph_view, + std::optional> distances_view = + std::nullopt, + bool return_distances = false) : ann::index(), res_{res}, metric_{raft::distance::DistanceType::L2Expanded}, diff --git a/cpp/include/raft/neighbors/refine-ext.cuh b/cpp/include/raft/neighbors/refine-ext.cuh index 7948a0e4f2..216e1b9ab5 100644 --- a/cpp/include/raft/neighbors/refine-ext.cuh +++ b/cpp/include/raft/neighbors/refine-ext.cuh @@ -29,24 +29,24 @@ namespace raft::neighbors { template -void refine(raft::resources const& handle, - raft::device_matrix_view dataset, - raft::device_matrix_view queries, - raft::device_matrix_view neighbor_candidates, - raft::device_matrix_view indices, - raft::device_matrix_view distances, - raft::distance::DistanceType metric = distance::DistanceType::L2Unexpanded) - RAFT_EXPLICIT; +[[deprecated("Use cuVS instead")]] void refine( + raft::resources const& handle, + raft::device_matrix_view dataset, + raft::device_matrix_view queries, + raft::device_matrix_view neighbor_candidates, + raft::device_matrix_view indices, + raft::device_matrix_view distances, + raft::distance::DistanceType metric = distance::DistanceType::L2Unexpanded) RAFT_EXPLICIT; template -void refine(raft::resources const& handle, - raft::host_matrix_view dataset, - raft::host_matrix_view queries, - raft::host_matrix_view neighbor_candidates, - raft::host_matrix_view indices, - raft::host_matrix_view distances, - raft::distance::DistanceType metric = distance::DistanceType::L2Unexpanded) - RAFT_EXPLICIT; +[[deprecated("Use cuVS instead")]] void refine( + raft::resources const& handle, + raft::host_matrix_view dataset, + raft::host_matrix_view queries, + raft::host_matrix_view neighbor_candidates, + raft::host_matrix_view indices, + raft::host_matrix_view distances, + raft::distance::DistanceType metric = distance::DistanceType::L2Unexpanded) RAFT_EXPLICIT; } // namespace raft::neighbors diff --git a/docs/source/conf.py b/docs/source/conf.py index 8b2040baa2..7a287b689f 100644 --- a/docs/source/conf.py +++ b/docs/source/conf.py @@ -71,7 +71,7 @@ .. attention:: - The vector search and clustering algorithms in RAFT are being migrated to a new library dedicated to vector search called `cuVS `_. We will continue to support the vector search algorithms in RAFT during this move, but will no longer update them after the RAPIDS 24.06 (June) release. We plan to complete the migration by RAPIDS 24.08 (August) release. + The vector search and clustering algorithms in RAFT are being migrated to a new library dedicated to vector search called `cuVS `_. We will continue to support the vector search algorithms in RAFT during this move, but will no longer update them after the RAPIDS 24.06 (June) release. We plan to complete the migration by RAPIDS 24.10 (October) release and they will be removed from RAFT altogether in the 24.12 (December) release. """ diff --git a/docs/source/raft_ann_benchmarks.md b/docs/source/raft_ann_benchmarks.md index fc11a56ac8..7bac2047fc 100644 --- a/docs/source/raft_ann_benchmarks.md +++ b/docs/source/raft_ann_benchmarks.md @@ -2,6 +2,10 @@ This project provides a benchmark program for various ANN search implementations. It's especially suitable for comparing GPU implementations as well as comparing GPU against CPU. +> [!IMPORTANT] +> The vector search and clustering algorithms in RAFT are being migrated to a new library dedicated to vector search called [cuVS](https://github.com/rapidsai/cuvs). As a result, `raft-ann-bench` is being migrated to `cuvs-bench` and will be removed from RAFT altogether in the 24.12 (December) release. + + ## Table of Contents - [Installing the benchmarks](#installing-the-benchmarks)