Expand description
oxicuda-recsys — Recommender system primitives for OxiCUDA.
Pure-Rust implementation of collaborative filtering, neural recommendation models, and graph-based recommenders, suitable for CPU simulation and PTX kernel generation for GPU execution.
§Architecture
oxicuda-recsys
├── factorization/ — ALS, BPR, NMF matrix factorization
├── ncf/ — Neural Collaborative Filtering
├── two_tower/ — Two-Tower retrieval model
├── deepfm/ — DeepFM and Wide & Deep models
├── sequential/ — SASRec self-attention sequential model
├── graph_recsys/ — LightGCN graph-based recommendation
├── multitask/ — MMoE / PLE multi-task learning
├── sampling/ — Uniform negative sampling
├── metrics/ — NDCG@k, Precision@k ranking metrics
├── handle — LcgRng (deterministic PRNG)
├── error — RecSysError / RecSysResult
└── ptx_kernels — GPU PTX kernel strings (7 kernels × 6 SM versions)Re-exports§
pub use crate::deepfm::dcn::CrossKind;pub use crate::deepfm::dcn::Dcn;pub use crate::deepfm::dcn::DcnConfig;pub use crate::dlrm::Dlrm;pub use crate::dlrm::DlrmConfig;pub use crate::factorization::ffm::Ffm;pub use crate::factorization::ffm::FfmConfig;pub use crate::factorization::ffm::FfmEntry;pub use crate::factorization::fism::Fism;pub use crate::factorization::fism::FismConfig;pub use crate::factorization::ials::Ials;pub use crate::factorization::ials::IalsConfig;pub use crate::fibinet::BilinearType;pub use crate::fibinet::Fibinet;pub use crate::fibinet::FibinetConfig;pub use crate::graph_recsys::graphrec::GraphRec;pub use crate::ranking::fairness_ranking::FairnessRanker;pub use crate::ranking::fairness_ranking::FairnessRankerConfig;pub use crate::sequential::cl4srec::Cl4sRec;pub use crate::sequential::cl4srec::Cl4sRecConfig;
Modules§
- deepfm
- dlrm
- DLRM — Deep Learning Recommendation Model (Naumov et al. 2019).
- error
- factorization
- fibinet
- FiBiNET — Feature Importance and Bilinear feature interaction NETwork.
- graph_
recsys - handle
- metrics
- multitask
- ncf
- ptx_
kernels - ranking
- Fairness-aware ranking and exposure-allocation utilities.
- sampling
- sequential
- two_
tower