Crate oxicuda_bayes 

oxicuda-bayes — Bayesian deep learning primitives for OxiCUDA.

Pure-Rust implementation of variational inference and Bayesian neural network building blocks suitable for CPU simulation and PTX kernel generation for GPU execution.

Architecture

oxicuda-bayes
├── layers/         — BayesLinear, BayesConv2d, Flipout layers
├── variational/    — ELBO, normalizing flows, mean-field, reparameterization
├── calibration/    — Temperature scaling, ECE/MCE/ACE, isotonic, Platt, Brier, NLL
├── uncertainty/    — MC Dropout, Deep Ensembles, SWAG, last-layer Laplace, BALD
├── error           — BayesError / BayesResult
├── handle          — BayesHandle (SmVersion + LcgRng)
└── ptx_kernels     — GPU PTX kernel strings

Modules

calibration

Confidence calibration of probabilistic classifier outputs.

error

Error types for oxicuda-bayes.

handle

Session handle for oxicuda-bayes.

layers

Bayesian layer implementations.

prelude

Convenience re-exports for common Bayesian deep learning types.

ptx_kernels

PTX GPU kernel sources for Bayesian deep learning operations.

uncertainty

Predictive uncertainty quantification for deep models.

variational

Variational inference primitives.