mamba-rs 0.7.3

Mamba SSM and Mamba-3 SISO in Rust with optional CUDA acceleration: inference and training (BPTT through the SSM state, AdamW) on CPU and GPU, custom NVRTC-compiled kernels, CUDA Graph capture, f32 / bf16 / f16 storage, deterministic batch-invariant GEMMs by default with explicit cuBLAS Fast and Pedantic modes.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14

//! Compatibility shim for the legacy `mamba3_gpu` module path.
//!
//! The 2313-line single-file `mamba3_gpu.rs` was split into:
//!   - [`super::state`]    — dimensions, saved acts, scratch, target scratch
//!   - [`super::forward`]  — gpu_forward_mamba3_layer / _backbone / _target_burnin
//!   - [`super::backward`] — gpu_backward_mamba3_layer / _backbone
//!
//! All public types and functions are re-exported here so existing
//! `use mamba3_siso::gpu::mamba3_gpu::Foo` paths continue to work.
//! New code should import directly from the split modules.

pub use super::backward::*;
pub use super::forward::*;
pub use super::state::*;