Mamba SSM and Mamba-3 SISO in Rust with optional CUDA GPU acceleration. Inference and training (BPTT through SSM state, AdamW), CPU + GPU paths, custom CUDA kernels, CUDA Graph capture, f32 / bf16 / f16. Batch-invariant bf16 inference — per-row output is bit-identical across batch sizes.
A transpiler which converts Mamba files to Python 3 files
State Space Model primitives for OxiCUDA: S4 (HiPPO-LegS / DPLR), Mamba selective scan (S6), Mamba-2 (SSD), and RWKV time-mixing — pure Rust, zero CUDA SDK dependency.
Mamba1 selective-SSM block and multi-backend driver; SSM core via rlx-ssm flow (SelectiveScan / mamba1_step)
Idiomatic Rust wrappers for the NVIDIA CUDA stack (Driver API, Runtime API, NVRTC, cuBLAS, cuDNN, NCCL, NVML, ...). Umbrella crate.
Neural Quantum Error Decoder (NQED) - GNN-based decoder with O(d^2) Mamba state-space architecture
Model architectures for Kizzasi AGSP - Mamba, RWKV, S4, Transformer
Unified WebAssembly bindings for 18+ attention mechanisms: Neural, DAG, Graph, and Mamba SSM
A simple wrapper around micromamba to ensure that it is installed.
BitMamba: 1.58-bit Mamba language model with infinite context window - includes OpenAI-compatible API server
A Rust-based LLM training framework built on Candle
CPU-first pure-Rust supervised trainer for Selective State Space Models with Hyperspherical Prototype Networks.
No description provided.
No description provided.
No description provided.
No description provided.
No description provided.
No description provided.
No description provided.