mamba-3

Native Blackwell (sm_100) tcgen05 training backward for the gated-linear-recurrence family (GDN-2/GLA/KDA/SSD), plus a contract-grade verifier that falsifies published GPU kernels. Six open Mamba-3 Triton kernels including the first open MIMO backward; GRPO trainer graded on contract correctness. state-spaces/mamba#904

Native MLX reimplementation of the Mamba-3 architecture from state-spaces/mamba, adapted for Apple Silicon (MLX). The implementation mirrors the block-selection logic, core Mamba-3 recurrence concepts, inference-cache semantics, and config structure, while adding MLX-native generation, benchmarking, serialization, and LoRA fine-tuning. Open Source.

Position Bias in Mamba and Hybrid Language Models | Evidence Position Bias Across Sequence Mixers in Long-Context Question Answering | NeurIPS'2026

A simple, minimalistic, and explainable JAX implementation of Mamba 2 & Mamba 3

Dynamic State Space Ensemble for Multi Horizon Market Forecasting