Portable xLSTM (sLSTM and mLSTM) kernels in C99: single header, no framework, no allocator, no OS.

xlstm s/m model and mingru - minlstm for parallel concurrency or threads using stacked layers or columns like the cerebral cortex, parallel or distributed concurrent networks, let's try everything except attention :)

Orthogonalizing an mLSTM's memory read is a removable training scaffold, not a memory gain — paper, code, and per-seed results.

Rust port of the official xLSTM repository integrating BitNet (1.58-bit), minGRU, minLSTM, and Transformers in Python and Rust.