Show HN: Linear RNN/Reservoir hybrid generative model, one C file (no deps.) (raw.githubusercontent.com) AI

The post presents “lrnnSMDDS,” a compact C implementation of a linear RNN/reservoir hybrid generative model intended to run efficiently on CPUs. It describes architectural components like SwiGLU channel mixing, multi-scale token shifting, data-dependent decay via low-rank factors, and a slot-memory reservoir with fast generation support, along with build/run commands for training and text generation. The included code covers core tensor ops, tokenizers (char/word), model parameter structures, and forward-pass primitives such as ALiBi-style position bias.