[NeurIPS 2021 Spotlight] & [IJCV 2024] SOFT: Softmax-free Transformer with Linear Complexity

Small-Dimensional Quadratic Programming in Linear Time

PyTorch implementation of Mamba: Selective State Space Models for efficient linear-time sequence modeling. An alternative to Transformers with O(N) complexity.

Algorithm for solving the Hamiltonian cycle problem deterministically and in linear time on all instances of discocube graphs (tested for graphs with over 8 billion vertices). Discocube graphs are 3-dimensional grid graphs derived from: a polycube of an octahedron | a Hauy construction of an octahedron with cubes as identical building blocks...

FluidLM is a Transformer-free language model replacing O(N²) self-attention with reaction-diffusion PDEs - achieving O(N) scaling, adaptive computation, and no KV-cache.

An algorithm for solving the Hamiltonian cycle problem deterministically and in linear time on all instances of discocube graphs (tested for over graphs with 1 billion vertices).

This program takes a string and multiple delimiter characters as input, and splits the string based on said delimiters. The function that performs this returns a vector of tokens, which is actually the input broken down into small strings according to the delimiters provided.

PyWave