Inventor(s)

Renee GagnonFollow

Abstract

This document discloses a novel technique called "Striped Mamba" for enhancing the parallelism and efficiency of State Space Models (SSMs) in neural network architectures. The Striped Mamba technique introduces a method of dividing input sequences into multiple "stripes" that can be processed in parallel, while maintaining the sequential dependencies crucial to SSMs. This approach aims to leverage the parallel processing capabilities of modern GPUs while preserving the temporal modeling strengths of SSMs.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.

Share

COinS