🧀 BigCheese.ai

Social

A Neuromorphic Hardware-Compatible Transformer-Based Spiking Language Model

🧀

The paper 'Sorbet: A Neuromorphic Hardware-Compatible Transformer-Based Spiking Language Model' introduces Sorbet, a transformer-based spiking language model tailored for neuromorphic hardware compatibility. It leverages PTsoftmax and bit-shifting power normalization (BSPN) for energy-efficient operations, distillation, and quantization to maintain competitive performance.

  • Authored by Kaiwen Tang, Zhanglu Yan, Weng-Fai Wong
  • Submitted on 4 Sep 2024
  • Utilizes PTsoftmax and BSPN
  • Targets small language models (SLMs)
  • Tested on GLUE benchmark