SRU
Simple Recurrent Units for Highly Parallelizable Recurrence
The SRU is introduced, which is a speed-optimized recurrent cell. It claims a ~10x speedup vs LSTM on a GPU or accelerator.
James Leopore
from a laptop in Sunnyvale
from a laptop in Sunnyvale
Simple Recurrent Units for Highly Parallelizable Recurrence
The SRU is introduced, which is a speed-optimized recurrent cell. It claims a ~10x speedup vs LSTM on a GPU or accelerator.