Floating-Point Format (bfloat16)

From GM-RKB
Jump to navigation Jump to search

A Floating-Point Format (bfloat16) is a Computer Number Format occupying 16 bits in computer memory.



References

2021

  1. Tensorflow Authors (2018-07-23). "ResNet-50 using BFloat16 on TPU". Google. Retrieved 2018-11-06.