Why are Eight Bits Enough for Deep Neural Networks?

Created on 2022-06-16T22:05:34-05:00

Return to the Index

This card pertains to a resource available on the internet.

This card can also be read via Gemini.

16-bit and 8-bit precision is often enough to still get competitive performance out of deep neural networks.

Fixed-point operations are also adequate as long as enough precision is available as scratch space.