-
Notifications
You must be signed in to change notification settings - Fork 103
Open
Labels
brainpy.dynissue belongs to brainpy.dyn moduleissue belongs to brainpy.dyn modulebrainpy.mathissue belongs to brainpy.math moduleissue belongs to brainpy.math moduleenhancementNew feature or requestNew feature or request
Description
Does BrainPy fully support half-precision floating point numbers? I have tried to changed some of my own BrainPy code from using brainpy.math.float32
to brainpy.math.float16
or brainpy.math.bfloat16
(by explicitly setting the dtype of all variables and using a debugger to make sure that they won't be promoted to float32
), but it seems that the GPU memory consumption and running speed is almost the same as using float32
.
Metadata
Metadata
Assignees
Labels
brainpy.dynissue belongs to brainpy.dyn moduleissue belongs to brainpy.dyn modulebrainpy.mathissue belongs to brainpy.math moduleissue belongs to brainpy.math moduleenhancementNew feature or requestNew feature or request