My Titan Xps are starting to show their age :( :( :(
We have a brand new guide on using Mixed Precision in TensorFlow 2.1: https://www.tensorflow.org/guide/keras/mixed_precision … Speed up training and inference on GPU by up to 3x (and up to 50% on TPU)
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
why did 32 bit neural nets become a thing in the first place? that is way more than necessary
-
Premature optimization is the root of all kinds of evil
End of conversation
New conversation -
-
-
This might be a stupid question, as I don’t have much knowledge on hardware level optimization, but could we also go down to 8 bit? Most RGB images used for applications like computer vision are in 8 bit, why not go even further then?
-
It's not about input data, it's the precision of the weights of the neural network. You can go half precision (16 bit) and the gradient descent will be faster to converge at the cost of precision in the solution (one earlier local minimum instead of a deeper local minimum).
- Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.