Running network with fixed point instead of float

tiny-dnn / tiny-dnn

header only, dependency-free deep learning framework in C++14

http://tiny-dnn.readthedocs.io

Other

5.85k stars 1.39k forks source link

Running network with fixed point instead of float #899

Open ghost opened 7 years ago

ghost commented 7 years ago

Hi, I would like to know if there's some options to run a pre-trained network in order to predict a value with fixed point weights. Because of some hardware environment constraints I have to run the entire network as fast as I can. Using other libraries (for example, FANN) when I save the network to file I could choose wheter to save with float weights or fixed. There's something like this also with tiny-dnn?

Thanks in advance.

beru commented 6 years ago

@giacavicchioli That isn't implemented as of now.