I'm offline, bit there is an issue on GitHub that tell how to do it. The first step is to use the new GPU backend gpuarray. Fred
Le 16 sept. 2016 22:37, <[email protected]> a écrit : > Hi there, > > I have been reading the documentation as well as the discussion on using > float16 instead of float32 to reduce memory footprint on a single GPU. > > My task is to train a 3D convolutional neural network which has a large > memory footprint (as a result of the feature cubes and their gradients). > > How should I shift from currently float32 to float16 if possible? > > What is the step-by-step guide to do so? > > Simply updating the configuration file with floatX = float16 breaks the > scipt. > > > Many thanks, > > Best, > > Wilson > > -- > > --- > You received this message because you are subscribed to the Google Groups > "theano-users" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
