Hi
I'm trying to do hyperparameters optimization on a GPU machine with tensorflow-gpu installed.
In my Keras model I manage the gpu memory with the following code snippet (without it, tensorflow occupies all available GPU memory by default) :
import tensorflow as tf
import keras
keras.backend.clear_session()
config_gpu = tf.ConfigProto()
config_gpu.gpu_options.allow_growth=True
sess = tf.Session(config=config_gpu)
keras.backend.set_session(sess)
However, as I have no idea about how gpflowopt uses tensorflow I can't manage its GPU memory usage, and I am running out of memory each time I launch optimization experiment.
Do you have any suggestion about how (or where) I can modify the gpflowopt code to manage gpu memory allocation ?
Hi
I'm trying to do hyperparameters optimization on a GPU machine with tensorflow-gpu installed.
In my Keras model I manage the gpu memory with the following code snippet (without it, tensorflow occupies all available GPU memory by default) :
However, as I have no idea about how gpflowopt uses tensorflow I can't manage its GPU memory usage, and I am running out of memory each time I launch optimization experiment.
Do you have any suggestion about how (or where) I can modify the gpflowopt code to manage gpu memory allocation ?