Is it possible to change the default Session configuration, either within Python or by setting environment variables, etc.?
Specifically I'd like
with tf.Session() as sess:
to use up much less memory when I'm running small side tests in parallel to other jobs. So I'd the above to behave the same as
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.1)
config = tf.ConfigProto(gpu_options=gpu_options)
with tf.Session(config=config) as sess:
Don't think there's a way to set process-wide default, but here's the pattern I use.
def create_session(): config = tf.ConfigProto(log_device_placement=True) config.gpu_options.per_process_gpu_memory_fraction=0.3 # don't hog all vRAM sess = tf.InteractiveSession("", config=config) return sess sess=create_session() a=tf.constant(1) b=tf.constant(2) sess.run([a+b])