(Neither affiliated with keras nor nvidia).
Little callback for
keras which will add desired statistics to the
output of keras. Other callbacks like TensorBoard will pick them up
and produce plots accordingly.
Will work only with nvidia devices, obviously.
Please note, that, if everything runs optimally, the metrics are not that useful (e.g. temperature will reach a steady-state and utilization will be constantly high). However, the values might show if the current ML setup is not utilizing the GPU properly, or serve as a remote (via TensorBoard) way to watch device health.
Usage:
from keras_nvidia_statistics import NvidiaDeviceStatistics
# ...
model.fit(X, Y, callbacks=[NvidiaDeviceStatistics(), ...])Yielding in TensorBoard:
py3nvml, which is a fork of the NVIDIA Management Library.
BSD