# Replace the two methods below in order to add custom SessionRunHooks to
# the training procedure.
@staticmethod
def train_hooks():
return []
@staticmethod
def eval_hooks():
return []
However, the problem is that these hooks are already living in a session.
I want to quantize a graph after it has been evaluated in order to re-evaluate the quantized version of that graph. For this I need to slightly modify the original graph which means I have to create another session which is why SessionRunHooks are not the way to go here since their callbacks are always within a session context.
So is there a clean way to register a callback in the t2t-trainer which gives me control after the model evaluation stage (or something like this)?
stefan-falk
commented
6 years ago
In
T2TModelthere isHowever, the problem is that these hooks are already living in a session.
I want to quantize a graph after it has been evaluated in order to re-evaluate the quantized version of that graph. For this I need to slightly modify the original graph which means I have to create another session which is why
SessionRunHooksare not the way to go here since their callbacks are always within a session context.So is there a clean way to register a callback in the
t2t-trainerwhich gives me control after the model evaluation stage (or something like this)?