You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
dp.Experiment can be parametrized with target_module which is just passed to dp.Propagator instances.
dp.Propagator has :type() function which is used to convert _loss (Criterion), but not target_module.
The problem is that when dp.Experiment:cuda() is called the target_module is not modified, so execution fails unless targetModule:cuda() call is done manually in addition to dp.Experiment:cuda()
It would be nice for dp.Propagator to handle target_module conversion too:
function Propagator:type(new_type)
if self._loss then
self._loss:type(new_type)
end
-- here goes support for target_module
if self._target_module then
self._target_module:type(new_type)
end
end
Alternatively this can be done in dp.Experiment.
Thanks.
The text was updated successfully, but these errors were encountered:
dp.Experiment can be parametrized with target_module which is just passed to dp.Propagator instances.
dp.Propagator has :type() function which is used to convert _loss (Criterion), but not target_module.
The problem is that when dp.Experiment:cuda() is called the target_module is not modified, so execution fails unless targetModule:cuda() call is done manually in addition to dp.Experiment:cuda()
It would be nice for dp.Propagator to handle target_module conversion too:
Alternatively this can be done in dp.Experiment.
Thanks.
The text was updated successfully, but these errors were encountered: