I have following question/suggestion, which I hope is okay and does not clash with the saving/loading behaviour of models: I added the dtype information to the config of Loss such that the default floatx behaviour can be maintained even when created from config. If dtype is specified, then the Loss will set dtype from config.
If the dtype should not be passed on to the config then this pull request should be canceled.
Thank you very much for the update in
Loss
.I have following question/suggestion, which I hope is okay and does not clash with the saving/loading behaviour of models: I added the dtype information to the config of
Loss
such that the default floatx behaviour can be maintained even when created from config. Ifdtype
is specified, then theLoss
will set dtype from config.If the dtype should not be passed on to the config then this pull request should be canceled.