_

[mchancan]

_

[jerrybai1995]

You probably shouldn't expect the model to work well on all tasks with just the default parameters. I would suggest some tuning on the number of layers and the dilation scheme. Maybe the optimizer and the dropout rates as well.

BTW, the language modeling task itself is a sequence task with thousands (or actually, tens of thousands) of classes, so it also depends on the specific application that you are using TCN for.