decomposable_attention
decomposable_attention copied to clipboard
Big BUG
In the section of build_graph(): you set every layer of MLPs with a relu activation including the final layer for predict. And something others.
I got the loss-nan-error;