0-dim t issue on multi gpus

[cb6e310]

Hi,
I ran into a problem when using nn.parallel model sampling. odeint's t parameter needs to be 1-D, and integrate() needs to be put into a t0, which is 0-dim. So this goes against the design of nn.parallel, which requires that scala not be passed in. How to solve this? Thanks.
This is my code.

This is the tricky spot.

Chunk error will rise when self.func is wrapped by parallel.