multi-GPU

[wanghechong]

i want to use multi-GPU to train my model, but it report error like"torch/install/share/lua/5.1/cunn/DataParallelTable.lua:348: attempt to call method 'type' (a nil value)"
i do not know why, but the cunn is our package dowloaded from net

[guillaumekln]

What code are you running? OpenNMT does not use nn.DataParallelTable.

[wanghechong]

the problem is that i do language generation, and use a mlp encoder and a LSTM decoder, because i expecially want to test the big batch so i have to train model on two GPUs because of out of memory.
so i have tried it many times, i found DataParallelTable is not suitable for my situation, i see someone use cutorch.setDevice() but i do not know very clearly about his logic relation, so i hope you can give me some advice!!
@