training crashes with too big data

[prazek]

Hi,
I get a lot of errors like:
torch/extra/cutorch/lib/THC/THCTensorIndex.cu:321: void indexSelectLargeIndex(TensorInfo<T, IndexType>, TensorInfo<T, IndexType>, TensorInfo<long, IndexType>, int, int, IndexType, IndexType, long) [with T = float, IndexType = unsigned int, DstDim = 2, SrcDim = 2, IdxDim = -2]: block: [37,0,0], thread: [0,0,0] Assertion srcIndex < srcSelectDimSize failed.

stack traceback:
[C]: in function 'addmm'
/home/prazek/torch/install/share/lua/5.1/nn/Linear.lua:66: in function 'func'
...e/prazek/torch/install/share/lua/5.1/nngraph/gmodule.lua:345: in function 'neteval'
...e/prazek/torch/install/share/lua/5.1/nngraph/gmodule.lua:380: in function 'forward'
train.lua:299: in function 'opfunc'
/home/prazek/torch/install/share/lua/5.1/optim/rmsprop.lua:35: in function 'optimizer'
train.lua:358: in main chunk
[C]: in function 'dofile'
...azek/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
[C]: at 0x00405db0

It fails like this with glove. On the other hand without glove swaps whole machine in weird way (probably on gpu). Any guesses?