Does lstm layer must have image data layer?

[Yatong Zhang]

I am trying to create a data layer that produces both image data and sequence data for lstm layer, which looks like combining  the image data layer and hdf5 data layer. But if I remove the image data layer, caffe always throws an error:

I0718 16:33:13.515290  3915 solver.cpp:337] Iteration 0, Testing net (#0)
I0718 16:33:13.551002  3915 solver.cpp:337] Iteration 0, Testing net (#1)

CUDA Exception: Warp Illegal Address

Program received signal CUDA_EXCEPTION_14, Warp Illegal Address.
[Switching focus to CUDA kernel 0, grid 3047, block (1849,0,0), thread (384,0,0), device 0, sm 0, warp 42, lane 0]
0x000000001a360748 in void caffe::EmbedForward<float>(int, float const*, float const*, int, int, int, float*)<<<(3907,1,1),(512,1,1)>>> ()
(cuda-gdb) bt
#0  0x000000001a360748 in void caffe::EmbedForward<float>(int, float const*, float const*, int, int, int, float*)<<<(3907,1,1),(512,1,1)>>> ()
(cuda-gdb)

Any advices are welcome.!

Thanks in advance

[Yatong Zhang]

All right, I figured it by myself. The data generating scripts in the examples has a bug. The scripts generate negative index values which is not allowed