Loading snapshot with GPU backend

[greenflash1357]

I have trained a network and want to load the parameters from a snapshot.
It works perfectly fine when using CPU backend. But for GPU backend, I get this error message:

TypeError: read_refs: in typeassert, expected Array{Array{T,N},1}, got Array{Array{T,N},1}
 in read_refs(::JLD.JldDataset, ::Type{Array{Array{T,N},1}}, ::Int32, ::Int32, ::Tuple{Int64}) at JLD\src\JLD.jl:491
 in read_array(::JLD.JldDataset, ::HDF5.HDF5Datatype, ::Int32, ::Int32, ::Tuple{Int64}) at JLD\src\JLD.jl:427
 in read(::JLD.JldDataset) at JLD\src\JLD.jl:392
 in read_ref(::JLD.JldFile, ::HDF5.HDF5ReferenceObj) at JLD\src\JLD.jl:518
 in macro expansion at JLD\src\jld_types.jl:451 [inlined]
 in jlconvert(::Type{JLD.AssociativeWrapper{AbstractString,Array{Array{T,N},1},Dict{AbstractString,Array{Array{T,N},1}}}}, ::JLD.JldFile, ::Ptr{UInt8}) at JLD\src\jld_types.jl:581
 in read_scalar(::JLD.JldDataset, ::HDF5.HDF5Datatype, ::Type{T}) at JLD\src\JLD.jl:418
 in read(::JLD.JldDataset)

The network has also been trained with GPU backend.
Network architecture is:

data_layer = MemoryDataLayer(name="data", tops=[:data], batch_size=64, data=Array[zeros(Float32,31,31,3,64)])
conv1_layer  = ConvolutionLayer(name="conv1", n_filter=96, kernel=(7,7), bottoms=[:data], tops=[:conv1], neuron=Neurons.ReLU())
pool1_layer  = PoolingLayer(name="pool1", kernel=(2,2), stride=(2,2), bottoms=[:conv1], tops=[:pool1])
conv2_layer = ConvolutionLayer(name="conv2", n_filter=256, kernel=(5,5), bottoms=[:pool1], tops=[:conv2], neuron=Neurons.ReLU())
pool2_layer = PoolingLayer(name="pool2", kernel=(2,2), stride=(2,2), bottoms=[:conv2], tops=[:pool2])
fc1_layer   = InnerProductLayer(name="ip1", output_dim=1024, neuron=Neurons.ReLU(), bottoms=[:pool2], tops=[:ip1])
fc2_layer   = InnerProductLayer(name="ip2", output_dim=2, bottoms=[:ip1], tops=[:ip2])
softmax_layer = SoftmaxLayer(name="class", tops=[:prob], bottoms=[:ip2])
mem_out = MemoryOutputLayer(name="output", bottoms=[:prob])

[greenflash1357]

JuliaIO/JLD.jl#170 seems to fix this.