load/save problem with CuArrays in tutorial/60.rnn.ipynb

[denizyuret]

No description provided.

[noyongkyoon]

Hello, Prof. Yuret.

When I choose not to train from scratch within the trainresults() function, opting instead for loading the model (and results) saved previously, I get the following error messages:

CUDA error: an illegal memory access was encountered (code 700, ERROR_ILLEGAL_ADDRESS)

Stacktrace:
[1] throw_api_error(res::CUDA.cudaError_enum)
@ CUDA ~/.julia/packages/CUDA/BbliS/lib/cudadrv/error.jl:89
[2] macro expansion
@ ~/.julia/packages/CUDA/BbliS/lib/cudadrv/error.jl:97 [inlined]
[3] cuMemAllocAsync
@ ~/.julia/packages/CUDA/BbliS/lib/utils/call.jl:26 [inlined]
[4] alloc(::Type{CUDA.Mem.DeviceBuffer}, bytesize::Int64; async::Bool, stream::CUDA.CuStream, pool::Nothing)
@ CUDA.Mem ~/.julia/packages/CUDA/BbliS/lib/cudadrv/memory.jl:83
[5] macro expansion
@ ~/.julia/packages/CUDA/BbliS/src/pool.jl:43 [inlined]
[6] macro expansion
@ ./timing.jl:383 [inlined]
[7] actual_alloc(bytes::Int64; async::Bool, stream::CUDA.CuStream)
@ CUDA ~/.julia/packages/CUDA/BbliS/src/pool.jl:41
[8] macro expansion
@ ~/.julia/packages/CUDA/BbliS/src/pool.jl:322 [inlined]
[9] macro expansion
@ ./timing.jl:383 [inlined]
[10] #_alloc#174
@ ~/.julia/packages/CUDA/BbliS/src/pool.jl:404 [inlined]
[11] #alloc#173
@ ~/.julia/packages/CUDA/BbliS/src/pool.jl:389 [inlined]
[12] alloc
@ ~/.julia/packages/CUDA/BbliS/src/pool.jl:383 [inlined]
[13] CUDA.CuArray{Float32, 3, CUDA.Mem.DeviceBuffer}(#unused#::UndefInitializer, dims::Tuple{Int64, Int64, Int64})
@ CUDA ~/.julia/packages/CUDA/BbliS/src/array.jl:42
[14] similar
@ ~/.julia/packages/CUDA/BbliS/src/array.jl:166 [inlined]
[15] similar(a::CUDA.CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}, dims::Tuple{Base.OneTo{Int64}, Base.OneTo{Int64}, Base.OneTo{Int64}})
@ Base ./abstractarray.jl:795
[16] _getindex(::CUDA.CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}, ::Base.Slice{Base.OneTo{Int64}}, ::Vararg{Any})
@ GPUArrays ~/.julia/packages/GPUArrays/9GYI6/src/host/indexing.jl:42
[17] getindex(::CUDA.CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}, ::Function, ::Matrix{UInt16})
@ GPUArrays ~/.julia/packages/GPUArrays/9GYI6/src/host/indexing.jl:38
[18] forw(::typeof(getindex), ::AutoGrad.Param{CUDA.CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}}, ::Vararg{Any}; kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
@ AutoGrad ~/.julia/packages/AutoGrad/1QZxP/src/core.jl:66
[19] forw
@ ~/.julia/packages/AutoGrad/1QZxP/src/core.jl:64 [inlined]
[20] getindex(::AutoGrad.Param{CUDA.CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}}, ::Function, ::Matrix{UInt16})
@ AutoGrad ./none:0
[21] (::Embed)(x::Matrix{UInt16})
@ Main ./In[7]:3
[22] (::Chain)(x::Matrix{UInt16})
@ Main ./In[5]:6
[23] tag(tagger::Chain, s::String)
@ Main ./In[24]:6
[24] top-level scope
@ In[26]:1

​This error message seems daunting to me.
Would there be a way to save this model and then loading it to use it later, without this apprent hiccup?
What does one have to do by way of resolving this issue? Any pointers?

Y. No

[denizyuret]

I think this is related to JuliaGPU/CUDA.jl#1833