Release model - error: unknown Torch class <onmt.CustomizedAttention>

[jkcchan]

When I run th tools/release_model.lua -model <model> -gpuid 1, this is what happens. (using a model trained from this paragraph model)

[02/06/18 06:38:42 INFO] Using GPU(s): 1
[02/06/18 06:38:42 WARNING] The caching CUDA memory allocator is enabled. This allocator improves performance at the cost of a higher GPU memory usage. To optimize for memory, consider disabling it by setting the environment variable: THC_CACHING_ALLOCATOR=0
[02/06/18 06:38:42 INFO] Loading model '../<foldername>/paragraph/model/840B.300d.rnn.para_epoch1_107.09.t7'...
/root/torch/install/bin/luajit: tools/release_model.lua:93: unable to load the model (/root/torch/install/share/lua/5.1/torch/File.lua:343: unknown Torch class <onmt.CustomizedAttention>). If you are releasing a GPU model, it needs to be loaded on the GPU first (set -gpuid > 0)
stack traceback:
	[C]: in function 'error'
	tools/release_model.lua:93: in function 'main'
	tools/release_model.lua:116: in main chunk
	[C]: in function 'dofile'
	/root/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
	[C]: at 0x00406670

Let me know if there's anything I can do (i.e. changing to different Torch class?)

[guillaumekln]

When loading a class, the Torch loader expects it to be available. I would recommend to simply copy the release script in nqg/paragraph and run it from there to ensure that all custom class definitions are available.

[jkcchan]

Thanks @guillaumekln ! I've been unavailable to try, but will report back.

[Prakhar0409]

@jkcchan Merely copying the release script into the nqg/paragraph folder did not work for me. What I rather did was copy the onmt/modules/* from the project you pointed to and copied them to onmt/modules/ folder of this project and than used the command. This should hopefully work for you too.