Kenton Lee

We haven't test this code on a multi-GPU setup. Are you sure it's using all available GPUs?

I just tried it with ``` tensorflow>1.13.1 tensorflow-hub>0.4.0 ``` and the demo still works for at least those versions. Are you still having trouble running the code?

The files should be set up such that the famous Pierre Vinken example (first sentence of the dev set) can be found via this path: neuralccg/data/ccgbank_1_1/data/AUTO/00/wsj_0001.auto Does this match something...

It looks like the failure is happening here: https://github.com/kentonl/EasySRL/blob/maven/src/edu/uw/easysrl/corpora/CCGBankDependencies.java#L386 There is likely a mismatch between the content of the online version and the disc version of CCGBank. You can debug...

The GPU version can be used with two simple changes: (1) Use [this model](http://lil.cs.washington.edu/resources/model_tritrain_finetune_long.tgz) instead. (2) Download [this binary](http://lil.cs.washington.edu/resources/libtaggerflow.so) and place it in the `lib` directory. An example of a...

Take a look at this branch under my fork of EasySRL: https://github.com/kentonl/EasySRL/tree/maven It uses maven instead to retrieve most of its dependencies (though some custom jars remain).

To follow-up, here's an example that uses EasySRL as a maven dependency: https://github.com/uwnlp/neuralccg/blob/master/neuralccg/pom.xml

Those dependencies are available here: https://github.com/kentonl/EasySRL/tree/maven/lib/edu/uw, but they should be automatically imported if you follow the example from neuralccg linked above.