Pre-trained models for structured prediction

Is there a detailed tutorial for structured prediction tasks like dependency/constituancy parsing, SRL etc.?
I use AllenNLP 1.0.0 demos to understand the structure of predictions but it is very obscure without API. Thanks in advance.

We don’t have walk-throughs of those models in the guide at this point, but we do have documentation for the models here. If you have questions that aren’t answered by the documentation, feel free to ask them.

Thanks for quick response!
I do have a question and I already asked on SO:

Any help is appreciated!

See here. Run some data through the model a few times so that predictions stabilize, then it should match what’s on the demo.

The paper in your reference does not include constituency parser results (only SRL which is a different parser)
Actually, the demo I linked above sites a different paper https://arxiv.org/pdf/1805.06556.pdf but it does not discuss pre-trained model performance.

The model that you’re using includes ELMo as a component. Any model that uses ELMo is going to exhibit the warmup behavior described in the page I linked above.

I see. But the different result in my case is not due to the warmup because it consistently the same. Perhaps, it is due to GUI rendering used by the demo. I guess I can build a demo from source and debug it.