vedantc6 / mtl-dts Goto Github PK

Running the main.yp with the standard config to train and evaluate the model using the default batch_size of 2 on CoNLL04 dataset, it takes roughly 30 hours to train the 10 epochs. Does somebody know if this is the expected behaviour?
Using cpu or cuda as the device wasnt a difference for me.

Hi, thanks a lot for sharing this repo

Do you plan to provide notebooks or snippets that explain how to use the model for inference on single strings, using a syntax of this kind:

s = "The sentence I want to extract the relations of"
nlp(s)
...

?

That would be great!

thanks for your work
i run the code but i found the NER f1 is 80.54
the ner f1 in the paper is 87
do you have some trick or extra config?
( i don't change any config in code)