Benchmark performance with BERT?

[yuchenlin]

Hi Allan,

Thanks for the great repo. I was wondering if you would have the results of "This Implementation + BERT" similar to the numbers that you reported in https://github.com/allanj/pytorch_lstmcrf/blob/master/docs/benchmark.md ?

[yuchenlin]

And is there a plan for incorporating flair embeddings just like ELMO? https://github.com/flairNLP/flair

[allanj]

Yes. Both are plans on the way. I tried BERT (without fine-tuning) before but the performance was a bit lower than the one with ELMo. (I was using huggingface's code with version 0.6.0)

Seems like I have to push myself harder to incorporate the transformers package into the repo. :P

Probably I will do it with Flair first

[yuchenlin]

you are the best :)

[allanj]

finally..lol. benchmark