Any way to load roBERTa model using kotlin? I am aware that certain language have implementation to ...

# datascience

[Ananiya]

Any way to load roBERTa model using kotlin?
I am aware that certain language have implementation to vectorize the text like RobertaTokenizerFast in pytho..
In kotlin
I was managed to get the size and some information using `OnnxInferenceModel`(kotlindl) but I am quite confused on vectorizing the sentence because i don't have access to the vocabulary, the approach used the vectorize the sent.. or the dimensions to implement the vectorizer manualy

So I was wondering if there's any possible way or library that could workaround

[Ilya Muradyan]

@Julia Beliaeva @zaleslaw

[zaleslaw]

At this moment I could not find a way to port tokenizers

[Ananiya]

Thanks @zaleslaw for the response
Is there any other model that I can utilize instead of roBerta
I tried importing pretrained keras model which doesn't workout due to the the lack of Embedding layer
Or maybe I should wait until the layers are ready

[roman.belov]

@Ananiya I guess https://github.com/londogard/londogard-nlp-toolkit/ should work for preprocessing

[Ananiya]

@roman.belov thanks for the response!
I looked at this library earlier but doesn't seems to provide same vectorizer as Roberta

[Hampus Londögård]

@Ananiya it’s something I’m working on adding through using DJL. You should be able to use DJL directly for now 🙂 It supports HuggingFace Tokenizers.

[Ananiya]

Interesting! thanks @Hampus Londögård for letting me know

[Hampus Londögård]

@Ananiya I’ve added support for ClassifierPipeline & TokenClassificationPipeline which means that Text. Classiification and Token Classiification (e.g. NER) is supported now 👍

Files can be loaded from file-system (ONNX & PyTorch) and from HuggingFace Hub (ONNX)

[Ananiya]

Fantastic @Hampus Londögård ! Excited to test it