Huggingface NER inference - aggregaton strategy

[cgpeltier]

Hi all,

We are looking to deploy a token classification/NER model trained with the Huggingface module (i.e., HuggingFacePredictor / HuggingFaceModel.

When we make inferences with our saved model locally we can use something like:

from transformers import pipeline
token_classifier = pipeline(
            "token-classification", 
            model=model,  
            tokenizer=tokenizer, 
            **aggregation_strategy="simple",**
        )

That allows us to group entity tokens in the inference. Does this or a similar option exist somewhere as an argument when we either deploy the model or query it?