Today I’ve published a model fine-tuned for token classification of legislation references to the Hugging Face Hub and set up a model card including a widget (romjansen/robbert-base-v2-NER-NL-legislation-refs · Hugging Face). Using Hugging Face’s inference API widget this model can be quickly tested on the provided examples.
However, the hosted inference API widget incorrectly presents the last token of a legislation reference as a seperate entity due to the workings of its ‘simple’ aggregation_strategy. While this model was fine-tuned on training data labelled in accordence with the BILOU scheme, the hosted inference API groups entities by merging B- and I- tags when the tag is similar (thereby omitting the L- tags). Does anybody know if I can adjust the aggregation_strategy to use the right tagging scheme?