I have fine tuned a custom NER model using pretrained model distilbert-base-uncased to extract magnitude of an earthquake event from a text.
Text: “Syria was hit by 8.1 magnitude earthquake.”
However, after fine tuning and generating predictions the model is able to extract magnitude values but omits the part after decimal point.
I have checked the values created by tokenizer and the magnitude values are not split into sub tokens. I checked but could not find any solution or cause for this.
Any suggestion on this would be really helpful.