NER model wrong output on Android

Hi guys, I am very new to the ML world. Need an urgent help, I am using a pre-built MobileBERT NER model from huggingface and converted to PyTorch Lite at .PTL file and it’s working great with python.

On Android I was able to create same tokenizer, prediction decoder but it’s very wrong on Android, same model performing well on python environment.

I tried reconverting the model with optimised and also not. Nothing is working, can anyone give me any hints please

Hint: The raw output of the model on Android are bigger numbers than than it’s showing on python environment