fashion_bert: masked_lm_ids

Hello,

I would like to train fashion_bert on a different dataset.
To construct the data, I need to input "masked_lm_ids".
I was wondering what this is?

Thank you for your help.

Start at the fashion_bert entry point and follow the data-construction path to identify where masked_lm_ids is consumed and what it represents. Document that explanation for users training on a different dataset, and verify that the guidance is sufficient to construct the required input.

Written by the indexing model from the issue text.