RoBERTa based model

Trained on 10% of the python code snippets in the codesearchnet dataset for code variable name suggestion. Use the 'fill-mask' inference pipeline to suggest variable names.