Train and test CodeBERT on an NL2PL Java code-comment pairs data

I want to use CodeBERT to generate Java code by using natural language as an input. The requirements are as follows:
1. Train CodeBERT on a sample NL2PL (natural language to programming language) data.
2. Show generated Java code.
3. Clearly explain the steps of the process.

The freelancer can use any nl2pl java pairs for training and testing CodeBERT. Simple dataset is preferred for understanding purposes.