How can we enable continuous learning with the Tiny Llama model ?

[TapendraBaduwal]

If we merge adapters then this work for continuous learning ? I expect the new train model to provide output for previously trained data as well.