Towards Interpretable Sequence Continuation: Analyzing Shared Circuits in Large Language Models