Multi-Agent deep reinforcement learning based task offloading and caching in computing power network | Synapse