Improving the performance of RLM for Long-Context Reasoning with the help of structured graph state and combining Mamba SSM | Synapse