Reducing Energy Footprint of LLM Inference Through FPGA-Based Heterogeneous Computing Platforms | Synapse