TokenPowerBench: Benchmarking the power consumption of LLM inference across engines and deployment configurations
