EdgeLLM: Fast On-Device LLM Inference With Speculative Decoding | IEEE Journals & Magazine | IEEE Xplore