Researchers from Snowflake and CMU Introduce SuffixDecoding: A Novel Model-Free Approach to Accelerating Large Language Model (LLM) Inference through Speculative Decoding