Accelerating LLM Inference with Lossless Speculative Decoding Algorithms for Heterogeneous Vocabularies Paper • 2502.05202 • Published Jan 31 • 1