Speculative Streaming: Fast LLM Inference without Auxiliary Models | Read Paper on Bytez