Merging tokens to accelerate LLM inference with SLERP | Towards Data Science

We can significantly accelerate LLMs next token generation by merging consecutive pairs of tokens using SLERP, reducing the computing power…