Compress prompts to speed up language models!