The data to be translated: Microsoft Research Team has recently introduced a technology called LLMLingua, which is renowned for its 20x compression ratio and accelerated model inference speed. LLMLingua is developed to address the issues posed by long prompts in large language models. It employs a series of critical strategies, including dynamic budget control, token-by-token iterative compression algorithms, and instruction tuning methods. Experimental results show that LLMLingua achieves significant performance in various scenarios, and can even achieve up to 20x compression. The emergence of LLMLingua provides a comprehensive solution to the difficulties brought by long prompts in large language models, greatly enhancing the effectiveness and affordability of the models.