Translated data: The LLM-Shearing pruning method developed by the team of Chen Danqi at Princeton University employs structured pruning and dynamic batch loading to reduce large language models into smaller, more efficient versions, significantly cutting down on computational resource requirements. The pruned models excel in various downstream tasks, showcasing high versatility and offering a new approach for constructing medium-scale powerful language models.