Translated data: SwiftInfer is a domestic open-source project that has recently succeeded in implementing infinite streaming input inference, enhancing the performance of large model inference by 46%. This provides an efficient and reliable implementation solution for multi-turn dialogue inference with large models. The Colossal-AI team has open-sourced SwiftInfer, aiming to reduce the development and application costs of training/fine-tuning/inference for AI large models, improve model task performance, and reduce GPU requirements.