Apple and NVIDIA have announced a collaboration to enhance the performance of large language models (LLMs) through a new technique called Recurrent Drafter (ReDrafter). This open-source method has been integrated into NVIDIA's TensorRT-LLM, resulting in a reported 2.7x increase in token generation speed for production-scale models on H100 GPUs. The partnership aims to optimize LLM inference workloads, moving towards more efficient processing methods that go beyond traditional batching. This development marks a significant advancement in generative artificial intelligence, with potential implications for various applications across industries.