Apple and NVIDIA partner to boost AI language models’ speed and efficiency

Apple has announced a new partnership with NVIDIA aimed at improving the performance of large language models (LLMs). This collaboration introduces a cutting-edge text generation technique that significantly speeds up AI applications, benefiting industries that rely on powerful language models.

Earlier this year, Apple unveiled a machine learning approach called Recurrent Drafter (ReDrafter). This technology combines two advanced methods: beam search and dynamic tree attention. Beam search works by evaluating multiple potential text sequences at once, allowing for more accurate and diverse results. Meanwhile, tree attention helps streamline these sequences, reducing redundant information and improving processing efficiency. The result is faster and more efficient text generation.