In a blog post today, Apple engineers have shared new details on a collaboration with NVIDIA to implement faster text generation performance with large language models.

Apple published and open sourced its Recurrent Drafter (ReDrafter) technique earlier this year. It represents a new method for generating text with LLMs that is significantly faster and “achieves state of the art performance.” It combines two techniques: beam search (to explore multiple possibilities) and dynamic tree attention (to efficiently handle choices).

While its research demonstrated strong results, Apple collaborated with NVIDIA to apply ReDrafter in production. As part of this collaboration, ReDrafter was integrated into NVIDIA TensorRT-LLM, a tool that helps run LLMs faster on NVIDIA GPUs.

Here are the results:

• To enable the integration of ReDrafter, NVIDIA added new operators or exposed existing ones, which considerably improved TensorRT-LLM’s capability to accommodate sophisticated models and decoding methods. ML developers using NVIDIA GPUs can now easily benefit from ReDrafter’s accelerated token generation for their production LLM applications with TensorRT-LLM.
• In benchmarking a tens-of-billions parameter production model on NVIDIA GPUs, using the NVIDIA TensorRT-LLM inference acceleration framework with ReDrafter, we have seen 2.7x speed-up in generated tokens per second for greedy decoding. These benchmark results indicate this tech could significantly reduce latency users may experience, while also using fewer GPUs and consuming less power.

“LLMs are increasingly being used to power production applications, and improving inference efficiency can both impact computational costs and reduce latency for users,” Apple’s machine learning researchers conclude. “With ReDrafter’s novel approach to speculative decoding integrated into the NVIDIA TensorRT-LLM framework, developers can now benefit from faster token generation on NVIDIA GPUs for their production LLM applications.”

Source: 9to5mac