Gengyu Wang

Title: Gengyu Wang and Patent Innovation in AI Caching

Introduction

Gengyu Wang is an inventor based in Long Island City, New York, whose patent work addresses efficient caching for large language model systems. He is associated with International Business Machines Corporation.

Latest Patents

Wang is named on 1 patent, titled “Caching Policy to Cache Key-Value Vectors.” The invention describes systems and methods for caching key-value vectors generated when a large language model processes a fixed prefix. Under a smart encoding policy, the system selectively stores those vectors when a temporal moving average for the prefix exceeds a defined threshold. This approach is designed to make reuse of frequently encountered prefixes more efficient.

Career Highlights

The patent reflects Wang’s work on computer systems and machine-learning infrastructure, particularly techniques for managing cached data produced by large language models. His disclosed approach combines request processing, usage measurements, and threshold-based cache selection.

Collaborations

Wang’s listed coworkers include Haode Qi and Nicholas James Hill. The patent is associated with International Business Machines Corporation.

Conclusion

Gengyu Wang’s patent focuses on a practical challenge in large language model deployment: efficiently caching key-value vectors for recurring fixed prefixes.