MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202641079302 A) filed by Shyena Solutions Private Limited on June 27, 2026, for System And Method For Managing Graphics Processing Unit (gpu)-Based Training.
Inventors include J, Akshay; Rao Ramasagara, Sankarshana; Krishna, Sanath; and Kadur, Srikantha.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: The present disclosure provides a system (102) and a method (300) for managing Graphics Processing Unit (GPU)-based training. The method (300) includes receiving one or more parameters associated with one or more telemetry signals transmitted by at least one GPU on-board agent deployed on at least one GPU node of a training cluster (104). The method (300) includes computing, failure probability estimates and associated uncertainty value for the at least one GPU node based on the telemetry signals. The method (300) includes writing a checkpoint state for training job based on the failure probability estimate and the associated uncertainty value exceeding predetermined threshold. The method (300) includes receiving the checkpoint state for upload to object store (108). The method (300) includes routing training task to target GPU on-board agent (110a) identified by identity condition and resuming the training job from the checkpoint state uploaded on the object store (108).
Disclaimer: Curated by HT Syndication.
