MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202441105209 A) filed by Samsung Electronics Co. , Ltd. on December 31, 2024, for Systems And Methods For Generating A Multilingual Large Language Model Using A Cascaded Model.
Inventors include Karjee, Jyotirmoy; Vala, Vanraj; and Mohanty, Debi Prasanna.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: Disclosed is a method (300) for generating a multilingual large language model (LLM) using a cascaded model (200). A first processing block (202) is trained based on a first multilingual dataset. One or more model artifacts associated with the first processing block (202) are generated based on the first multilingual dataset. A second processing block (204), comprising at least a multilingual LLM with a plurality of embeddings, is trained based on a second multilingual dataset. Each of the plurality of embeddings is associated with a distinct language. A compressed multilingual LLM is generated from the multilingual LLM. A third processing block (206) is trained with a third multilingual dataset. Moreover, an optimized multilingual LLM is generated from the compressed multilingual LLM based on removing one or more embeddings, from among the plurality of embeddings, corresponding to one or more languages excluded from the third multilingual dataset.
Disclaimer: Curated by HT Syndication.
