ALEXANDRIA, Va., Aug. 5 -- United States Patent no. 12,699,841, issued on Aug. 4, was assigned to Intuit Inc. (Mountain View, Calif.).
"Importance aware selective correction for speculative decoding in large models" was invented by Ofir Ben Shoham (Petah Tikva, Israel), Sagiv Antebi (Petah Tikva, Israel), Shai Ardazi (Petah Tikva, Israel) and Gad Markovits (Petah Tikva, Israel).
According to the abstract* released by the U.S. Patent & Trademark Office: "Importance aware selective correction for speculative decoding in large models include obtaining, via a prompt interface, an input prompt to a large model (LM) and intercepting the input prompt by a draft model. The draft model generates a set of output tokens in response to the input prom...