US 12,169,513 B2
Method and system for multi-level artificial intelligence supercomputer design
Vijay Madisetti, Alpharetta, GA (US); and Arshdeep Bahga, Chandigarh (IN)
Assigned to Vijay Madisetti, Alpharetta, GA (US)
Filed by Vijay Madisetti, Alpharetta, GA (US)
Filed on Dec. 20, 2023, as Appl. No. 18/391,127.
Application 18/391,127 is a continuation of application No. 18/348,692, filed on Jul. 7, 2023, granted, now 12,001,462.
Claims priority of provisional application 63/463,913, filed on May 4, 2023.
Claims priority of provisional application 63/469,571, filed on May 30, 2023.
Prior Publication US 2024/0370472 A1, Nov. 7, 2024
Int. Cl. G06F 16/332 (2019.01); G06F 40/284 (2020.01)
CPC G06F 16/3329 (2019.01) [G06F 40/284 (2020.01)] 15 Claims
OG exemplary drawing
 
1. A system for enhancing large language models (LLMs), comprising:
an input component configured to:
receive an input text from an input data stream; and
split the input text into one or more tokens;
a batch processing component comprising one or more LLMs, the one or more LLMs being fine-tuned based on an aggregation of input data received from the input data stream over a predetermined time duration and configured to generate an output responsive to receiving a token of the one or more tokens, the output being based upon the token and associated with the token;
a ranking component configured to:
score outputs from the one or more LLMs of the batch processing component; and
rank the one or more tokens responsive to the scores of the outputs from the one or more LLMs of the batch processing component;
a clustering component configured to:
select a subset of highest-ranked tokens; and
consolidate the subset of the highest-ranked tokens into refined context token batches;
a control component configured to iteratively provide the refined context token batches to the one or more LLMs comprised by the batch processing component for further training of the one or more LLMs; and
a query component configured to:
receive input queries via a user interface;
generate derived queries responsive to the input queries;
transmit the derived queries to the batch processing component;
receive responses from the batch processing component responsive to the derived queries;
at least one of score the responses and rank the responses from the batch processing component; and
transmit one or more responses responsive to at least one of scoring the responses and ranking the responses via the user interface.