US 12,596,736 B2Granted

Systems and methods for using prompt dissection for large language models

AssigneeInstabase, Inc.

InventorsVineeth Chinmaya Murthy, Rafal Powalski

Life of this patent

Estimating the term…

Loading patent…

Abstract

Systems and methods for using prompt dissection to prompts one or more machine learning models for a set of one or more documents are disclosed. Exemplary implementations may: store a set of document segments from the set of one or more documents; present a user interface to obtain a compound query that includes a first subquery and a second subquery; create subquery vectors that semantically represent the subqueries; determine a subset of a set of semantic vectors based on comparisons with the subquery vectors; create a combination of the individual document segments that are associated with the subset; provide one or more prompts to the one or more machine learning models, using the combination of the individual document segments as context; present replies from the one or more machine learning models, and/or perform other steps.

Claims18

Dependent claims

A system configured to use prompt dissection to prompt one or more machine learning models for a set of one or more documents, the system comprising:

electronic storage configured to electronically store information, wherein the stored information includes a set of document segments, wherein individual document segments are included in the set of one or more documents, wherein the stored information further includes a set of semantic vectors, wherein individual semantic vectors are associated with individual document segments;
and
one or more hardware processors configured by machine-readable instructions to:
effectuate a presentation of a user interface, the user interface being configured to obtain a compound query from a user, wherein the compound query includes two or more subqueries including a first subquery and a second subquery;
create, using the one or more machine learning models, a first subquery vector representing the first subquery;
determine a subset of the set of semantic vectors, wherein the determination is based on comparisons, wherein the comparisons include a first comparison using the first subquery vector;
create a combination of the individual document segments that are associated with the subset of the set of semantic vectors as determined;
provide one or more prompts to the one or more machine learning models that prompt the one or more machine learning models to generate one or more replies, using the created combination of the individual document segments as context for the one or more machine learning models, wherein the one or more prompts are based on the compound query, wherein the second subquery constrains the one or more replies, wherein an individual machine learning model from the one or more machine learning models is limited to a predetermined numbers of tokens as the context for the one or more prompts, and wherein the combination of the individual document segments is created such that the predetermined number of tokens is not exceeded;
and
present to the user, through the user interface, the one or more replies obtained from the one or more machine learning models in reply to the one or more prompts, wherein the one or more replies have been constrained by the second subquery.

7 more elements

A method of using prompt dissection to prompt one or more machine learning models for a set of one or more documents, the method comprising:

electronically storing information, wherein the stored information includes a set of document segments, wherein individual document segments are included in the set of one or more documents, wherein the stored information further includes a set of semantic vectors, wherein individual semantic vectors are associated with individual document segments;
and
effectuating a presentation of a user interface that obtains a compound query from a user, wherein the compound query includes two or more subqueries including a first subquery and a second subquery;
creating, using the one or more machine learning models, a first subquery vector representing the first subquery;
determining a subset of the set of semantic vectors, wherein the determination is based on comparisons, wherein the comparisons include a first comparison using the first subquery vector;
creating a combination of the individual document segments that are associated with the subset of the set of semantic vectors as determined;
providing one or more prompts to the one or more machine learning models that prompt the one or more machine learning models to generate one or more replies, using the created combination of the individual document segments as context for the one or more machine learning models, wherein the one or more prompts are based on the compound query, wherein the second subquery constrains the one or more replies, wherein an individual machine learning model from the one or more machine learning models is limited to a predetermined numbers of tokens as the context for the one or more prompts, and wherein the combination of the individual document segments is created such that the predetermined number of tokens is not exceeded;
and
presenting to the user, through the user interface, the one or more replies obtained from the one or more machine learning models in reply to the one or more prompts, wherein the one or more replies have been constrained by the second subquery.

6 more elements

Dependencies are inferred from the claim text.