Multi-round representation builder for large language model
AssigneeSap SE
InventorDavid Kunz
Life of this patent
Estimating the term…
Abstract
In an example embodiment, rather than use large language model (LLM) to directly generate desired computer code, an intermediate representation is generated by the LLM. The LLM is used to generate the portion of the computer code that cannot be computed programmatically (which may be called the “creative” part for purposes of the present disclosure). The intermediate representation can then be fed into a separate programmatic component that compiles the intermediate representation into compilable computer code. This fine-tuning may involve, for example, sanitizing the intermediate representation, enhancing the intermediate representation, and formatting the intermediate file, as well as modifying the intermediate representation based on a feature set.
Claims20
A system comprising:
- at least one hardware processor;
- and
- a computer-readable medium storing instructions that, when executed by the at least one hardware processor, cause the at least one hardware processor to perform operations comprising:
- accessing a schema and metadata about the schema, the schema defining a plurality of properties, the metadata comprising, for each property of the plurality of properties, a type of information to generate an instance of a corresponding property;
- generating a prompt using the schema and the metadata;
- passing the prompt to a large language model (LLM);
- receiving a response from the LLM, the response comprising one or more questions;
- causing the one or more questions from the response to be presented to a user;
- receiving one or more answers to the one or more questions;
- generating another prompt using the schema, the metadata, and the one or more answers;
- passing the another prompt to the LLM;
- repeating the receiving a response, causing, receiving one or more answers, generating another prompt, and generating the another prompt until, for each property of the plurality of properties, the corresponding type of information to generate an instance of the corresponding property has been received;
- generating a representation prompt comprising the schema and the received answers;
- passing the representation prompt to the LLM;
- and
- receiving, from the LLM, a generated representation in response to the representation prompt.
A method comprising:
- accessing a schema and metadata about the schema, the schema defining a plurality of properties, the metadata comprising, for each property of the plurality of properties, a type of information to generate an instance of a corresponding property;
- generating a prompt using the schema and the metadata;
- passing the prompt to a large language model (LLM);
- receiving a response from the LLM, the response comprising one or more questions;
- causing the one or more questions from the response to be asked to a user;
- receiving one or more answers to the one or more questions;
- generating another prompt using the schema, the metadata, and the one or more answers;
- passing the another prompt to the LLM;
- repeating the receiving a response, causing, receiving one or more answers, generating another prompt, and generating the another prompt until, for each property of the plurality of properties, the corresponding type of information to generate an instance of the corresponding property has been received;
- generating a representation prompt comprising the schema and the received answers;
- passing the representation prompt to the LLM;
- and
- receiving, from the LLM, a generated representation in response to the representation prompt.
A non-transitory machine-readable medium storing instructions which, when executed by one or more processors, cause the one or more processors to perform operations comprising:
- accessing a schema and metadata about the schema, the schema defining a plurality of properties, the metadata comprising, for each property of the plurality of properties, a type of information to generate an instance of a corresponding property;
- generating a prompt using the schema and the metadata;
- passing the prompt to a large language model (LLM);
- receiving a response from the LLM, the response comprising one or more questions;
- causing the one or more questions from the response to be asked to a user;
- receiving one or more answers to the one or more questions;
- generating another prompt using the schema, the metadata, and the one or more answers;
- passing the another prompt to the LLM;
- repeating the receiving a response, causing, receiving one or more answers, generating another prompt, and generating the another prompt until, for each property of the plurality of properties, the corresponding type of information to generate an instance of the corresponding property has been received;
- generating a representation prompt comprising the schema and the received answers;
- passing the representation prompt to the LLM;
- and
- receiving, from the LLM, a generated representation in response to the representation prompt.
Dependencies are inferred from the claim text.