Effect detection via convex hull similarity
AssigneeSalesforce, Inc.
InventorsRomain Cosentino, Sarath Shekkizhar, Yacov Salomon
Life of this patent
Estimating the term…
Abstract
An agent service can perform effect detection using a convex hull representation of an input. The agent service includes at least one agent large language model (LLM) trained to perform a plurality of tasks. A representation of the input is extracted from an attention layer of the at least one agent LLM. A value space representation is determined based on the extracted representation. A task is identified from the plurality of tasks to be performed by the agent based on the value space and an embedding corresponding to the task.
Claims20
A non-transitory machine-readable storage medium that provides instructions that, if executed by a processor, are configurable to cause the processor to perform operations comprising:
- obtaining an input to a agent service, wherein the agent service includes at least one agent large language model (LLM) trained to perform a plurality of tasks;
- extracting a representation of the input from an attention layer of the at least one agent LLM;
- determining a value space representation based on the extracted representation;
- and
- identifying a task from the plurality of tasks to be performed by the agent LLM based on the value space representation and an embedding corresponding to the task.
A method comprising:
- obtaining an input to a agent service, wherein the agent service includes at least one agent large language model (LLM) trained to perform a plurality of tasks;
- extracting a representation of the input from an attention layer of the at least one agent LLM;
- determining a value space representation based on the extracted representation;
- and
- identifying a task from the plurality of tasks to be performed by the agent LLM based on the value space representation and an embedding corresponding to the task.
An apparatus comprising:
- a processor;
- and
- a non-transitory machine-readable storage medium that provides instructions that, if executed by the processor, are configurable to cause the apparatus to perform operations comprising,
- obtaining an input to a agent service, wherein the agent service includes at least one agent large language model (LLM) trained to perform a plurality of tasks;
- extracting a representation of the input from an attention layer of the at least one agent LLM;
- determining a value space representation based on the extracted representation;
- and
- identifying a task from the plurality of tasks to be performed by the agent LLM based on the value space representation and an embedding corresponding to the task.
Dependencies are inferred from the claim text.