公开(公告)号：US20250103300A1

公开(公告)日：2025-03-27

申请号：US18424372

申请日：2024-01-26

Applicant: Salesforce, Inc.

Inventor： Hung Le ,  Hailin Chen ,  Amrita Saha ,  Akash Gokul ,  Doyen Sahoo ,  Shafiq Rayhan Joty

IPC: G06F8/30 ,  G06F40/40

Abstract: The embodiments are directed to generating source code for a program from a problem description. One or more pre-trained code large language models (LLMs) generate sub-modules from a problem description in a natural language. The sub-modules are filtered based on testing criteria and encoded into sub-module encodings in an embedding space. The sub-module encodings are clustered into multiple clusters. A subset of sub-modules encoding that are close to the centroids of the clusters are selected. The sub-set of sub-modules is decoded into representative sub-modules. The problem description is augmented with the representative sub-modules and fed into one or more pre-trained code LLMs and new sub-modules are generated. The iterations continue until a program is generated from the representative sub-modules.

公开(公告)号：US20250139411A1

公开(公告)日：2025-05-01

申请号：US18498229

申请日：2023-10-31

Applicant: Salesforce, Inc.

Inventor： Rithesh Murthy ,  Shelby Heinecke ,  Juan Carlos Niebles Duque ,  Zhiwei Liu ,  Le Xue ,  Weiran Yao ,  Yihao Feng ,  Zeyuan Chen ,  Akash Gokul ,  Devansh Arpit ,  Ran Xu ,  Lik Mui ,  Huan Wang ,  Caiming Xiong ,  Silvio Savarese

IPC: G06N3/0455 ,  G06N3/084

Abstract: Embodiments described herein provide a large language model (LLM) based AI agent that adopts Monte-Carlo Tree Search (MCTS) to execute a task. The LLM is prompted with a task description and it responds with its first attempted list of actions. Based on the success or failure of the first attempt, the LLM is prompted with an updated prompt which includes feedback from the first attempt based on a determined reward. The prompt may include a relative “score” for each action taken at each step. A numeric score may be mapped to a set of pre-defined text labels, such as “high” or “low” value putting the score in a form more suited for an LLM prompt. In this way, the LLM is iteratively given prompts which are updated with the scores from each action taken at each previous iterations so that it traverses different paths on the tree in each iteration.