公开(公告)号：US20250094687A1

公开(公告)日：2025-03-20

申请号：US18758441

申请日：2024-06-28

Applicant: Oracle International Corporation

Inventor： Zheng Wang ,  Yazhe Hu ,  Mengqing Guo ,  Tao Sheng ,  Jun Qian ,  Vinod Murli Mamtani

IPC: G06F40/166 ,  G06F40/253

Abstract: Techniques for generating repetition-free text using a large language model (LLM) are provided. In one technique, textual content that was generated by an LLM is accessed, where the textual content comprises a plurality of sub-components including a first sub-component and a second sub-component. A first embedding that represents the first sub-component is generated and a second embedding that represents the second sub-component is generated. Based on a similarity between the first embedding and the second embedding, it is determined whether the second sub-component is repetitious with respect to the first sub-component. In response to determining that the second sub-component is repetitious with respect to the first sub-component, at least a portion of the second sub-component is removed from the textual content.

公开(公告)号：US20250094686A1

公开(公告)日：2025-03-20

申请号：US18758321

申请日：2024-06-28

Applicant: Oracle International Corporation

Inventor： Zheng Wang ,  Yazhe Hu ,  Mengqing Guo ,  Tao Sheng ,  Jun Qian ,  Vinod Murli Mamtani

IPC: G06F40/166 ,  G06F40/279

Abstract: Techniques for modifying a narrative point of view for content generated by a machine-learned model, such as a large language model (LLM), are provided. In one technique, a first textual content that was generated by an LLM is accessed. A narrative point of view (NPOV) detection operation is performed on a first portion of the first textual content to identify a first NPOV corresponding to the first portion of the first textual content. Based on an output, of the NPOV detection operation, that indicates that the first NPOV does not meet one or more NPOV criteria, the first portion of the first textual content is modified to generate a modified textual content. The modified textual content is submitted to the LLM, causing the LLM to generate a second textual content.

公开(公告)号：US20250094455A1

公开(公告)日：2025-03-20

申请号：US18885347

申请日：2024-09-13

Applicant: Oracle International Corporation

Inventor： Umanga Bista ,  Ying Xu ,  Aashna Devang Kanuga ,  Xin Xu ,  Vishal Vishnoi ,  Charles Woodrow Dickstein

IPC: G06F16/332 ,  G06F16/33

Abstract: Techniques are disclosed herein for contextual query rewriting. The techniques include inputting a first user utterance and a conversation history to a first language model. The first language model identifies an ambiguity in the first user utterance and one or more terms in the conversation history to resolve the ambiguity, modifies the first user utterance to include the one or more terms identified to resolve the ambiguity to generate a modified utterance, and outputs the modified utterance. The computing system provides the modified utterance as input to a second language model. The second language model performs a natural language processing task based on the input modified utterance and outputs a result. The computing system outputs a response to the first user utterance based on the result.

公开(公告)号：US20250094411A1

公开(公告)日：2025-03-20

申请号：US18598122

申请日：2024-03-07

Applicant: Oracle International Corporation

Inventor： Graham Ivey ,  Shih-Yu Huang ,  Yunrui Li ,  Shampa Chakravarty

IPC: G06F16/23

Abstract: The present disclosure relates to adaptively overlapping redo writes. A log writer, while operating in a thin mode, may assign a first log writer group of a plurality of log writer groups to write one or more first redo log records to an online redo log in response to determining that a pipelining parameter is satisfied. The thin mode may be associated with one or more target sizes that are less than one or more target sizes associated with a thick mode. The log writer may determine to operate the thick mode based at least in part on at least a portion of the plurality of log writer groups being unavailable to write one or more second redo log records to the online redo log. The log writer, while operating in the thick mode, may assign a second log writer group of the plurality of log writer groups to write one or more second redo log records from the log buffer to the online redo log in response to determining that an amount of redo log records in the log buffer meets one of the one or more target sizes associated with the thick mode. The log writer, while operating in the thick mode, may assign a third log writer group of the plurality of log writer groups to write one or more second redo log records from the log buffer to the online redo log in response to determining that a highest busy group number meets or exceeds a core threshold.

公开(公告)号：US20250094399A1

公开(公告)日：2025-03-20

申请号：US18885639

申请日：2024-09-14

Applicant: Oracle International Corporation

Inventor： Aurosish Mishra ,  Shasank Kisan Chavan ,  Tirthankar Lahiri ,  Agnivo Saha ,  Rohan Aggarwal ,  Weiwei Gong ,  Juan R. Loaiza

IPC: G06F16/22

Abstract: Techniques for automatically selecting a type of vector index are provided. In one technique, in response to determining to generate a vector index based on a base table that stores a plurality of vectors, a number of the plurality of vectors is identified. Based at least on the number of the plurality of vectors, a particular type of vector index is identified from among a plurality of types of vector indexes. Examples of the plurality of types include an HNSW index and an IVF index. A vector index of the particular type is generated for the base table. Another criterion in identifying a type of vector index to generate is the number of neighbors that is a parameter in generating a certain type of vector index.

公开(公告)号：US20250094398A1

公开(公告)日：2025-03-20

申请号：US18885531

申请日：2024-09-13

Applicant: Oracle International Corporation

Inventor： Aleksandra Czarlinska ,  Saurabh Naresh Netravalkar ,  Denis B. Mukhin ,  Harichandan Roy ,  Zhen Hua Liu ,  Sebastian de la Hoz Luna ,  Beda Christoph Hammerschmidt ,  George R. Krupka ,  Bo Xia ,  David Chih-Wei Jiang

IPC: G06F16/22 ,  G06F16/93

Abstract: Techniques for a unified relational database framework for hybrid vector search are provided. In one technique, multiple documents are accessed and a vector table and a text table are generated. For each accessed document, data within the document is converted to plaintext, multiple chunks are generated based on the plaintext, an embedding model generates a vector for each of the chunks, the vectors are stored in the vector table along with a document identifier that identifies the accessed document, tokens are generated based on the plaintext, the tokens are stored in the text table along with the document identifier. Such processing may be performed in a database system in response to a single database statement to create a hybrid index. In response to receiving a hybrid query, a vector query and a text query are generated and executed and the respective results may be combined.

公开(公告)号：US20250094318A1

公开(公告)日：2025-03-20

申请号：US18467187

申请日：2023-09-14

Applicant: Oracle International Corporation

Inventor： Timothy Clegg

IPC: G06F11/36 ,  G06F8/41

Abstract: Examples provide a computer system including an electronic processor configured to obtain a set of source code and a plurality of test scenarios. Each of the plurality of test scenarios specifies a respective build architecture. For each respective test scenario of the plurality of test scenarios, the electronic processor is configured to instantiate a respective build environment according to the respective build architecture, compile the set of source code in the respective build environment to generate a respective binary file, and generate a respective set of one or more metrics for the respective binary file.

公开(公告)号：US20250094223A1

公开(公告)日：2025-03-20

申请号：US18676248

申请日：2024-05-28

Applicant: Oracle International Corporation

Inventor： Ming Fang ,  Simo Lin ,  Jinguo Zhang ,  Wei Gao

IPC: G06F9/50 ,  H04L9/08

Abstract: A system and computer-implemented method include receiving a request for allocating graphical processing unit (GPU) resources for performing an operation. The request includes metadata identifying a client identifier (ID) associated with a client, throughput, and latency of the operation. A resource limit is determined for performing the operation based on the metadata. Attributes associated with each GPU resource of a plurality of GPU resources available for assignment are obtained. The attribute is analyzed that is associated with each GPU resource with respect to the resource limit. A set of GPU resources is identified from the plurality of GPU resources based on the analysis. A dedicated AI cluster is generated by patching the set of GPU resources within a single cluster. The dedicated AI cluster reserves a portion of a computation capacity of a computing system for a period of time and the dedicated AI cluster is allocated to the client associated with the client ID.

公开(公告)号：US20250094210A1

公开(公告)日：2025-03-20

申请号：US18632055

申请日：2024-04-10

Applicant: Oracle International Corporation

Inventor： Rondy C. Ng ,  Shyam Sundar Santhanam ,  Sundar Narayanan

IPC: G06F9/46

Abstract: A system accesses transaction data associated with a plurality of transactions, and based on characteristics of the transaction data, determines a set of functions to be applied to the transaction data at different corresponding levels of granularity. Determining the set of functions includes determining parallel processing requirements corresponding to the set of functions and determining an execution order corresponding to the set of functions based on the parallel processing requirements. The system schedules parallel execution of (a) a first function on the transaction data at a first level of granularity to generate a first dataset having the first level of granularity, and (b) a second function on the transaction data at a second level of granularity to generate a second dataset having the second level of granularity.

公开(公告)号：US20250094138A1

公开(公告)日：2025-03-20

申请号：US18743866

申请日：2024-06-14

Applicant: ORACLE INTERNATIONAL CORPORATION

Inventor： Yazhe HU ,  Mengqing GUO ,  Zheng WANG ,  Tao SHENG ,  Jun QIAN ,  Vinod MAMTANI

IPC: G06F8/35 ,  G06N3/0895

Abstract: Systems, methods, and other embodiments associated with automated fine-tuning of software code generation by large language models are described herein. In one embodiment, a method accesses a collection of software code samples that intermix sample code and human language description. The method generates prompts to an LLM to write code that performs as described by the human language description of the sample code. The method fine-tunes a large language model to generate software code based on a code generation loss function that evaluates code generated by the LLM in response to the prompts. The method generates an evaluation score for performance of the tuned large language model as a code generator based on code generation loss for second generated code. And, the method automatically signals that fine-tuning of the tuned large language is complete in response to the evaluation score satisfying a threshold.