Zilliz Inc.

Starlord (cayman) Limited

Chao Xie

Li Liu

Xiaofan Luan

Chao Gao

Xiaomeng Yi

Rentong Guo

Xun Huang

Xiangyu Wang

Xuan Yang

Weizhi Xu

Songlin Wu

Yuchen Gao

Bolong Zheng

Mengzhao Wang

Ziyang Yue

Yu Xie

Bang Fu

Qi Hu

Min Tian

Jie Hou

Qianya Cheng

Enwei Jiao

Anyang Wang

The method is applied to a question-answering system including large language model. The question-answering system can store history query information to vector database, and store history query results corresponding to each history query information to scalar database. When receiving the current query request, the question-answering system may first match the current query request with the history query information. If the history query information contains target history query information matching the current query request, the query result of the current query request can be obtained based on the target history query result corresponding to the target history query information. Hence, the question-answering system does not need to obtain the query result by LLM inferring, which is beneficial to reducing the delay of obtaining the query result by the question-answering system and reducing the operation cost of the question-answering system.

Apparatus, method and storage medium for data query

Vector dataset index parameter determination is performed by building an index of a vector dataset according to each build parameter group among a plurality of build parameter groups, generating search parameter groups, submitting each vector query in a vector query set to each index according to the candidate search parameter values from each search parameter group to obtain, for each submission, a result content and a result delay value, obtaining, for each submission, a vector distance between at least one result vector of the result content and the submitted vector query, determining a relative accuracy value for each submission by comparing the vector distance of the result content to the vector distance of at least one other submission, correlating build parameter values and search parameter values with the relative accuracy values and the result delay values, and fixing build parameter values based on the correlating.

Vector dataset index parameter determination

Vector dataset index parameter determination is performed by obtaining a statistical variance of each vector dataset among a plurality of vector datasets, grouping the plurality of vector datasets into a plurality of vector dataset groups based on the statistical variance, determining, for each vector dataset group, build parameter values, obtaining the statistical variance of a target vector dataset, and applying, to the target vector dataset, the determined build parameter values corresponding to a vector dataset group selected based on the statistical variance.

A data query apparatus, method, and storage medium include a database system that determines a data collection and a calculating model corresponding to a received query task request, according to the number of bytes occupied when the data collection or the calculating model is stored, or the number of bit widths occupied when the data collection or the calculating model is transmitted, the database system determines a transmission rate of the data collection or the calculating model. The query task request is transmitted to a model node or a data node with a lower transmission rate, and the data node or the model node with a higher transmission rate is dispatched to perform data calculation.

Data query apparatus, method and storage medium

The present disclosure relates to the field of computers, and provides an apparatus, a method, and a storage medium for database query, which are applied to a vector database, by acquiring a query request from a user, determining query parameters corresponding to each of multiple shards based on the query request using a preset neural network model, wherein the query parameters control query complexity by affecting a query range of database data of the query request for the corresponding shard; querying database data in each shard based on the query request and the query parameters to obtain a target query result. Accordingly, corresponding query parameters are adjusted for each shard, so that redundant queries are effectively avoided, and database performance is effectively improved.

Apparatus, method and storage medium for database query

An apparatus, a method, and a storage medium for database pagination. The apparatus, method, and storage medium are configured to perform following processes, including: acquiring a pagination request from a user; determining a query parameter based on the pagination request; determining a corresponding pagination pattern and a request data amount based on the query parameter; determining a target query mode based on the request data amount; generating a query task based on the pagination pattern and the target query mode; obtaining at least one first query result based on the query task; determining a target result set based on the at least one first query result; and paging the target result set based on the pagination pattern and the target query mode to obtain multiple pages of pagination data.

Apparatus, method and storage medium for database pagination

A database system that, after receiving the query request, adopts a pre-trained deep learning model to predict probing cardinality corresponding to the query request based on the query vector in the query request, the number of vectors of the query result, and the vector corresponding to the distances between the query vector and the center vector of a plurality of data partitions stored in the storage device. The system then determines the target data partitions according to the probing cardinality, and obtains the query result from the target data partitions. Since the probing cardinality is dynamically determined based on the query request, decreases to both the query efficiency due to a larger setting of the probing cardinality, and the query recall due to a smaller setting of the probing cardinality are avoided, which is beneficial in reducing the average number of target data partitions and improving query efficiency.

Data query apparatus, method, and storage medium

An apparatus, a method, and a storage medium for database query. The apparatus, method, and storage medium are configured to: determine a syntax tree corresponding to an SQL query statement includes a preset subtree, wherein the preset subtree is used to indicate a query mode for querying vector data; determine a query mode according to the preset subtree and the SQL query statement; query data in a database based on the query mode to determine query results. By using an SQL query statement including a preset subtree to query a vector database, to realize an efficient approximate query, so as to avoid high access costs caused by accessing the database by using an API. Moreover, it can reduce user's understanding costs on unstructured query.

A retrieval apparatus, a method, and a storage medium are disclosed. The retrieval apparatus includes a memory and a processor, wherein the processor is configured to acquire a retrieval request, and the retrieval request includes the query vector; according to the query vector, generate and execute a first access request corresponding to the first memory according to the first graph index and the index node in the candidate pool; determine and process the data acquired by the first access request corresponding to the first memory, corresponding to the absence of redundant data in the previous storage pool, and store the result in the result pool; and output the data in the result pool corresponding to the candidate pool does not include unreachable index nodes. The retrieval apparatus improves the efficiency of the data reading process and further improves the efficiency of the retrieval process.

Retrieval apparatus, methods, and storage medium

Vector dataset index parameter determination is performed by building an index of a vector dataset according to each build parameter group among a plurality of build parameter groups, generating search parameter groups, submitting each vector query in a vector query set to each index to obtain, for each vector query, a result content and a result delay value, wherein an accuracy value is determined based on the result content, correlating build parameter values and search parameter values with accuracy values and result delay values, fixing build parameter values based on the correlating, and determining a mathematical relationship expressing a correlation of the search depth parameter values, the numbers of requested results, and the accuracy values obtained in response to vector queries submitted to the index having the fixed build parameter values.

A data backup method, a data recovery method, and an electronic device are disclosed. The data backup method includes: determining a data set to be backed up, performing a status update operation on a data block in the data set, determining the time points when the data block changes from the first phase to the second phase, and when the data block changes from the second phase to the third phase, and performing data backup on the data set to generate backup files. With this method, upon data seal time point, no fresh data can be reached in any data block, and the data block back-up persists, so that the time point of the data backup operation is more accurate, the cut point of which deleted data should be included in the back up is easy to locate, and the deleted data may be accurately processed.

Growing community of inventors

Redwood City, CA, United States of America

Chao Xie