53results about How to "Reduce visits" patented technology

The invention discloses an accelerator for accelerating the reasoning process of an unstructured sparse large language model and an operation method of the accelerator. The accelerator comprises a processor module, a direct memory access module, a memory controller module, a DDR memory module, an interconnection bus module, a calculation core module and a router module. The interconnection bus module is connected with the direct memory access module, the memory controller module and the computing core module; the processor module is connected with the direct memory access module; the DDR memory module is connected with the memory controller module; the router module is connected with the computing core module and the adjacent router module; and the calculation core module obtains to-be-calculated data from the off-chip DDR memory module through the interconnection bus module. According to the method, the limit that data reuse is carried out only along a single row or a single column on a traditional systolic array is broken through, cross-row and cross-column data reuse is achieved, the pressure of hardware equipment is relieved, the positions of non-zero elements are coded through bit masks, and the calculation efficiency is improved.

The invention relates to the field of artificial intelligence chips, in particular to an artificial intelligence chip integrated with an on-chip memory array, a computing task processing method and electronic device.The chip comprises a controller and a plurality of processor-memory modules, obtaining a plurality of sub-calculation tasks; the processor-memory module comprises a processing unit group, a memory and a scheduler, the processing unit group comprises a plurality of processing units, and the processing units process the sub-calculation tasks; the memory stores data related to the plurality of processing units in the processing unit group when executing the corresponding sub-calculation tasks; the scheduler is connected to the plurality of processing units and the memory and is used for scheduling access operation of the plurality of processing units to the memory, the processing unit group is arranged in a processing layer of the artificial intelligence chip, the memory is arranged in a storage layer of the artificial intelligence chip, and the processing unit group is used for processing the processing units. The processing layer and the storage layer are arranged in a stacked mode.

The invention discloses an automatic code vectorization optimization method and device and a medium, belongs to the technical field of high-performance computing, and aims at solving the problems that existing automatic code vectorization is inconvenient, the manual operation proportion is large, vectorization of irregular codes or codes with a load dependency relationship is difficult to support, and the vectorization efficiency is high. And improvement of calculation performance and reduction of energy consumption are not facilitated. The method comprises the steps of obtaining a scalar instruction diagram of a calculation task according to a calculation kernel of parallel calculation; performing protocol conversion processing on the scalar instruction diagram to obtain a protocol instruction diagram; according to the high-level organization structure of the instruction, grouping processing related to mapping vectors is carried out on the protocol instruction graph, and a grouping graph is obtained; based on the limited vector length of hardware, performing splitting matching processing on instruction vectors in the grouping graph to obtain a vector matching grouping graph; and executing configuration on each group of elements in the vector matching grouping graph to obtain a vector operation graph.

The neural network forward prediction hardware accelerator based on the programmable logic device can be applied to the technical field of neural network hardware acceleration. The accelerator comprises a multi-layer hardware accelerator module unit, and the hardware accelerator module unit comprises an input feature map cache module and a data processing module; the accelerator module units are connected in an assembly line structure mode; multi-channel feature maps can be processed in parallel; and the output data of the hardware accelerator module unit of each layer is stored in the input feature map cache module of the hardware accelerator module unit of the next layer for the operation of the next layer. By adopting an assembly line structure, the throughput rate of the hardware accelerator is improved; a multi-channel parallel processing mode is adopted, and the running speed of a hardware accelerator is increased; and on-chip storage is adopted, so that the off-chip access times are reduced, and the power consumption of the hardware accelerator is reduced.

The invention provides a reasoning calculation method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a target text to be reasoned through a large language model; the target text is preprocessed, and all first tokens in the target text are obtained; through a target data loading strategy obtained by limiting parameters of each second token in N second tokens which need to be loaded and are processed recently and uniformly storing row and column data of a feedforward neural network layer of the large language model, target parameters used for reasoning are loaded from a persistent storage medium storing the large language model; wherein the second token comprises the first token and / or a token obtained by reasoning based on the first token; and performing reasoning calculation on the integer ID through the target parameter to complete reasoning of the new second token. According to the invention, the problems of high data transmission cost and low reasoning speed in the prior art are solved.

The invention discloses a neural network accelerator based on eDRAM in-memory calculation and a control method thereof, the neural network accelerator comprises a top layer controller and an extensible and reconfigurable SR-CIM macro array, and the extensible and reconfigurable SR-CIM macro array is operated in a global system mode. The method comprises the following steps: acquiring weight data and input activation data; setting a working mode of the extensible and reconfigurable SR-CIM macro array; and according to the working mode of the extensible and reconfigurable SR-CIM macro array, performing operation on the weight data and the input activation data, and transmitting an operation result to the off-chip neural network. According to the embodiment of the invention, the computing efficiency of the neural network can be improved by improving the reuse rate of the data and reducing the access times of the off-chip data in different data flow modes. The method can be widely applied to the technical field of neural network accelerators.

The invention relates to the technical field of power acquisition terminal data storage, and discloses a data hybrid storage method and device and computer equipment, which are applied to a memory database, and the method comprises the following steps: when electric energy meter reading data is written into a main database of a power acquisition terminal, receiving a data change notification sent by the main database; recording the meter reading data into a data cache by adopting a key value mode based on the data change notification; and preferentially querying the corresponding meter reading data in the data cache based on a preset data retrieval requirement, and if the corresponding meter reading data is not queried, switching to the main database to continue data retrieval. According to the method, a large amount of data is prevented from simultaneously accessing the main database at the same moment, so that the load of the main database is reduced, load balancing is realized, the memory database is preferentially queried during data retrieval, and the main database is switched to retrieve when the memory database cannot be queried, so that the access frequency of the main database is greatly reduced, and the data retrieval efficiency is improved. And the performance bottleneck during batch data query of the relational database is solved.

The invention provides a multi-feature fusion-based black box attack detection method (MFBA), device and system, different features are extracted by using a multi-model fusion strategy, the effectiveness and quality of opposite samples are improved, and the MFBA comprises the following three key steps: firstly, obtaining an initial data set from a target model; a decision boundary is approached by using a multi-feature fusion model; secondly, the stability of the model is enhanced through multiple rounds of training application data expansion and incremental learning technologies; and finally, generating opposite examples by using the optimized fusion model. By constructing the comprehensive attack framework, how to combine various model features to improve the performance against the resistance attack is demonstrated; compared with a single substitution model, the antagonism examples generated by the MFBA have higher mobility under different disturbance levels and have better attack effects; according to the method, an effective solution is provided for black box antagonism attacks, and the model in the generation of the advantage opposite paradigm of multi-feature fusion is verified.

The invention provides a model reasoning method and device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence. Inputting the input vector into a target model to obtain a model reasoning result corresponding to the input vector output by the target model; wherein the target model comprises at least one hybrid expert module; the hybrid expert module comprises a time sequence feature extraction layer, a routing selection layer, an expert model layer and a result output layer which are connected in sequence; the expert model layer comprises a plurality of state space models; the input vector is generated based on at least one of text, speech, image, or video. According to the method and the device provided by the invention, the storage consumption of the large model is reduced, the access frequency of the large model to the storage is reduced, and the reasoning efficiency of the large model is improved.

The invention discloses a method for inversely sorting and assembling tree data, which relates to the field of software background development and comprises the following steps: S1, preparing; s2, a sorting step; s3, a cache establishment step; and S4, traversing and assembling. Compared with a traditional assembling mode of arranging from root nodes in sequence and recurring layer by layer, the method for assembling the tree structure data in the inverted sorting mode has the remarkable beneficial effects that the basic data are traversed in the inverted sorting mode, calculation is started from the last-stage node, the lower-stage nodes are set, and the efficiency is improved. According to the method, the complete tree structure can be assembled only through one-time circulation, the number of times and depth of recursion calling are greatly reduced, the time complexity problem (O (1)-O (n2)) caused by too deep recursion hierarchy and uncertain tree branch is avoided, and therefore the response time of a tree interface is remarkably shortened, and the overall running smoothness of software is improved.

The present disclosure provides a processing device, method and artificial intelligence chip for transposition operation, which relates to the field of artificial intelligence technology. The processing device for transposition operation includes K groups of registers, the kth group of registers includes M registers, and each register includes N storage units. The i-th register in the kth group of registers is configured to receive the i-th row element of the kth matrix to be transposed, and store the j-th element in the i-th row element in the j-th storage unit of the i-th register in the kth group of registers, the kth matrix to be transposed is an M×N matrix, and the N elements in the i-th row element are arranged in sequence from 1 to N; the M registers in the kth group of registers are configured to output the N column elements of the kth matrix to be transposed in sequence from 1 to N or from N to 1, and the j-th column element includes the j-th element in the j-th storage unit of each register in the kth group of registers.

The invention provides an image processing method and device based on element-by-element operation, equipment and a medium, and relates to the technical field of GPUs (Graphics Processing Unit), and the method comprises the following steps: according to the number of a plurality of elements contained in a plurality of input tensors needing to participate in operation and hardware resource information of GPU equipment, determining the number of the plurality of elements contained in the plurality of input tensors needing to participate in operation; determining the number of thread blocks required to be used by the plurality of input tensors in the element-by-element operation process; according to the number of the thread blocks, the number of the elements and the memory layout of the multiple elements, kernel functions of element-by-element operation class operators corresponding to the multiple input tensors needing to participate in operation are determined on the basis of a predefined template structure body, and parameters of the template structure body comprise element types and the number of the elements; and based on the number of the thread blocks, initiating a plurality of threads to enable the plurality of threads to obtain the plurality of elements according to the kernel function, and executing an element-by-element operation process of the plurality of input tensors to obtain an image processing result.

The present invention provides a multi-version concurrency control method for a database, which can be applied to the field of database technology. The multi-version concurrency control method for a database includes: determining the difference information of a target data row according to a first version and a second version of the target data row, wherein the second version of the target data row is obtained by modifying the first version through a first transaction; writing the difference information as the latest row version information into an uncommitted version data chain corresponding to the target data row, wherein the uncommitted version data chain includes at least one row version information, and the at least one row version information is obtained by executing a first transaction; upon receiving a feedback signal indicating that the first transaction is successfully committed, generating the committed row version information corresponding to the target data row according to the uncommitted version data chain, and writing the committed row version information into the committed row version chain corresponding to the target data row.

The invention provides a real-time attendance checking method and system based on cache preheating and hotspot peak clipping. The real-time attendance checking method based on cache preheating and hotspot peak clipping comprises the following steps: before a real-time attendance checking system is started or the concurrent access quantity exceeds a preset first concurrent access quantity threshold value, loading hotspot data into a cache region in advance; monitoring the concurrent access quantity of the attendance system in real time, when the concurrent access quantity exceeds a preset second concurrent access quantity threshold value, setting a target concurrent access processing quantity, and performing asynchronous processing on the access request by utilizing the target concurrent access processing quantity; and according to the data processing response duration offset coefficient corresponding to the employee card punching data processing response duration, determining whether data processing abnormity exists, and when the data processing abnormity exists, carrying out abnormity alarm. The system comprises modules corresponding to the steps of the method.

The application discloses a control method and device of a neural network execution, equipment and a storage medium, relates to the technical field of neural network inference, and comprises the following steps: acquiring an activated neuron set of a current neural network layer and corresponding weight data of the activated neuron set; performing calculation of the current neural network layer based on the activated neuron set and the weight data; in the process of performing calculation, determining an activated neuron set of a next neural network layer based on an intermediate result of calculation, and accordingly, pre-fetching weight data corresponding to the activated neuron set of the next neural network layer. According to the application, the activated neuron set of the next layer is determined in advance and the corresponding weight data is pre-fetched based on the intermediate result in the calculation process of the current layer, so that only the weight data related to the activated neuron needs to be loaded, the data access amount is reduced, and efficient inference is realized under the condition of limited bandwidth.

The invention discloses an API (Application Program Interface) request processing method, system and device and a storage medium, which are applied to a cache management center and are used for reducing the dependence on a server, reducing the delay of the server and improving the user experience. The method comprises the following steps: acquiring an API request initiated by a micro-service application, wherein the micro-service application is formed by aggregating a plurality of single applications; judging whether target data requested by the API request is cached or not; if the target data is cached, sending the target data to the micro-service application; if the API request is not cached, sending the API request to a server; and caching the target data returned by the server in response to the API request.

The invention relates to a quintuple matching method and device, computer equipment and a readable storage medium. The method comprises the steps that under the condition that a target quintuple of a to-be-processed message contains an internet protocol IP address, the IP address contained in the target quintuple is matched with a target IP filtering table to obtain an initial query address, the target IP filtering table stores the IP address associated with each IP matching rule table and the query address in a memory, and the initial query address is stored in the memory; the IP addresses of the quintuple matching rules associated with the same IP matching rule table are the same, determining a hierarchical query strategy of the target quintuple according to the quintuple information type contained in the target quintuple, and performing rule matching on the target quintuple by adopting the hierarchical query strategy and the initial query address. By adopting the method, the matching efficiency of the quintuple can be improved on the basis of reducing the access times of the memory.

The present invention discloses a multi-level flow table matching method and device based on Bloom filter. The method comprises the following steps: obtaining a data frame of a computer network data flow to be matched, extracting key fields from the data frame to obtain the value of the key field; inputting the values ​​of matching keywords corresponding to each level of the flow table into a preset counting Bloom filter in order of priority from high to low for detection, until the detection is passed; matching the values ​​of matching keywords corresponding to the flow table that passes the detection with a preset multi-bucket hash table based on DDR, and outputting the matching result; the device comprises a frame processing module, a serial matching module, a counting Bloom filter and a DDR memory; the present invention adopts a DDR memory to store the multi-bucket hash table, thereby increasing the flow table scale supported by the matching circuit; adopting a Bloom filter when matching flow table items, thereby reducing the number of accesses to the off-chip DDR storage and improving the matching speed of the data frame.

The present invention proposes a processor with pipeline processing design and an execution path prediction method. The processor with pipeline processing design includes a memory controller, an instruction cache module, a sending unit, a scheduler and an execution unit; the execution path prediction method includes: taking the execution of the instruction as the predicted satisfied condition of the conditional jump; when the instruction with a jump request is read, detecting whether the satisfied condition of the conditional jump corresponding to the conditional jump prediction instruction occurs; when the satisfied condition of the conditional jump does not occur, the pipeline continues to execute the next instruction; when the satisfied condition of the conditional jump occurs, the program pointer moves to the specified position requested by the satisfied conditional jump instruction, and the pipeline continues to execute the instruction from the specified position requested by the conditional jump instruction. The present invention can reduce the effect loss due to incorrect prediction by designing the cache column with a jump address, thereby improving the execution efficiency of the program.

The invention discloses a distributed system data access method and device, equipment and a storage medium, and relates to the technical field of data processing. The method comprises the following steps: determining target type data in a distributed system, and storing the target type data by utilizing a distributed cache database; the target type data is high-frequency access type data; creating a corresponding timed updating task for each piece of target type data, and adding a data acquisition method of each piece of target type data to a data updating service; when the timed updating task is executed, a corresponding data obtaining method in the data updating service is called to obtain latest data of the target type data, and the distributed cache database is updated according to the latest data; and querying the distributed cache database according to a data query instruction, and if a query result is null, querying a main database of the distributed system according to the query instruction. The data access efficiency of the distributed system can be improved.

The embodiment of the application belongs to the field of financial technology, and relates to a data batch request method and device, computer equipment and a storage medium. The method comprises the following steps: obtaining batch request parameters according to a batch request instruction; obtaining preset parameter saving information, saving the batch request parameters in a predetermined time period according to the parameter saving information to obtain effective request parameters; requesting cached data according to the batch request parameters, and detecting whether the return result of the cached data is empty; if the return result of the cached data is not empty, obtaining update information of the cached data, updating the cached data according to the update information to obtain batch request data; and if the return result of the cached data is empty, screening the effective request parameters according to the update information to obtain to-be-requested parameters, and sending the to-be-requested parameters to a database for data request to obtain the batch request data. The application can effectively improve the data reading efficiency when data is requested in batches.

The invention relates to an SQL (Structured Query Language) statement processing method and device, a medium, equipment and a product. The method comprises the following steps: acquiring a plurality of indexes under an index group; a syntax tree corresponding to each index in the index group is generated, the syntax tree is used for splitting the index to obtain a plurality of nodes corresponding to the index, and leaf nodes in the syntax tree are nodes which cannot be split; combining nodes in the syntax tree corresponding to the indexes to obtain a node relationship corresponding to the index groups; and generating SQL statements corresponding to the index groups according to the node relationships corresponding to the index groups.

The invention provides a model optimization and task processing method and device, electronic equipment and a storage medium. The model optimization method comprises the following steps: constructing a tree search graph according to a preset operator operation scheme of a preset task processing model; a current search node set is determined according to the current in-degree of each original node of the tree search graph, current degree expansion nodes of the tree search graph are constructed according to candidate operation schemes corresponding to each original node in the current search node set, and the candidate operation schemes comprise node merging operation and / or node concurrent operation; determining a local optimization node from the current degree expansion nodes according to the operation overhead of the global operation scheme corresponding to the current degree expansion nodes; determining a global optimization node according to a local optimization node determined from each degree expansion node; and optimizing a preset operator operation scheme according to a global operation scheme corresponding to the global optimization node. The optimization method is high in universality, simple and easy to implement, and hardware resources of equipment deployed by the model can be fully utilized.

The invention discloses a data security processing method, a client, electronic equipment and a storage medium, and relates to the technical field of information security. The method comprises the steps that a file uploading request of an application program is intercepted through an adaptation library of a user mode file system; obtaining a physically isolated master key in the hardware cryptographic device, and deriving a unique file encryption key in combination with the target file information; calling a national secret algorithm library, performing symmetric encryption on file data by using the file encryption key, and performing asymmetric encryption on the file encryption key by using a public key; and generating an encrypted file header containing the encryption parameter and the ciphertext file key, and uploading and storing the encrypted file header. Through deep fusion of the hardware cryptographic device and the user mode file system, hardware-level physical isolation protection of the key, national cryptographic algorithm compliance support and transparent encryption and decryption are realized, key leakage is effectively prevented, and data security is improved.

The utility model discloses a data acquisition communication recorder which is integrated with a GPS positioner and a 4G module, the front end face of a shell is provided with a 4G antenna interface, and the 4G module is connected with a mobile communication antenna through the 4G antenna interface, so that the data acquisition communication recorder can acquire accurate position information of a ship in real time through the GPS positioner and the 4G module. The accuracy of a navigation route is ensured, direction losing or entering a dangerous area can be avoided, the problem of incomplete data acquisition caused by equipment position change in a navigation environment is solved, the limitation of a traditional data recorder is broken through, data acquisition and real-time reporting in a dynamic environment are realized, and meanwhile, the accuracy of the navigation route is improved. The data acquisition communication recorder can be connected to a mobile communication network through the arrangement of the 4G module and the 4G antenna interface, and real-time data transmission and communication between the data acquisition communication recorder and a remote monitoring center or other ships are achieved.

The present invention proposes a real-time attendance method and system based on cache preheating and hotspot peak shaving. The real-time attendance method based on cache preheating and hotspot peak shaving includes: before the real-time attendance system is started or the number of concurrent accesses exceeds the preset first concurrent access number threshold, pre-loading the hotspot data into the cache area; real-time monitoring of the number of concurrent accesses to the attendance system, when the number of concurrent accesses exceeds the preset second concurrent access number threshold, setting the target concurrent access processing number, and using the target concurrent access processing number to asynchronously process the access request; judging whether there is a data processing anomaly based on the data processing response time offset coefficient corresponding to the employee punch-in data processing response time, and issuing an anomaly alarm when there is a data processing anomaly. The system includes modules corresponding to the steps of the method.

This application discloses a method, apparatus, and system for email association processing, solving the technical problem that existing discrete storage of email data leads to multiple independent queries of associated emails and redundant memory space usage. The method includes: receiving first data for a first email, including a first serial number identifying the first email and its sender, and at least one second serial number identifying the first email and its recipient; receiving second data for a second email, including a third serial number identifying the second email and its sender, and at least one fourth serial number identifying the second email and its recipient; determining a target serial number from the first data; the target serial number being a second serial number corresponding to either the sender's identifier or the recipient's identifier; and writing the target serial number into the second data. This application establishes data links between emails, converting multiple random I / O operations into sequential I / O operations, reducing processor load and memory access frequency.

The invention discloses a block chain-based cross-chain data retrieval system, which comprises a data acquisition module, a data transmission module, a data storage module and a user terminal module, and is characterized in that the data acquisition module is used for acquiring tracing data of crops; the data transmission module is used for transmitting the data acquired in the data acquisition module to the data storage module; the data storage module is used for storing the traceability two-dimensional code information and the traceability data by adopting a block chain node; and the user terminal module is used for querying the traceability data stored in the data storage module according to the traceability two-dimensional code information. The system has the beneficial effects that the crop classification processing unit can perform classification processing on crops according to purposes, and the crop classification numbering unit performs unique numbering on the classified crops for avoiding unnecessary retrieval steps during subsequent retrieval, so that the retrieval speed is improved.

The invention relates to the technical field of video processing, and provides a two-dimensional video processor, a video processing system, a video processing method and a program product, the two-dimensional video processor comprises more than two video processing sub-cores, each video processing sub-core comprises a plurality of image processing modules, and each image processing module is used for executing an image processing operation; when more than two video processing sub-cores are cascaded, the first video processing sub-core is used for reading a first video frame from the memory; the image processing modules in different video processing sub-cores are used for executing more than two image processing operations on the first video frame to obtain a processed video frame; and the at least one second video processing sub-core is used for writing the processed video frame into the memory, so that the number of access times to the memory during video frame processing can be effectively reduced, the video frame processing rate is accelerated, and the power consumption is reduced.

The invention discloses an intelligent auxiliary system based on an instrument sharing platform in the technical field of information, and the system comprises an information collection module which is used for collecting basic information of an instrument sharing platform user and instrument use information; the database module is used for establishing a basic database based on the information acquired by the information acquisition module and managing and optimizing the information of the basic database; the statistical analysis module is used for performing statistical analysis on instrument use condition information and fund account information according to the collected instrument use information; the screening and error correction module is used for carrying out screening and error correction on the instrument use condition information according to the basic database; and the reporting module is used for forming a report form according to the information completing screening and error correction for reporting. The system is intelligent and efficient, the MATLAB is used for constructing a database to count and analyze the instrument use condition information and the required fund account information, intelligent data screening, error correction and reporting are carried out, the working efficiency of statistical analysis is improved, and orderly and efficient operation of an experiment teaching platform is guaranteed.