26results about How to "Reduce memory access" 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 discloses a method for identifying and counting multiphase flow bubbles based on improved YOLOv8. The method comprises the following steps: acquiring a multiphase flow bubble image data set; dividing the multiphase flow bubble image data set into a training set and a verification set; yOLOv8 is improved to obtain an improved YOLOv8 network model, a counting module is added in the improved YOLOv8 network model, and a multiphase flow bubble recognition and counting model based on the improved YOLOv8 is obtained; training through the training set to obtain an optimal bubble recognition and counting model; and performing performance evaluation on the optimal bubble identification and counting model through the verification set, and when the performance evaluation is passed, performing identification and counting of multiphase flow bubbles through the optimal bubble identification and counting model. According to the method, under the condition that the model parameter quantity and the calculation quantity are greatly reduced, the bubble detection effect with high operation precision and high speed can be achieved, and equipment deployment and application are more convenient.

The invention discloses a lightweight fabric surface defect detection method and device, and relates to the technical field of target detection. The method comprises the following steps of: performing data preprocessing and division on a pre-acquired fabric image data set; secondly, the model is improved, down-sampling Conv in a backbone and a check is replaced with ADown, StarBlock is introduced into a backbone part C2f to form a C2fStar module, and PCDeect is introduced into a head part of a detection head; the fabric defect picture data set is trained through the built YOLOv8 + Aown + C2fStart + PCDeect detection network, and a model pre-training weight is obtained; inputting a to-be-detected fabric picture into the optimal weight model for fabric defect detection, and outputting a detection result; the device comprises a processor, and when the processor runs the computer program, the steps of the lightweight fabric defect detection method are executed. According to the method, the parameter quantity and the calculation quantity are effectively reduced in a fabric defect detection task, meanwhile, the industrial defect detection requirement can be met, and the model precision is improved.

The invention discloses a large model reasoning method and device, equipment and a program product, and relates to the technical field of artificial intelligence, and the method comprises the steps: selecting a sparse large model as a to-be-deployed large model during deployment, further splitting the large model into a plurality of sub-models, and storing the sub-models in a memory of a terminal; when the large model reasoning task is executed, only the target sub-model can be loaded from the memory to the reasoning chip to execute the reasoning task according to the target sub-model required by the current reasoning task, and all sparse large models do not need to be loaded, so that the number of model parameters required to be loaded is greatly reduced, the access stock is reduced, the reasoning time is also reduced, and the reasoning efficiency is improved. And the memory wall problem is effectively relieved.

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 invention belongs to the technical field of wood defect detection, and particularly relates to a wood surface defect detection method based on improved YOLOv81, electronic equipment and a storage medium. The method comprises the following steps of: improving the Backbone of the traditional YOLOv81 by using Faster Net, and reducing the parameter quantity and the calculation quantity by using the design of partial convolution PConv in the Faster Net; an enhanced weighted bidirectional feature pyramid network EFPN is provided, information exchange between different feature layers is enhanced through an auxiliary branch, and an expansion re-parameter module DRM is designed to fully capture features on different scales. The improved YOLOv81 is more suitable for detecting the multi-scale defects on the surface of the wood, the detection precision is improved, the parameter quantity and the calculated quantity are smaller, and an efficient and accurate solution is provided for detecting the defects on the surface of the wood.

The invention relates to a data reading and writing method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: receiving a data reading request; if the resource type corresponding to the data reading request is a pixel buffer object, aligning the horizontal offset of the pixel buffer object with a target buffer position to obtain a first mapping logic address; and converting the first mapping logic address into a first physical address, reading first target data from the pixel buffer object according to the first physical address, and storing the first target data to the target cache. By adopting the method, the data reading efficiency can be improved.

The invention provides an eDRAM-based high-density near-storage calculation and in-storage calculation hybrid architecture and a calculation method. The architecture comprises an eDRAM near-storage and storage calculation hybrid array, a sensitive amplifier, an ADC circuit, a storage calculation result register, a characteristic value updating drive circuit, a controller, a row decoder and a column decoder. According to the calculation method, the architecture is adopted for calculation. By combining a near memory and an in-memory calculation structure, the characteristics of the near memory and the in-memory calculation structure are fully matched to carry out optimization design on an array structure and a basic unit and complement the independent defects of the two architectures, meanwhile, a deep groove capacitor is adopted to replace a plane capacitor, and the advantages of low power consumption, low time delay, high calculation efficiency, high density and the like are achieved; the bandwidth limitation of a traditional near storage circuit and the process limitation of an in-storage computing circuit are effectively overcome, and the method can be applied to a deep learning accelerator computing core and is particularly suitable for edge end hardware deployment with high requirements for the integration level and the power consumption performance.

The invention discloses an intelligent driving front-view image perception algorithm based on heterogeneous convolution, and the algorithm comprises the steps: S1, carrying out the basic feature extraction of an original image through an initial convolution layer, and enabling the original image to enter a plurality of parallel convolution units which are integrated together; s2, performing feature fusion on the output of the plurality of parallel convolution units in a weighted fusion mode; s3, stacking a plurality of parallel convolution units into a backbone network, and performing multi-scale target detection through an FPN feature pyramid; and S4, after the detection of the targets of multiple scales is finished, completing the training of the intelligent driving front-view image perception algorithm model. According to the intelligent driving front-view image perception algorithm based on heterogeneous convolution, the reasoning efficiency and the feature extraction capability of the model can be remarkably improved, and the balance problem of the lightweight target detection model between the calculation complexity and the feature extraction capability is solved.

The invention relates to the technical field of data processing, in particular to a multi-dimensional data processing method and device, equipment and a medium, which are used for performing instantiation processing on a template according to an input tensor parameter to obtain an offset calculation instance. And performing initialization operation on the offset calculation instance to determine the step length and the shape of each input array on each output dimension, and recording the step length and the shape in a set storage space. And starting an offset calculation instance to carry out index calculation on the step length and the shape of each input array recorded in the storage space on each output dimension so as to obtain the offset corresponding to the output array. And calculating the data corresponding to the offset according to a set calculation rule to obtain an output result. By pre-calculating the step length and shape of each input array on each output dimension, unnecessary memory access can be reduced. The index calculation process is simplified by constructing the offset calculation instance, and data access and calculation can be efficiently and accurately achieved.

The invention discloses a global nerve drawing method and system based on a programmable rasterization engine, and belongs to the technical field of computer graphics, and the method comprises the steps: at the programmable rasterization engine, analyzing a rasterization descriptor according to a rasterization instruction, and extracting vector microoperation and control parameters; maintaining a task state machine according to the parameters and distributing a control signal, selecting an execution entry from a vector kernel table according to the control signal, and instantiating an operation into a parallel vector thread; in a vector thread execution process, tracking data dependence of a vector register and a synchronization state of a direct memory access unit, executing vector loading / storage operation so as to carry data between the register and an on-chip shared memory according to the data dependence and the synchronization state, and dynamically scheduling vector micro-operation to an execution component so as to complete rasterization calculation; and outputting a result to the neural rendering network to complete global neural rendering. According to the method, the multi-representation neural rendering load can be uniformly and efficiently supported on the AI accelerator, the memory access overhead is remarkably reduced, and the calculation efficiency is improved.

The invention discloses a distributed street lamp control method and system for urban street lamps, and relates to the field of Internet of Things. The method is applied to a central server, and comprises the following steps: when a load value of a first agent node reaches a threshold value, obtaining an assistance request sent by the first agent node; based on the assistance request, evaluating the computing power resource demand quantity of the first agent node; querying load information of a plurality of idle agent nodes; according to the load information of the plurality of idle agent nodes and the computing power resource demand quantity of the first agent node, service information of the first agent node is sent to the plurality of idle agent nodes for processing, and a plurality of processing results are obtained; and returning the plurality of processing results to the first agent node, so that the first agent node controls the operation states of the plurality of city street lamps according to the plurality of processing results. By implementing the technical scheme provided by the invention, the overload problem caused by excessive data volume of the urban street lamp agent node is solved.

The invention provides a multi-dimensional self-adaptive large model inference engine method and system, and relates to the technical field of artificial intelligence, and the method comprises the steps: collecting hardware identification information of calculation hardware, and calling a hardware fingerprint database for matching to generate a hardware type result and a hardware characteristic parameter; constructing a hardware abstraction layer, loading a hardware abstraction driving plug-in according to a hardware type result, and virtualizing computing hardware into a unified logic computing power unit through a hardware abstraction interface; when a large model reasoning task is triggered, dynamically routing to a back-end executor according to a hardware type result by querying a back-end registry; on the basis of hardware characteristic parameters and a hardware abstract interface, a hardware compiling tool chain is called through a rear-end executor, graph optimization and operator fusion are executed on a large model calculation graph, and therefore self-adaptive deployment and efficient reasoning of a large model on heterogeneous hardware can be achieved; the problems that in the prior art, heterogeneous hardware is insufficient in compatibility, configuration is tedious and rigid, and compilation optimization barriers and performance losses are caused during model cross-platform migration are solved.

The present disclosure discloses a computing device, a method for performing convolution operation by using the computing device, and related products. The computing device can be included in a combined processing device, which can further include an interface device and other processing devices. The computing device interacts with the other processing devices to jointly complete a user-specified computing operation. The combined processing device can further include a storage device connected to the computing device and the other processing devices respectively, for storing data of the computing device and the other processing devices. The scheme of the present disclosure optimizes the convolution operation and improves the operation processing efficiency.

The application provides an AI device and an electronic equipment based on sparsity, the AI device comprises a first storage module for storing parameters of a large model, a second storage module for storing intermediate data, and a calculation module, the calculation module comprises a neural network accelerator (NPU) based on in-memory calculation, the neural network accelerator (NPU) comprises an in-memory calculation matrix, the in-memory calculation matrix is used for neural network calculation according to input data and data from the first storage module and the second storage module; the calculation module is further used for predicting neuron parameters to be activated in the parameters of the large model in the first storage module according to the input data and a sparsity prediction model, the calculation module reads the predicted neuron parameters to be activated from the first storage module according to the prediction result, and the calculation module performs neural network calculation according to the read neuron parameters and the input data. The AI device provided by the application can better solve the problems of computing power, energy efficiency, heat dissipation, storage capacity and bandwidth.

The invention discloses a post-quantum digital signature key generation method based on AVX2 acceleration and matrix multiplication optimization, which comprises the following contents: on the basis of original coding, parallel optimization of signature compression and decompression processes is realized by using an AVX2 instruction set, the coding and decoding efficiency is remarkably improved, and the method is particularly suitable for a large-scale data processing scene. A matrix multiplication optimization method for an algorithm is provided, and recursion operation is avoided by only performing matrix partitioning once. And in combination with a register optimization technology, the memory access is effectively reduced while the calculation accuracy is ensured, and the calculation efficiency in the signature generation process is remarkably improved. A register optimization strategy is introduced in matrix multiplication and addition operation, and a calculation result is stored in a register as much as possible by reducing memory access operation, so that the number of times of writing back to a memory is reduced, the execution efficiency of matrix operation is further improved, and the method has important significance on efficient implementation of a post-quantum digital signature scheme.

The invention provides a corn early leaf recognition method, device and equipment and a storage medium, and belongs to the technical field of image recognition, and the method comprises the steps: integrating a FasterNet module in a C2F module of a backbone network of a YOLOv8 model, the FasterNet module comprising a partial convolution module, integrating an LSKNet network in a neck network, constructing an improved YOLOv8 model, and through a historical image data set comprising corn early leaves, carrying out the recognition of the corn early leaves. Training the improved YOLOv8 model to obtain a target detection network; and identifying the early leaf of the corn through the target detection network. According to the method, the Faster Net module is integrated in the C2F module of the backbone network of the YOLOv8 model, the LSKNet network is integrated in the neck network, and the backbone network and the neck network of the original YOLOv8 model are improved to identify the early corn leaves, so that the accuracy of identifying the early corn leaves is improved.

A traffic tracking and detection system for drones. It belongs to the field of target tracking technology, and specifically to the field of lightweight target detection and tracking technology for intelligent transportation and drones. The system includes a target detection module, a target motion trajectory matching module, and a target motion trajectory output module; the target detection module processes the video frame image to obtain the detection frame corresponding to the tracked target and the confidence score corresponding to each detection frame; the target motion trajectory matching module divides the detection frame according to the confidence score, and matches the detection frame with the predicted trajectory after division; the target motion trajectory output module outputs the predicted trajectory as a confirmed trajectory after the same frame detection frame is successfully matched with the predicted trajectory for three consecutive times in the target motion trajectory matching module. It solves the problem of low tracking accuracy caused by the existence of small-sized targets, complex backgrounds, and occlusion in the previous drone tracking of targets.

The present invention relates to the field of image processing technology, and in particular to an improved method for license plate detection and recognition under high-position video, comprising: first, using a high-definition smart camera to obtain image information of a parking section, using the LPD‑YOLOv8 algorithm to process the image information, and preliminarily detecting and selecting the license plate area; secondly, performing image preprocessing on the extracted license plate area; finally, using the improved LPR‑NET algorithm to recognize the corrected license plate image, and uploading the recognized license plate image information to the background. The method proposed in the present invention has a lightweight architecture and a novel method, and can be deployed with only one high-definition smart camera. Compared with traditional methods, this method has achieved significant improvements in detection speed and recognition accuracy, providing an efficient and accurate license plate recognition solution for roadside parking management on urban roads.

The embodiment of the present invention discloses a hybrid expert model reasoning acceleration method, device, equipment, medium and program, wherein the method includes: loading the target association parameters of the hybrid expert model; wherein the target association parameters include the non-expert network parameters and the expert network reference vector of the hybrid expert model; predicting the target expert network of the network layer to be run according to the current layer output feature vector of the current running network layer of the hybrid expert model and the expert network reference vector of the network layer to be run; preloading the target expert network of the network layer to be run to the device storage running the hybrid expert model to perform the model reasoning process. The technical solution of the embodiment of the present invention can reduce the runtime behaviors such as data movement and memory access of the hybrid expert model, save memory usage, and improve the computing performance and operation speed of the model, thereby improving the utilization of hardware resources and the execution efficiency of the model.

This application provides a GPU-based sparse matrix-vector multiplication optimization method and apparatus for large-scale computing. The method includes rearranging and reorganizing the sparse matrix to make the distribution of non-zero elements in adjacent columns of the rearranged sparse matrix more similar; dividing the non-zero elements in each row of the sorted sparse matrix into several slices, placing the slices into containers that satisfy constraints, namely, the starting column index of the newly added slice is consecutive to the column indices of other existing slices; each container corresponds to a synchronous execution unit, and several synchronous threads perform the multiplication and accumulation of non-zero elements with the corresponding components of the input vector. Thus, matrix rearrangement makes it more likely that a synchronous execution unit will process data from the same row; by dividing non-zero elements with the same row index and consecutive column indices into slices and placing them into containers, the continuity of data processed by the synchronous execution unit in the column direction reduces non-merged memory accesses and enhances data reuse capabilities.

The invention discloses a low-delay streaming voice wake-up method and device for an embedded terminal, is applied to the technical field of data processing, and adopts a GRU as a backbone architecture for the requirements of low power consumption and high real-time performance of voice wake-up of the embedded terminal. The original audio is segmented and reconstructed through time domain lightweight preprocessing, and the FFT overhead is eliminated through linear layer mapping; building a multi-layer GRU full-causal streaming coding network, wherein no sliding window cache exists, and the memory bandwidth occupies O (1); outputting Token probability distribution containing Blank through a CTCLoss training module, and directionally extracting a target Token probability; constructing a target index sequence based on the target wake-up word and a CTC protocol, and setting three types of state transition rules to generate anti-interference decoding information; and a non-Top-K pruning decoding mechanism is adopted, an attenuation factor and threshold judgment is combined, a judgment result with constant calculation complexity and low false and missing wakeup is generated, and subsequent interaction is triggered if the judgment result succeeds.

The application provides a point cloud data processing method, system and device based on spatial serialization, the method comprising: sorting point cloud data by using a six-way standard sequence sorting method to obtain point cloud data sequences; dynamically and automatically selecting focal point blocks in the sequenced data according to a view cone and screen projection error, and dynamically adjusting the LOD level of each focal point block to obtain sequenced LOD data; after the sequenced LOD data is managed by VIBO block management, performing triangle optimization reconstruction; using a UV space remapping algorithm, performing patch and texture alignment processing on the point cloud surface after triangle optimization reconstruction, and performing multi-format texture packing to complete rendering output. The application reduces unnecessary calculation and memory access through dynamic adjustment of the LOD of the focal point block and VIBO block caching. Through dynamic LOD adjustment and optimized triangle reconstruction technology, combined with efficient texture packing and decompression, the rendering frame rate can be significantly improved while details are maintained.

The present invention discloses a method for identifying and counting multiphase flow bubbles based on improved YOLOv8, comprising the following steps: obtaining a multiphase flow bubble image data set; dividing the multiphase flow bubble image data set into a training set and a verification set; improving YOLOv8 to obtain an improved YOLOv8 network model, and adding a counting module in the improved YOLOv8 network model to obtain a multiphase flow bubble identification and counting model based on improved YOLOv8; training through the training set to obtain an optimal bubble identification and counting model; performing performance evaluation on the optimal bubble identification and counting model through the verification set, and when the performance evaluation passes, identifying and counting multiphase flow bubbles through the optimal bubble identification and counting model. The present invention can achieve a bubble detection effect with high calculation accuracy and high speed while greatly reducing the amount of model parameters and the amount of calculation, and is more convenient for equipment deployment and application.

The invention discloses a hot rolled steel surface defect automatic detection system based on improved YOLOv8m. The system comprises an industrial personal computer, a PLC, a first sliding rail, a second sliding rail, a hot rolled steel detection table, a line scanning camera and a spraying machine. The line scanning camera moves forwards along with the first sliding rail, and meanwhile an image collected by the line scanning camera in real time is detected by calling a hot rolled steel surface defect detection algorithm through the industrial personal computer; if defects are detected, defect images and data are displayed in hot rolled steel surface defect detection software, meanwhile, defect position information is sent to a lower computer PLC through network communication, at the moment, the PLC recognizes the data, then a second sliding rail and an automatic spraying machine switch are controlled to conduct defect identification spraying, and then round-trip circulation detection is conducted. According to the detection system and the detection method, hot rolled steel defect detection and automatic mark spraying auxiliary work can be achieved, subjectivity of human eye detection is reduced, and detection efficiency and accuracy are improved.

The present disclosure discloses a computing device, a method for performing convolution operation by using the computing device, and related products. The computing device can be included in a combined processing device, which can further include an interface device and other processing devices. The computing device interacts with the other processing devices to jointly complete a user-specified computing operation. The combined processing device can further include a storage device connected to the computing device and the other processing devices respectively, for storing data of the computing device and the other processing devices. The scheme of the present disclosure optimizes the convolution operation and improves the operation processing efficiency.