1300 results about "Computation process" patented technology

The invention discloses a large model reasoning efficiency dynamic optimization and hardware sensing compression method. The method comprises the following five steps: S1, generating an input complexity signal representing calculation complexity; s2, synchronously monitoring a hardware resource index of the operation platform, and generating a hardware state signal reflecting a real-time load; s3, inputting the input complexity signal and the hardware state signal into a dynamic strategy selector, and generating a compression control signal through a pre-trained decision model; s4, according to the compression control signal, dynamic reconfiguration operation is executed on the large model weight and the activation value of the current reasoning task; and S5, performing reasoning calculation by using the reconfigured large model, and feeding back a hardware resource index to the step S2 in real time in the calculation process to form a closed-loop optimization link. According to the large model reasoning efficiency dynamic optimization and hardware perception compression method, the problems of low resource utilization rate, delay fluctuation and energy efficiency imbalance caused by a static compression method in dynamic input and heterogeneous hardware environments can be solved.

The invention discloses a visual language model illusion suppression method based on self-adaptive dynamic attention intervention. The method is used for reducing the problem that a visual language model generates wrong associated information. The method comprises the steps that text input, visual input and historical response are acquired, and an attention accumulation vector is initialized; in the layer-by-layer calculation process of the language model, dynamically adjusting a non-normalized attention matrix, enhancing the weight of a visual sensitive attention head, and performing visual Token pruning in a deep network to optimize cross-modal information interaction; and finally, generating an output Token based on the adjusted attention mechanism, and carrying out loop iteration until a complete response is generated. According to the method, a method of combining text deviation correction through self-adaptive attention head modification and visual attention convergence-based Token pruning is adopted, so that the performance of the model in a multi-modal task is remarkably improved, and the illusion phenomenon caused by a language modal dominant reasoning process is effectively relieved.

The invention discloses a large-model-driven intelligent calculation method and system for a water conservancy mechanism model. According to the method, natural language input, structured conversion and intelligent optimization calculation of a scheduling target are realized by integrating a field-enhanced large language model and a water conservancy professional mechanism model. The method comprises the steps of receiving a calculation target expressed by a user in a natural language, and analyzing and converting the calculation target into a constraint condition and a target function which can be recognized by a water conservancy mechanism model; a hydrological model, a hydraulic model, a hydrodynamic model and other models are called based on a workflow engine, and reverse calculation is carried out by adopting a hybrid optimization strategy of'coarse adjustment-fine adjustment-verification '; synchronously and visually displaying the parameter change and the result convergence state in the calculation process; and outputting a calculation result including parameter adjustment logic, standard conformity analysis and multi-scheme comparison. The system comprises a natural language interaction module, a target conversion module, an intelligent calculation engine module, a visualization module and a result generation module, and supports multiple application scenes such as multi-target scheduling, emergency decision making and ecological guarantee. Compared with a traditional scheme, the method has the advantages that the model use threshold is lowered, the dispatching efficiency and calculation transparency are improved, and the method is suitable for complex hydraulic engineering calculation tasks such as reservoir dispatching, cross-basin water transfer and flood control emergency.

The invention discloses a time series anomaly detection method based on combination of hierarchical adaptive attention and Mamba. According to the method, a multi-granularity token routing strategy is provided, the strategy dynamically allocates computing resources in a time context, adaptively concentrates processing capacity on an information segment, and keeps wider perception at the same time, so that attention computing can be dynamically focused on different time scales and modes according to the complexity of input data; according to the method, the Mama is improved, so that parameters of the Mama can be dynamically adjusted according to characteristics of an input sequence, the long-distance dependency relationship is effectively simulated, and meanwhile, the modeling capability of a nonlinear time mode is enhanced. The abnormal score calculation process comprises three stages: reconstruction error calculation, error normalization and hierarchical score fusion. Different from a traditional method using a fixed threshold, the self-adaptive threshold selection strategy constructed by the method considers time context and data set features, the efficiency and precision of anomaly detection are improved, and effective support is provided for development of time sequence anomaly detection.

The invention discloses a high-precision continuous tracking method for a strong-maneuvering infrared weak and small target under a space-based detection visual angle, and the method comprises the steps: obtaining the uncertainty measurement of a target at a non-maximum suppression stage of a detection result, and enabling the uncertainty measurement to act on a target state updating and data association stage; therefore, the experience distribution of the detector is fully transmitted to the tracking process, and the tracking accuracy is improved. In order to cope with a complex target maneuvering state, an interactive multi-model technical route is adopted to design a tracking algorithm; in order to reduce the dependence on a prior motion model, a dynamic Markov transfer matrix construction method is designed, and the model transfer probability is updated in a mode of comprehensively modulating historical dynamic information and current static information; in a data association stage, targets with different scales are associated by integrating advantages of IoU and NWD, uncertainty is transmitted to a cost calculation process, and tracks and targets which are indefinitely matched are processed based on scale invariant and energy invariant hypotheses, so that high-precision continuous tracking of the targets is realized.

The invention provides a multi-party data sharing data sandbox and method. The multi-party data sharing data sandbox comprises a preprocessing module, a security access module, a sandbox core module and an output control module, the preprocessing module is used for desensitizing, fragmenting and encrypting original data; the security access module is used for establishing an encrypted transmission channel and executing identity authentication; the sandbox core module is used for executing a cross-domain calculation task, and data are kept in an encrypted state in the calculation process; and the output control module is used for desensitizing a calculation result, embedding an invisible watermark, triggering a multi-party approval process and outputting data after approval is passed. The technical problem that in the prior art, the requirements of data non-landing, multi-party joint governance and dynamic authority control are difficult to meet at the same time can be effectively solved.

The invention relates to the technical field of computing resource scheduling, and discloses a heterogeneous multi-task computing power dynamic scheduling method and system.The method comprises the steps that in response to multiple computing task requirements of a user, the priority of computing tasks is analyzed and sorted, and a task priority sequence is obtained; classifying the heterogeneous computing resources according to the resource states of the heterogeneous computing resources, and respectively placing different categories of computing resources in corresponding resource pools to obtain multiple categories of resource pools; in combination with the task priority sequence and the multi-class resource pool, allocating computing resources of a corresponding class to each task through a preset resource matching mechanism; in the task calculation process, when the calculation resources of the first task are insufficient, the calculation resources of the second task are preempted through a preset resource preemption mechanism, the priority of the second task is improved after resource preemption, and the calculation resources are redistributed; according to the method and the device, the corresponding computing resources can be allocated in combination with the task priorities, and the resources are dynamically scheduled when the computing resource gap appears.

The invention provides an end side model reasoning method and device based on an RWKV architecture, electronic equipment and a storage medium, and the method comprises the steps: obtaining a target input request of a target object, and converting the target input request into target model input data; loading a historical reasoning state corresponding to the target input request in a preset state storage space; determining a corresponding RWKV core operator according to the hardware platform type of the terminal equipment; based on the RWKV core operator, performing reasoning calculation on the target model input data and the historical reasoning state to obtain an output token sequence; wherein in the reasoning calculation process, the real-time reasoning state of the large language model is stored in a preset state accelerator memory for multiplexing; converting the output token sequence into a text format and outputting the output token sequence; and updating the historical reasoning state according to the real-time reasoning state after reasoning calculation. According to the method, calculation optimization and hardware acceleration can be carried out on the large language model of the RWKV architecture, so that the reasoning performance of the RWKV architecture model is improved on the end side.

This invention introduces a Hierarchical Tree-Based Attention (HTA) mechanism to optimize transformer-based large language models (LLMs) for processing hierarchical documents. HTA leverages a lineage-based approach to model parent-child and sibling relationships, preserving document hierarchy while reducing memory and computational demands. A novel data processing pipeline segments content into blocks, establishes hierarchical relationships, and produces annotated input for LLMs. During attention calculation, embeddings for lineage-related blocks compress information outside the immediate hierarchy, ensuring scalability without sacrificing accuracy. HTA enables efficient applications in structured document processing, such as legal, healthcare, and education, while improving generative tasks like summarization and question answering. This approach advances hierarchical NLP with superior fidelity and reduced latency.

The invention relates to a method for improving the calculation speed of a model based on a mercuric chloride AI processor. The method comprises the following steps: deploying the mercuric chloride AI processor and a deep learning model in a server, and carrying out adaptation and optimization on the mercuric chloride AI processor; before the data in the first buffer area is read, predicting and preloading the data to be processed, and loading the data from the global memory to the second buffer area in advance; a double-buffer mechanism is arranged in the mercuration AI processor, interrupt and event trigger points are set, the operation of loading data to a next buffer area is immediately started when a specified calculation stage is finished, and the time sequence of data circulation is accurately controlled; and a control parameter is automatically adjusted based on monitoring data of the real-time monitoring module, the calculation process is decomposed into a plurality of stages, different buffer areas are allocated for each stage, and access to the shared memory is optimized by setting a specified cache replacement strategy and the size of a cache line. According to the process, the calculation speed of the model in the deep learning field is improved, the resource utilization rate is improved, and the defect of manual adjustment and optimization is overcome.

The invention relates to the technical field of heat transmission, in particular to a heat transmission data storage and management system based on an industrial big data platform, and the system comprises a self-adaptive acoustic baseline modeling module which generates a self-adaptive baseline model library for storing the mapping relation between a working condition area and a model; the abnormal deviation degree calculation module is used for calculating and generating an abnormal deviation degree; the health state evaluation module is used for generating a comprehensive health index representing the long-term service performance of the pipe network; and the closed-loop correction and scheduling module is used for determining a comprehensive risk level according to the comprehensive health index and the change trend thereof, generating an operation and maintenance scheduling instruction for dynamically adjusting an abnormal deviation degree calculation process and pipe network operation parameters, and realizing closed-loop feedback control. According to the invention, accurate identification and positioning of abnormal events such as leakage, third-party damage and the like are realized.

The invention discloses an attention calculation implementation method and device, a medium, equipment and a product, and the method comprises the steps: loading the current to-be-calculated ith query block from a shared memory to a first register group of a consumer thread group, and carrying out the internal splitting calculation of the query block and a key block, so as to obtain corresponding attention score blocks; sequentially obtaining MK attention score blocks, and executing attention fusion calculation of mixing precision with the corresponding value blocks to obtain an attention output block corresponding to the ith query block; and after the attention output blocks are logically divided into N2 batches, a specified register group for storing target precision type data in the fusion calculation process is multiplexed and executed according to batches, target precision type conversion is carried out, and an output result after conversion of each batch is written back to a shared memory. According to the method, the existing hardware resources can be efficiently utilized to improve the calculation performance, and the method is particularly suitable for large-size query block and key block scenes.

The invention relates to an automatic guided vehicle (AGV) adaptive path tracking method based on MPC. The method comprises the following steps: S1, establishing a kinematic discretization error model based on the kinematic characteristics of the two-wheel differential AGV, processing a continuous kinematic equation by adopting an Euler discretization method, expressing a dynamic change relationship between a transverse deviation distance and an angle deviation in a state-space equation form, and generating a kinematic discrete state-space model; the method has the advantages that the continuous equation is processed by establishing the kinematics discretization error model and adopting the Euler discretization method, the deviation dynamic relation is expressed through the state space, the calculation process is simplified, the precision is kept, sensor data are fused, the wheel type odometer, IMU and laser data are integrated through the extended Kalman filtering algorithm, and the precision is improved. The real-time position and angle deviation are calculated, the positioning accuracy is improved, a model prediction controller objective function is designed, a weight matrix and boundary constraint are introduced according to constraint conditions, and the effect of obstacle avoidance constraint is combined.

The invention belongs to the technical field of artificial intelligence, is applied to the field of financial science and technology and the field of medical health, and discloses a rule engine configuration method, device and equipment based on machine learning and a storage medium. The method comprises the steps that historical calculation data and performance indexes in historical calculation tasks are collected, and a training data set is constructed; a machine learning model is trained, and algorithm parameters of the rule engine are dynamically optimized through the training data set; configuring an expression and a user-defined function of a rule engine, and realizing analysis and execution of scripts through JEXL (JavaScript Exchange Library); establishing a dynamic mapping relationship between the algorithm configuration library and the business variables, mapping algorithm variable names and business variable names, and generating configurable algorithm rules; and analyzing task load and resource requirements based on a pre-trained machine learning model, distributing system resources, and repairing abnormal behaviors in a calculation process. According to the invention, the problems of low processing efficiency, high maintenance cost, unreasonable resource allocation and poor expandability in the prior art are solved.

The invention belongs to the technical field of artificial intelligence, and particularly relates to a neural network automatic pruning method based on GRPO reinforcement learning, and the method comprises the steps: introducing a dynamic scaling factor into a batch normalization layer of a to-be-pruned neural network, and calculating the importance score of each convolution layer channel of the to-be-pruned neural network in combination with an attention mechanism; s2, constructing a multi-dimensional state vector containing layer structure features based on an importance calculation result in the step S1; s2, inputting the multi-dimensional state vector constructed in S2 into a strategy network of a GRPO reinforcement learning agent, generating a pruning action by the strategy network according to state information of a current network layer, and defining the action to represent a pruning rate of the layer; according to the method, a GRPO reinforcement learning algorithm is adopted, a traditional Critic model is abandoned, the strategy calculation process is simplified through a group sampling-relative advantage estimation mechanism, and memory occupation is remarkably reduced.

The embodiment of the invention discloses a method and device for constructing a field consanguinity tree, a storage medium and a terminal. Firstly, a target field, a database query statement of the target field and upstream blood relationship information are obtained, the target field serves as a root node, a first executable calculation function packaging calculation logic of the target field is generated, and the calculation process is dominant and modularized. By analyzing input parameters of the function, an upstream field which directly depends on is automatically identified and determined to serve as a child node. And then recursively processing each sub-node, generating a corresponding calculation function, analyzing the dependency of the calculation function, tracing layer by layer until all leaf nodes (namely original bottom table fields), outputting a complete blood relationship tree containing all the nodes and the corresponding calculation function, and displaying the complete blood relationship tree through a visual interface. According to the method, automatic and accurate tracing from a target field to an original data source is realized, a computable and reusable blood relationship knowledge framework is constructed, and a solid foundation is provided for data understanding, problem investigation and intelligent data service.

The invention provides a model training method and device, equipment and a storage medium, and relates to the technical field of computers, in particular to the technical field of neural network models and model training. The specific implementation scheme is as follows: a calculation unit executes quantization matrix multiplication based on Hadamard pre-transformation on an activation tensor and a weight tensor of a target model stored in a memory so as to generate an output tensor of a linear layer based on a low-precision tensor with smaller data bit width; using the output tensor and a subsequent network layer of the target model to complete forward propagation so as to obtain a loss value; and according to the loss value, updating model parameters of the target model stored in a memory through a back propagation algorithm. By means of the technical scheme, on the premise that the model training precision is guaranteed, memory resource occupation and the calculation amount in the calculation process can be remarkably reduced, and the training cost is reduced.

The invention discloses a flexible circuit board production quality monitoring method and system based on data feedback, and the method comprises the steps: collecting the physical parameters of development etching, drilling and copper plating processes, such as line width, line distance, aperture, roundness and copper thickness, and workshop temperature and humidity and dust concentration data; generating a single-variable control chart for a single-process physical parameter, generating a multivariable T2 control chart for a multi-coupling parameter process, and calculating a process stability index and a comprehensive fluctuation index; inputting the abnormal data into a BP neural network, identifying an abnormal mode type and outputting characteristic parameters; generating a parameter correction instruction according to the abnormal mode and the characteristic parameters, and adjusting equipment parameters in real time; recalculating the process capability index based on the adjusted data, and triggering secondary feedback if the process capability index does not reach the standard; and dynamically adjusting the threshold value of the control chart according to the standard deviation and the mean value of the process capability indexes of the continuous batches. Production key parameters are comprehensively covered, abnormity is found in time, accurate and comprehensive detection is achieved, real-time adjustment is achieved, stability and controllability are ensured, and the product quality is improved.

The present disclosure provides a computer-implemented system for automated augmentation of differential equation models. The system stores differential equations representing mechanistic behavior of physical or computational processes and constructs a hybrid computational solver by embedding a trainable universal approximator with adjustable parameters into the differential equations, where approximator outputs augment time derivatives during numerical integration. An iterative training process adjusts parameters through numerical integration, monitors integration failures, assigns infinite penalty values to loss functions when failures occur, and computes gradients using automatic differentiation otherwise. The system computes sensitivity metrics via Jacobian matrix evaluation, classifies input / output subsets as significant based on threshold-exceeding sensitivity metrics, generates a reduced approximator operating on classified subsets, and replaces the universal approximator with the reduced version to create an optimized solver.

The invention provides a storage battery health state assessment method fusing ultrasonic data and hybrid machine learning, and belongs to the technical field of storage battery health state assessment. The method focuses on performance comparison of an ANN model and a GWO-ANN model in storage battery SOH regression prediction, and finds that the GWO-ANN model accurately adjusts and optimizes ANN parameters by means of a grey wolf optimization algorithm, so that the storage battery health state assessment accuracy is improved. Compared with the prior art, the method has the advantages that higher prediction precision and generalization ability are shown, the R2 value is close to perfect, the error is extremely low, and the ANN model is remarkably surpassed, the HHT-GWO-ANN model is innovatively introduced, the model integrates the advantages of HHT and GWO, the calculation process is accelerated, the robustness and prediction precision of the model are enhanced, efficient and stable model construction is achieved, and the method is suitable for large-scale popularization and application. And the test performance of the method is consistent with that of the training stage, so that the practical potential of the method in storage battery SOH nondestructive testing is verified. The ultrasonic nondestructive testing method proposed by the invention shows unique advantages in SOH recognition of the storage battery, and overcomes the limitation of a traditional method.

The invention provides a deep reinforcement learning-based method and a deep reinforcement learning-based system for integrating communication, sensing and calculation in unmanned aerial vehicle edge calculation, and the method comprises the steps: S1, constructing a system model, including constructing a network model which comprises a central unmanned aerial vehicle and a plurality of auxiliary unmanned aerial vehicles; s2, constructing a service process model; s3, constructing a sensing process model; s4, constructing a communication process model; s5, constructing a calculation process model; s6, constructing an energy consumption model; s7, constructing a joint optimization problem based on the network model, the service process model, the sensing process model, the communication process model, the calculation process model and the energy consumption model; s8, based on the joint optimization problem, constructing a Markov decision process conversion model; and S9, according to the Markov decision process conversion model, constructing a network architecture based on an A3C algorithm, and carrying out distributed DRL agent design and training, so that the affiliated unmanned aerial vehicle and the central unmanned aerial vehicle directly generate a resource allocation strategy based on a local state.

A distributed model training method includes that: the training of an internal iteration with a preset number of internal iterations is performed on a preset model through a computation process, to obtain a first node model parameter value of a current global iteration; a second node model parameter value of the current global iteration of a second computing node in a distributed system is acquired through a communication process running in parallel with the computation process, and a first ALLReduce model parameter value of the current global iteration is determined according to the first node model parameter value and the second node model parameter value; and external iteration is performed through the computation process by using a second ALLReduce model parameter value of a last global iteration, to obtain a target model parameter value of the current global iteration.

The invention relates to the technical field of reactive power compensation of a power system, in particular to a stable voltage compensation control method applied to a micro-grid cluster, which specifically comprises the following steps of: analyzing principal components of all items of grid data of each node through a principal component analysis algorithm, extracting the first three principal components, and performing data weight distribution through an entropy weight method to obtain a data weight distribution result; constructing reactive compensation priorities of the nodes; the future reactive compensation priority of the node is predicted through the historical reactive compensation priority of the node, the fluctuation rate of prediction data is analyzed, and the dynamic compensation urgency of the node is constructed in combination with the clustering condition of the node at the current moment; a hierarchical resource allocation strategy is implemented according to the dynamic compensation urgency degree, the calculation process is updated in real time according to the reactive compensation feedback result, a self-optimization closed loop is formed, the problems that a traditional method is insufficient in quantization precision and poor in real-time performance are solved, the reactive demand matching precision and the voltage maintaining stable effect are improved, and the method is suitable for large-scale popularization and application. And the communication load and the calculation complexity are obviously reduced.

The invention relates to the field of data analysis, in particular to an internet-based convergence media data relation analysis system. The fusion media data relation analysis system based on the Internet comprises a fusion media data acquisition module, a fusion media data processing module and a hot topic analysis module. According to the method, the keywords of the convergence media data are extracted, the corresponding clustering clusters are constructed, then the correlation degree of the two pieces of convergence media data is calculated on the basis of the clustering clusters, and the convergence media data higher than the correlation threshold value are combined to be used for analyzing the hot topic condition of the current time period; and the correlation degree is calculated based on the similarity of the clustering clusters and the TF-IDF value, and the distribution condition of keywords corresponding to the clustering clusters is also considered in the calculation process, so that the accuracy of fusion media data matching can be improved.

The invention discloses a power dispatching method and system based on a topology dispatching graph, a storage medium and electronic equipment. The method comprises the following steps: acquiring meteorological data, historical load data and user behavior data; and inputting the meteorological data, the historical load data and the user behavior data into a convolutional neural network for feature extraction and feature fusion to obtain fusion features. Inputting the fusion features into a long-short-term memory network to obtain predicted future load data of the current moment in a period of time in the future; constructing a topological characteristic scheduling graph based on future load data, topological information of the power system and a historical scheduling decision; updating the topological feature scheduling atlas through an adversarial neural network; and decoupling the updated topological feature scheduling atlas to obtain a current scheduling decision. The efficiency of the scheduling strategy calculation process is improved, the topological structure of the power system is effectively fused, and the safety and adequacy of the power system are improved.

The invention provides a model training method. The method comprises the steps of collecting information of a first model; and making a re-calculation strategy according to the communication time of the first model in the communication stage and the execution time required by the plurality of operators in the first model. The information of the first model comprises execution time respectively required by a plurality of operators in the first model, the recalculation strategy comprises at least one recalculation operator in the plurality of operators and opportunity for executing a recalculation process by the at least one recalculation operator, the recalculation operator is an operator used for executing the recalculation process in the plurality of operators, and the opportunity for executing the recalculation process by the at least one recalculation operator in the plurality of operators is the opportunity for executing the recalculation process by the at least one recalculation operator. The opportunity at which the at least one recalculation operator performs the recalculation process includes performing the recalculation process in parallel with the communication phase of the first model. The recalculation process is parallel to the communication process of the model, so that the model training time is shortened, and the throughput of model training is improved.

The invention provides a multi-agent self-triggering control method and system, each agent calculates a next triggering moment according to the information of the current moment and the information of an upstream neighbor agent, the calculated triggering moment is taken as a pre-triggering moment, and if other upstream neighbors are triggered before the pre-triggering moment, the other upstream neighbors are triggered before the pre-triggering moment. If yes, updating and calculating the pre-triggering moment, and otherwise, performing triggering control on the intelligent agent by taking the current pre-triggering moment as the triggering time; an event trigger parameter and a time trigger parameter are fused in the calculation process of the trigger moment. According to the method, the advantages of event triggering and time triggering are combined, communication and calculation in a specified mode only need to be carried out at the triggering moment, continuous communication and calculation are not needed, the communication efficiency is improved, the system stability is improved, and the energy consumption ratio is reduced.

The invention discloses a user risk coefficient calculation process visualization method and system. The method comprises the following steps: acquiring user behavior data, and constructing a risk assessment dimension system; determining node types of non-leaf nodes, and constructing a rule tree structure; performing risk assessment on behavior characteristics of the user behavior data by using the risk prediction model, and extracting risk factors; analyzing the risk factors by using an SHAP model, determining feature influence relation data and a high-risk rule combination between the risk factors, and filling non-leaf nodes with non-leaf node attributes; determining node weights of non-leaf nodes to update attributes of the non-leaf nodes; based on leaf node attributes and updated non-leaf node attributes, breadth-first traversal is carried out on the rule tree structure, and a risk scoring result of a traversal path is calculated by adopting a bottom-up backtracking mechanism; and carrying out visual display on the risk scoring result of the rule tree structure and the traversal path. The risk identification accuracy can be improved.

The invention relates to the technical field of code development, and provides an AI code effective proportion statistical method and device, a medium and equipment. The method comprises the steps of obtaining related information of codes submitted by a user; according to a user name in the related information, searching log information of an AI code generated by a corresponding user through adoption of a code generation tool; under the condition that the file name of the code submitted by the user is matched with the file name in the log information, searching a corresponding submitted code segment from the code submitted by the user according to the mark information in the log information; and calculating the similarity between the AI code and the submitted code segment, and counting the effective proportion corresponding to the AI code according to the similarity. Therefore, the calculation of the effective proportion considers the quality of the AI code, so that the effective proportion can accurately reflect the real contribution of the AI code, and the calculation process is automatically realized, thereby avoiding the tedious, time-consuming and labor-consuming conditions of manual labeling.

Provided in the embodiments of the present description are a data processing method, a text processing method, an image processing method, a computing device, a computer storage medium and a computer program product. The data processing method comprises: determining a target computation graph of a target model and inputting target data of the target model; determining from the target computation graph the current node for processing the target data; and determining a data processing computation graph corresponding to the current node, processing the target data on the basis of the current node and / or the data processing computation graph, and obtaining a data processing result. By means of flexibly calling a data processing computation graph, it can be ensured that a target model performs an efficient and accurate pre-processing or post-processing operation on target data, a computation process is dynamically inserted into a target computation graph, and secondary compilation is not needed, such that the overheads of computation graph compilation are saved on, the requirement and consumption for computer hardware resources are reduced, and the model inference efficiency is improved.