228 results about "Task parallelism" patented technology

The invention discloses a method for supporting fast switching of multiple models by NPU (Network Processing Unit) firmware. According to the invention, the system real-time performance and the resource utilization rate in a multi-task scene are obviously improved. Through a dynamic hierarchical caching strategy and a priority scheduling mechanism, the system can intelligently allocate cache resources and preferentially guarantee rapid loading of a high-frequency and high-urgency model, for example, in an automatic driving scene, switching delay of a path planning model can be reduced to a millisecond level, and non-perceptual switching of key tasks is guaranteed. The incremental parameter loading technology and firmware-level context management are deeply fused, only model difference data are transmitted, and hardware is utilized to accelerate and recover a calculation state, so that bandwidth waste caused by traditional full-amount loading is greatly reduced, the model switching efficiency of edge computing equipment is improved by more than 10 times when multiple tasks such as voice recognition and image processing are carried out in parallel, and the efficiency of the edge computing equipment is improved. And meanwhile, the calculation precision is kept lossless.

The invention discloses an instruction execution method and device for an artificial intelligence chip, and relates to the technical field of artificial intelligence chips, and the method comprises the steps: carrying out the deep learning driven feature recognition and resource demand prediction of an input task, and generating a demand prediction report of the task for computing resources through the analysis of a computational graph and a data dependency relationship of the task; a computing unit and memory resources are intelligently scheduled, an optimal instruction execution path is dynamically selected, and meanwhile a caching strategy is optimized; automatically generating a micro instruction set corresponding to the task according to the computing resource demand, the computing characteristic and the intelligent scheduling result of the task; when multiple tasks are executed in parallel, the execution sequence of the multiple tasks is dynamically adjusted according to the calculation load and the resource sharing condition of the tasks, and resource allocation is optimized. According to the method, the computing resources and the memory bandwidth required by each task can be accurately predicted through the deep learning driving analysis of the task computing graph, so that the allocation of the computing resources is optimized.

The invention discloses a voice interactive console based on artificial intelligence, and relates to the technical field of intelligent consoles, a large language model is combined with context tracking and logic analysis technologies, a composite instruction is disassembled into independent tasks, a task chain is generated, a semantic weight is added to each task through a weight regression model, and the intelligent consoles are obtained. The task execution priority and the logic dependency relationship are identified, the limitation of a traditional system on the complex semantic analysis capability is overcome, a composite instruction can be accurately understood, a logic tree analysis method based on rules is adopted, a task chain is converted into a directed graph structure, the priority is dynamically adjusted, and a task execution sequence table is generated. In the scheduling process, multi-task parallel processing is carried out on tasks without logic dependence, and the resource utilization rate is increased; tasks with a dependency relationship are gradually scheduled by adopting topological sorting, so that the logic consistency is ensured.

The invention provides a resource intelligent bus and task parallel hydrogen energy data scheduling optimization method and system, relates to the technical field of energy, and aims to solve the problems of resource heterogeneity, lack of real-time performance and dynamic adjustment, insufficient energy efficiency optimization, poor adaptability to abnormal working conditions and the like in the prior art. According to the method, a resource intelligent bus architecture is constructed, and the resource intelligent bus architecture comprises a resource virtualization encapsulation layer, a resource state sensing layer and a global prediction and scheduling layer; in the resource state sensing layer, performing multi-granularity task decomposition on hydrogen energy business requirements, and establishing an inter-task dependency relationship model by adopting a directed acyclic graph; in the global prediction and scheduling layer, a multi-objective optimization model is constructed based on a dependency relationship between resource states and tasks, and resources and tasks are dynamically matched according to an optimal solution of the model; and dynamically scheduling the resources according to a matching result of the resources and the tasks. The problems in the prior art are solved, and the hydrogen energy data scheduling optimization performance is improved.

The invention provides a task parallel processing intelligent scheduling method and system oriented to large model services, and belongs to the technical field of service computation.The method comprises the steps that semantic features of instructions and texts of parallel subtasks are extracted through a LaBSE model, embedded vectors are generated, the vector dimension of the embedded vectors is reduced through a compression module, and the embedded vectors are distributed in parallel; inputting the compressed embedded vector into a regression device to predict the generation length; constructing a segmentation algorithm based on semantic integrity constraint, and segmenting the text into a text fragment list; dividing the segmented text fragments in the text fragment list into different batches of tasks by adopting a multi-constraint task batching algorithm according to a similar length constraint and a same instruction constraint to obtain a batch task set; and optimizing the batch task set by adopting a task allocation algorithm oriented to heterogeneous resources, and determining an optimized sub-task scheduling scheme. According to the method, efficient parallel processing of large model services is realized, and the reasoning speed, the resource utilization rate and the intelligent level of task scheduling are remarkably improved.

The invention belongs to the technical field of machine learning, and discloses a machine learning training task communication scheduling method and device, equipment and a storage medium. The method comprises the steps of obtaining a to-be-trained task in a task queue; training feature extraction is carried out on the to-be-trained task, and a communication operation workflow and the priority of each communication operation in the workflow are obtained according to the training features obtained through analysis; and obtaining a communication scheduling scheme of the to-be-trained task according to the priority of each communication operation, and executing communication scheduling of the whole training task through the scheduling scheme. By means of the mode, accurate prediction of task computing power load and communication requirements and fine-grained modeling of communication scheduling between the nodes are achieved, scheduling planning can be conducted on the communication behavior of the level of communication between the nodes, efficient overlapping of calculation and communication is achieved, the training iteration time is shortened, and the scheduling efficiency is improved. The method is suitable for scenes of large-scale model training and multi-task parallel training, and has remarkable economic benefits and application values.

The invention relates to the technical field of artificial intelligence, and provides a model distributed training optimization method, electronic equipment and a storage medium, and the method comprises the steps: respectively constructing a calculation flow and a communication flow based on the calculation operation and the communication operation in a model; in the training iteration of the model, executing staggered pipeline scheduling for at least two micro-batches of tasks; wherein the staggered pipeline scheduling comprises the following steps: scheduling a first micro-batch of calculation tasks to a calculation flow for execution, and scheduling a second micro-batch of communication tasks to a communication flow for execution, or scheduling a first micro-batch of communication tasks to the communication flow for execution, and scheduling the second micro-batch of calculation tasks to the calculation flow for execution, the tasks of the first micro-batch and the tasks of the second micro-batch are processed in parallel. According to the method, the parallel processing of the calculation task and the communication task is realized by executing the staggered assembly line scheduling on the tasks of different micro-batches, and the hardware utilization rate and the model training efficiency are improved.

The invention relates to a fine-grained dynamic unit task scheduling strategy, aims to solve the problem of real-time scheduling of DAG (directed acyclic graph) tasks on a multi-core platform, and belongs to the field of real-time embedded system task scheduling. Static priority distribution is adopted in an existing priority preemptive scheduling method, the parallel execution degree of tasks is limited, and response time is possibly increased. According to the method, the DAG vertex is divided into a plurality of unit nodes, the independent priority is allocated to each unit node, execution is performed with the shorter worst case execution time (WCET), the priority is dynamically adjusted, the limitation of the static priority is broken through, the task parallelism degree is improved, and the calculation limit of the worst response time (WCRT) is optimized. Meanwhile, a unit node priority distribution algorithm is designed, and the task scheduling efficiency is further improved.

The invention provides a safety measure generation method and system based on a power distribution network multi-task parallel work ticket, and the method comprises the steps: receiving a to-be-executed work ticket containing a plurality of parallel tasks, and extracting the task information of each task; obtaining a plurality of task groups and task group associated equipment sets corresponding to the task groups; obtaining a task group minimum operation sequence set; obtaining a task group topology boundary equipment set, and performing topology search by using a topology boundary equipment set subset to obtain a safety measure set of each task group; and based on the minimum range maintenance decision model, optimizing the safety measure set of each task group to obtain a final multi-task parallel safety measure set. According to the scheme, the tasks which are electrically connected adjacently are combined into one task group, so that repeated operation on the same equipment can be avoided, unnecessary workload is reduced, and the safety measure generation efficiency of the safety measure generation method based on the multi-task parallel work ticket of the power distribution network is improved.

The invention provides a multi-task parallel processing method for user problems under an AI platform, and belongs to the technical field of digital data processing of the AI platform. Task resource requirements are accurately calculated through a video memory pre-estimation function and a memory pre-estimation function, a resource consumption mode of concurrent execution is analyzed through a multi-task resource prediction model; the load capacity of the system is evaluated based on a video memory utilization rate gain index, an optimal task segmentation strategy is determined through a data set splitting degree calculation function, multi-stage video memory sub-pools are constructed to realize differentiated resource management, and a Nash equilibrium point of resource allocation is solved by adopting a data set video memory allocation game model. Collaborative optimization allocation of resources is achieved through the video memory and memory coupling allocation equation set, the task state is monitored in real time in the multi-task concurrent execution process, the resource allocation weight is dynamically adjusted, and the technical problem that the system resource utilization rate is low during AI platform multi-task parallel execution is solved.

The invention discloses a multi-task parallel data stream processing method and device based on an FPGA (Field Programmable Gate Array), which are applied to the technical field of data processing. Carrying out normalization processing on the decoded data to generate a data stream in a unified format; performing coding and ID identification code embedding processing on the normalized data to generate to-be-stored data with an identifier; carrying out FIFO (First In First Out) buffering and write-in priority judgment on the coded data to generate priority queue data; dDR writing and BANK scheduling processing are carried out on the priority queue data, and stably stored parallel data are generated; carrying out reading and output coordination processing on DDR storage data to generate a continuous output data stream, and triggering a reading application by decoding ID identification code positioning data, reading FIFO according to depth scheduling and monitoring data volume by channel temporary storage FIFO; and performing dynamic bandwidth allocation and energy efficiency optimization on the whole data stream processing process to generate an efficient parallel processing result.

The invention relates to the technical field of intelligent manufacturing, in particular to an intelligent building system and method based on BIM, and the system comprises a key node sorting module, a continuity index calculation module, a node coupling judgment module, a resource period regulation and control module and a node path rearrangement module. According to the method, through aggregation processing of the total number of construction node stages, the number of follow-up dependency levels and risk exposure values, node key levels are ranked through an analytic hierarchy process model, a construction task priority basis is constructed, a continuity index is formed in combination with task interval duration, a task switching time ratio and an equipment idle period, and state calibration is carried out; a node coupling imbalance relationship is judged through a continuity index difference value and a threshold value, a resource utilization rate is calculated based on a resource calling number, a path load and a neutral position number to judge a compression and release state, a conflict is identified based on a task parallelism degree and an equipment occupation conflict rate, and a task sequence structure is recombined by utilizing a genetic algorithm. And the adaptive capacity of the path to resource configuration is enhanced.

The invention relates to a task execution method and device based on a hybrid expert model, equipment, a storage medium and a program product, and relates to the technical field of artificial intelligence. The calculation utilization rate and the calculation efficiency of the equipment can be improved. The method comprises the steps that the number of lexical units distributed to each expert in equipment is determined, the number of lexical units of each expert group in the equipment is obtained according to the number of lexical units of each expert, the expert group comprises one or more experts in the equipment, and an execution sequence corresponding to each expert group is determined according to the relative size of the number of lexical units of each expert group in the equipment, different devices are configured with the same number of expert groups, each expert group comprises the same number of experts, the expert groups in the same execution sequence in the different devices have the same relative size in the respective devices, and the communication task and the calculation task of each expert group are sequentially executed according to the execution sequence. The computing task of each expert group in the device is parallel to the communication task of another expert group in the device.

The invention relates to the technical field of industrial automatic control, and discloses an industrial configuration software cross-platform cooperative control method and system based on a swan gap, which realize flexible initialization and dynamic task distribution of a control terminal by fully utilizing distributed communication and microkernel characteristics of a swan gap system. Through structured analysis and task graph construction of a configuration control model, granularity recombination and task parallel execution of a control process are supported; a task terminal recommendation mapping table is further generated in combination with the execution requirement of the control task and the terminal capability label information, and the adaptation matching degree of the control task and the execution terminal is effectively improved; and meanwhile, a distributed cooperative channel is constructed, a communication binding channel between the master control terminal and the slave control terminal is established, and a guarantee is provided for rapid scheduling and execution of tasks. Compared with the prior art, the heterogeneous compatibility, the task scheduling efficiency and the protocol adaptation flexibility can be remarkably improved, and the method is suitable for high-complexity and multi-terminal collaborative industrial automation control scenes.

The invention provides a graph card-based hybrid scheduling rendering method and system, and the method comprises the steps: dividing a frame buffer area, binding the divided frame buffer area with a task, completing the rendering of each card through a hybrid task scheduling strategy, and finally synthesizing and outputting a complete graph; a frame buffer area of a rendering graph is divided into a plurality of graph cards according to display criticality, each graph card is bound with an independent rendering task and a video memory area, a unique task division and resource allocation mode is formed, and the mode is different from a traditional overall rendering mode; and in combination with a hybrid scheduling strategy of high-priority task preemptive scheduling and dynamic card task parallel rendering, the problems of rendering task priority conflict and resource utilization are effectively solved.

The invention provides an MCM-GPU-oriented resource storage optimization method, device and equipment, and the method comprises the steps: carrying out the static analysis of an operator-level data flow diagram of a calculation task in a compiling stage, and extracting the data dependence relation and life cycle characteristics of each operator; generating a storage optimization scheme based on the data dependency relationship and life cycle characteristics of each operator; in the operation stage, task and data collaborative allocation is carried out according to a data prefetching strategy and a data replacement strategy, task and data collaborative allocation and the calculation task are executed in parallel, and a data flow diagram and an operator dependency relationship of the task are analyzed by utilizing a compiling period; in combination with the data locality characteristics of the MCM-GPU multi-level storage architecture and the concurrency requirements of different modules for data access, an appropriate GPU module is selected to carry out data optimization management, so that frequent data exchange between a GPU and a CPU is reduced, and the performance potential of the MCM-GPU architecture in a super-large-scale task is brought into full play.

The invention provides an inference method and device for a large language model, equipment and a medium, and belongs to the technical field of artificial intelligence. The method comprises the steps that a CPU and a GPU are used for cooperatively completing multiple iterations, and each iteration comprises the steps that a first task is used for determining a plurality of uncompleted sequences and lexical element identifiers generated finally in the CPU; the second task is used for determining model input on the CPU based on the lexical element identification finally generated by the plurality of sequences; the third task is used for calculating probability distribution of to-be-generated next lexical elements of the plurality of sequences on the GPU; the fourth task is used for sampling on the GPU to obtain an identifier of a next lexical element; the fifth task is used for updating the completion states of the multiple sequences on the CPU based on the identification of the next lexical element, the (n + 1) th iteration and the nth iteration are executed asynchronously, the first task and the second task of the (n + 1) th iteration and the third task of the nth iteration are executed in parallel, and the third task of the (n + 1) th iteration and the fifth task of the nth iteration are executed in parallel.

This invention discloses a method and apparatus for processing multi-task parallel data streams based on FPGA, applied in the field of data processing technology. The invention constructs a multi-format raw data receiving and decoding system; normalizes the decoded data to generate a unified format data stream; encodes and embeds ID identification codes into the normalized data to generate tagged data to be stored; performs FIFO buffering and write priority determination on the encoded data to generate a priority queue data; performs DDR write and BANK scheduling on the priority queue data to generate stable parallel data; and coordinates the reading and output of the DDR stored data to generate a continuous output data stream. Specifically, data is located by decoding the ID identification code, reads from the FIFO by depth scheduling, and the channel temporary storage FIFO monitors the data volume to trigger read requests. Dynamic bandwidth allocation and energy efficiency optimization are performed throughout the data stream processing process to generate efficient parallel processing results.

The invention provides a query optimization method and system for a distributed database. The method comprises the following steps: determining a high-frequency data unit set in a distributed database, and generating a sub-query task division scheme for each data unit in the high-frequency data unit set; according to the dependency relationship between the sub-query tasks represented by the sub-query task division scheme, determining the interaction frequency between the data units, and determining a data distribution scheme according to the interaction frequency; according to a storage relationship indicated by the data distribution scheme, determining an index set of each node for querying the stored data unit; for each index set, respectively calculating execution costs of different sub-query tasks according to an access mode supported by a query data unit supported by the index set, and generating a sub-query task parallel scheme according to the execution costs; and generating a node scheduling scheme meeting the execution sub-query task parallel scheme. The query speed of the distributed database and the overall performance of the system can be improved.

The invention provides a furniture board processing equipment control method. The furniture board processing equipment control method comprises the steps that board cutting tasks received in a polling mode are distributed to processing equipment; calculating the total length of a processing track required by each device for completing a new task and not executing a task, and determining a priority by combining with a cutting feeding speed; path scheduling is sequentially carried out according to priorities, during scheduling, whether tool interference or path overlapping exists in machining schemes between adjacent cutting coordinate points or not is judged on the basis of equipment state information, including tool paths of tasks being executed and tool paths of tasks not being executed, and a next execution node is selected from adjacent points without conflicts, the equipment state information comprises cutter positions, time points and motion tracks, conflict detection is realized by calculating cutter spacing and motion route directions, and priorities are sequenced according to track length and task completion time estimated by speed. According to the method, the scheduling efficiency and the machining precision of furniture plate machining equipment can be improved, the tool interference risk is avoided, and optimization of multi-task parallel processing is achieved.

The invention discloses a dynamic priority dyeing task scheduling and conflict resolution method, which comprises the following steps of: pre-checking task conflicts, extracting all step sequences of a newly-added task and reversing, and detecting whether irreversible opposite node conflicts exist between the step sequences of the newly-added task and all the step sequences of the executed task or not; a to-be-updated execution task queue is generated, the task dip dyeing state of each dyeing site is traversed, to-be-updated execution tasks are screened out, task weights are calculated, and the to-be-updated execution task queue is generated according to descending sort of the weights; a dynamic priority decision is made, and the task executed by the mechanical arm in the next step is determined according to the weight sorting result of the to-be-updated execution task and the step precision level; and a mechanical arm operation instruction set is generated, whether space-time conflicts exist or not is judged, if yes, a new to-be-updated execution task is selected to generate the mechanical arm operation instruction set, and if not, the mechanical arm operation instruction set is executed. Multi-task parallel processing is supported, and the rapid processing requirement of large-scale pathological samples is met.

The invention discloses a CAE (Computer Aided Engineering) workflow control subsystem and method based on cloud computing, and belongs to the technical field of crossing of computer simulation and cloud computing, and the subsystem comprises a workflow configuration analysis module which is used for converting a CAE workflow configuration file possibly containing cyclic dependence into a directed acyclic graph state machine through strong connectivity component analysis; the workflow state machine operation module is used for dynamically deriving task groups which can be executed in parallel in the next stage in the state machine according to the current state; and the workflow scheduling module is used for submitting the tasks to a cloud solving platform in parallel and realizing fault recovery through state persistence. Through deep fusion of a graph theory method and a cloud computing architecture, automatic scheduling, efficient parallel execution and reliable fault tolerance of the CAE workflow are realized, and the computing resource utilization rate and the system reliability are remarkably improved.

The invention discloses a GPU cluster resource allocation method and device, equipment and a storage medium, and relates to the technical field of GPU cluster resource allocation. According to the method, a hardware performance description vector is generated by fusing static hardware parameters and dynamic micro-benchmark test data, the limitation that the GPU performance is represented only by depending on the static parameters is broken through, and the actual operation capacity of different architecture GPUs in a heterogeneous cluster is accurately matched; task feature vectors are generated by extracting task calculation features and resource demands, and quantitative description of reasoning task resource consumption features is achieved; information of hardware, tasks and load dimensions is integrated through a machine learning model, performance degradation characteristics during multi-task parallel are effectively captured, and the accuracy of execution time prediction is improved; scheduling decision operation including candidate node screening, comprehensive cost evaluation and resource reservation backfilling is executed in combination with the predicted execution time, and the task execution efficiency and cluster load balancing are both considered.

The invention provides an efficient operation and maintenance software architecture setting method for multi-task parallel scheduling, and relates to the technical field of operation and maintenance automation. The method comprises the steps that resource state data of operation and maintenance nodes in an available area are collected in real time; dynamically calculating a comprehensive load index of each available area operation and maintenance node based on the resource state data to obtain a calculation result; according to the calculation result, identifying a high-load available area with the comprehensive load index exceeding a preset threshold value and a low-load available area with the comprehensive load index lower than the preset threshold value; and based on the high-load available area, performing priority grading on operation and maintenance tasks to be processed in the high-load available area according to task types to form a high-priority core task queue and a low-priority conventional task queue. According to the invention, accurate matching and efficient utilization of cross-regional resources are realized.

The invention provides an asynchronous parallel-based service processing method, system and device and a storage medium, and the method comprises the steps: initializing a service context object according to a service type identifier in a request; a preprocessing rule script corresponding to the service type is loaded and executed through a dynamic rule engine, and parameter verification, service pre-check and context initialization are carried out; a plurality of asynchronous processing rule scripts corresponding to the service types are loaded and executed in parallel, each asynchronous processing rule script corresponds to one service task and is packaged into an asynchronous task, the asynchronous tasks are executed in parallel, and a result is written into a service context object; and loading and executing a result processing rule script corresponding to the service type through a dynamic rule engine, aggregating results of the different step tasks, and generating and returning a service response after applying a final service logic. Parallel processing can be realized through a three-stage asynchronous processing architecture, multi-core CPU resources are fully utilized, and the throughput and response speed of the system are improved.

The invention provides a watershed spatio-temporal data storage and retrieval method and system based on multistage indexing, and the method comprises the steps: S1, obtaining watershed multi-source heterogeneous spatio-temporal data, and carrying out the logic partitioning of watershed spatial raster data; s2, constructing a global, local and memory multi-level index architecture; s3, executing a query task based on the multi-level index, scheduling the query task to a target node and performing parallel processing; comprising the following steps: positioning a target logic sub-block according to a query range, scheduling a task to a data storage node in combination with dynamic re-partitioning, and realizing end-to-end efficient retrieval through multi-stage indexing; and S4, executing partition adjustment according to the node load, namely dynamically triggering partition refining or merging operation according to the resource use condition of each node in the task running process, and optimizing the matching degree of data locality and task parallelism. According to the method, a multi-level collaborative index and dynamic re-partitioning mechanism is constructed, so that quick response to a complex query request and efficient scheduling of computing resources are realized.

The present application provides a query optimization method and system for a distributed database. The method includes: determining a high-frequency data unit set in a distributed database, and generating a sub-query task division scheme for each data unit in the high-frequency data unit set; determining the interaction frequency between each data unit according to the dependency relationship between sub-query tasks represented by the sub-query task division scheme, and determining a data distribution scheme based on the interaction frequency; determining an index set for each node to query the stored data units according to the storage relationship indicated by the data distribution scheme; for each index set, calculating the execution cost of different sub-query tasks according to the access mode supported by the data units supported by the index set, and generating a sub-query task parallel scheme based on the execution cost; generating a node scheduling scheme that satisfies the execution of the sub-query task parallel scheme. The present application can improve the query speed of a distributed database and the overall performance of the system.

The invention discloses a bank batch processing system and a working method thereof. The system comprises a task scheduling module, the task scheduling module is configured to automatically trigger batch processing tasks according to a preset scheduling strategy, and a task fragmentation module is configured to divide the batch processing tasks into a plurality of subtasks for parallel processing according to a fragmentation strategy and dynamically adjust the fragmentation number according to system resources. The task execution and fault-tolerant processing module is configured to capture task execution anomalies and execute retry or skipping operation on tasks which are executed abnormally based on a preset exception execution strategy, the monitoring and log module is configured to monitor task execution states in real time and record log information, and the performance optimization module is configured to optimize task execution performance. Through deep fusion of technical innovation and financial scenes, systematic improvement is realized in the aspects of efficiency, fault tolerance, extension, operation and maintenance, safety and the like, and an efficient, reliable and compliant solution is provided for bank batch business processing.

The invention relates to the technical field of large-model lightweight fine tuning, and discloses a parallel optimization method of a low-rank adapter and a task awareness scheduling system.The parallel optimization method comprises the steps that an increment matrix of the low-rank adapter is fragmented to a tensor parallel group according to rows, the tensor parallel group comprises a plurality of computing devices, and the computing devices are used for computing the increment matrix of the low-rank adapter; the fragmentation granularity is dynamically determined according to the equipment hardware capability and the dimension of the increment matrix; dynamically scheduling the tasks with strong conflicts to different task parallel groups based on the calculated inter-task gradient conflict coefficient; and for a plurality of tasks scheduled to the same task parallel group, adapter calculation is merged into unified large matrix operation by adopting a micro-batch processing technology, and LayerNorm and adapter projection calculation are merged into a single calculation kernel by adopting a kernel fusion technology. According to the method, communication redundancy and synchronization overhead in distributed training are reduced, task interference during multi-task parallel is effectively eliminated, and the problem of computing resource fragmentation is solved.

The invention provides a DCG-based agent arrangement and task scheduling method and device, relates to the technical field of artificial intelligence, and solves the technical problem of low task management flexibility due to an existing task scheduling mode. The method comprises the following steps: acquiring a plurality of task nodes; determining a dependency relationship among the plurality of task nodes according to the directed edge in the DCG, and controlling an execution sequence of the plurality of task nodes according to the dependency relationship, so that tasks corresponding to the task nodes are executed according to the sequence corresponding to the dependency relationship; according to the dependency relationship and the current state of the intelligent agent, dynamically allocating a plurality of tasks to the corresponding suitable intelligent agent for execution; according to the dependency relationship and the load condition of the intelligent agent, the multiple tasks are controlled to be executed in parallel, so that execution of the workflow is achieved; in the execution process of the workflow, the execution path of the task is dynamically adjusted according to the data for monitoring the task in real time so as to realize the real-time distribution of the task.