Data processing method, device and equipment of distributed computing system and readable medium

Distribution and resource allocation of large-model computing tasks through distributed computing systems and blockchain technology, solving the high cost and resource concentration problems of large-model training and inference, and achieving efficient, low-cost and sustainable computing resource utilization.

CN120144268APending Publication Date: 2025-06-13JD DIGITS HAIYI INFORMATION TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311714783.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-12-13
Publication Date
2025-06-13

AI Technical Summary

Technical Problem

The high cost and resource concentration of large model training and inference limit the participation of SMEs or individual developers.

Method used

Through a distributed computing system, blockchain technology is used to distribute, resource allocation and reward allocation of large-model computing tasks, realize the on-chain processing of computing tasks, the determination of attribute information of computing resources, and the allocation of tasks and the feedback of results.

Benefits of technology

It reduces the cost of large-scale model training and inference, improves computing efficiency, promotes the participation of small and medium-sized enterprises or individual developers, and achieves efficient, low-cost and sustainable computing resource utilization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120144268A_ABST
    Figure CN120144268A_ABST
Patent Text Reader

Abstract

The invention provides a data processing method, device and equipment for a distributed computing system and a readable medium, and the method comprises the steps: responding to a received processing request containing a computing task, and carrying out the uplink processing of the computing task; determining calculation attribute information of calculation resources associated with the distributed system; distributing the calculation task to the calculation resource according to the demand information of the calculation task and the calculation attribute information; and determining a calculation result of the calculation task according to a result fed back by the calculation resource. According to the embodiment of the invention, high-efficiency, low-cost and sustainable development of large model training and reasoning is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Background Art

[0002] Currently, the training and inference of large models are mainly completed by a large number of high-performance GPUs clusters. On the one hand, the cost of using these GPU clusters is extremely high, and it usually costs millions of dollars to complete the training of a large model. On the other hand, the computing resources are relatively concentrated and mainly monopolized by large companies, which limits the participation of small and medium-sized enterprises or individual developers.

[0003] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present disclosure, and thus may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0004] The purpose of the present disclosure is to provide a data processing method, device, equipment and readable medium for a distributed computing system, which is used to overcome at least to some extent the problem of great difficulty in implementing large model training or inference caused by the limitations and defects of related technologies.

[0005] According to the first aspect of the embodiments of the present disclosure, a data processing method for a distributed computing system is provided, including: in response to a received processing request including a computing task, performing an on-chain processing on the computing task; determining the computing attribute information of the computing resources associated with the distributed system; allocating the computing task to the computing resources according to the requirement information of the computing task and the computing attribute information; and determining the computing result of the computing task according to the result feedback by the computing resources.

[0006] In an exemplary embodiment of the present disclosure, determining the computing attribute information of the computing resources associated with the distributed system includes:

[0007] Determining the network status index, computing performance index, and load usage status index of the computing resources;

[0008] Performing a weighted processing on the network status index, the computing performance index, and the load usage status index;

[0009] Determining the computing attribute information according to the result of the weighted processing, and performing an on-chain processing on the computing attribute information.

[0010] In an exemplary embodiment of the present disclosure, allocating the computing task to the computing resources according to the requirement information of the computing task and the computing attribute information includes:

[0011] Determining the computing complexity of the computing task according to the requirement information;

[0012] Determining the priority of the computing task according to the computing complexity and the computing attribute information;

[0013] Allocate the computing task to the computing resource according to the priority.

[0014] In an exemplary embodiment of the present disclosure, allocating the computing task to the computing resource according to the requirement information of the computing task and the computing attribute information includes:

[0015] Determine a corresponding computing model according to the requirement information of the computing task;

[0016] Judge whether the size of the computing model is less than a preset model size;

[0017] Allocate the computing task and the computing model to the corresponding computing resource according to the determination result of the size of the computing model.

[0018] In an exemplary embodiment of the present disclosure, allocating the computing task and the computing model to the corresponding computing resource according to the determination result of the size of the computing model includes:

[0019] If it is determined that the size of the computing model is less than the preset computing model size, allocate the computing model to the computing resource that can carry the computing task according to the computing attribute information.

[0020] In an exemplary embodiment of the present disclosure, allocating the computing task and the computing model to the corresponding computing resource according to the determination result of the size of the computing model further includes:

[0021] If it is determined that the size of the computing model is greater than or equal to the preset computing model size, split the tensors of the computing model according to the computing attribute information;

[0022] Evenly allocate the split tensors and the corresponding preset weights to multiple computing resources for the multiple computing resources to calculate the computing task in a tensor parallel manner.

[0023] In an exemplary embodiment of the present disclosure, splitting the tensors of the computing model according to the computing attribute information includes:

[0024] When it is determined that the architecture of the computing model is a transformer architecture, if the preset weights are split by column, the input items of the computing model are not split;

[0025] When it is determined that the architecture of the computing model is a transformer architecture, if the preset weights are split by row, the input items of the computing model are split.

[0026] According to a second aspect of the embodiments of the present disclosure, there is provided a data processing device for a distributed computing system, including:

[0027] A blockchain uploading module configured to perform blockchain uploading processing on the computing task in response to a received processing request including the computing task;

[0028] A determination module configured to determine computing attribute information of computing resources associated with the distributed system;

[0029] An allocation module configured to allocate the computing task to the computing resources according to the requirement information of the computing task and the computing attribute information;

[0030] A feedback module configured to determine the computing result of the computing task according to the result feedback by the computing resources.

[0031] According to a third aspect of the present disclosure, there is provided an electronic device, including: a memory; and a processor coupled to the memory, the processor being configured to execute the method as described in any one of the above based on instructions stored in the memory.

[0032] According to a fourth aspect of the present disclosure, there is provided a computer-readable storage medium having a program stored thereon, which when executed by a processor implements the data processing method of the distributed computing system as described in any one of the above.

[0033] In the embodiments of the present disclosure, by performing blockchain uploading processing on the computing task in response to a received processing request including the computing task, determining the computing attribute information of the computing resources associated with the distributed system, then allocating the computing task to the computing resources according to the requirement information of the computing task and the computing attribute information, and finally determining the computing result of the computing task according to the result feedback by the computing resources, that is, using blockchain technology for large model computing task distribution, resource allocation and reward distribution, thereby realizing the efficient, low-cost and sustainable development of large model training and inference.

[0034] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present disclosure and used together with the specification to explain the principles of the present disclosure. Obviously, the accompanying drawings in the following description are only some embodiments of the present disclosure, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0036] Figure 1A schematic diagram of an exemplary system architecture of a data processing solution for a distributed computing system to which embodiments of the present invention can be applied is shown;

[0037] Figure 2 It is a flowchart of a data processing method for a distributed computing system in an exemplary embodiment of the present disclosure;

[0038] Figure 3 It is a flowchart of a data processing method for a distributed computing system in an exemplary embodiment of the present disclosure;

[0039] Figure 4 It is a flowchart of a data processing method for a distributed computing system in an exemplary embodiment of the present disclosure;

[0040] Figure 5 It is a flowchart of a data processing method for a distributed computing system in an exemplary embodiment of the present disclosure;

[0041] Figure 6 It is a flowchart of a data processing method for a distributed computing system in an exemplary embodiment of the present disclosure;

[0042] Figure 7 It is a flowchart of a data processing method for a distributed computing system in an exemplary embodiment of the present disclosure;

[0043] Figure 8 It is a flowchart of a data processing method for a distributed computing system in an exemplary embodiment of the present disclosure;

[0044] Figure 9 It is a schematic diagram of the architecture of a data processing solution for a distributed computing system in an exemplary embodiment of the present disclosure;

[0045] Figure 10 It is a schematic diagram of tensor splitting in a data processing solution for a distributed computing system in an exemplary embodiment of the present disclosure;

[0046] Figure 11 It is a block diagram of a data processing device for a distributed computing system in an exemplary embodiment of the present disclosure;

[0047] Figure 12 It is a block diagram of an electronic device in an exemplary embodiment of the present disclosure. Detailed implementation manners

[0048] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art. The features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of the present disclosure. However, those skilled in the art will recognize that one or more of the specific details may be omitted, or other methods, components, devices, steps, etc. may be used. In other cases, well-known technical solutions are not shown or described in detail to avoid obscuring the various aspects of the present disclosure.

[0049] In addition, the accompanying drawings are only schematic illustrations of the present disclosure, and the same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0050] Figure 1 A schematic diagram of an exemplary system architecture showing a data processing solution of a distributed computing system to which embodiments of the present invention can be applied is shown.

[0051] As Figure 1 shown, the system architecture 100 may include one or more of the terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0052] It should be understood that Figure 1 the numbers of terminal devices, networks, and servers in

[0053] are merely illustrative. According to the implementation requirements, there may be any number of terminal devices, networks, and servers. For example, the server 105 may be a server cluster composed of multiple servers, etc.

[0054] In some embodiments, the data processing method of the distributed computing system provided by the embodiments of the present invention is generally executed by the server 105. Correspondingly, the data processing device of the distributed computing system is generally disposed in the terminal device 103 (which may also be the terminal device 101 or 102). In other embodiments, some terminals may have functions similar to those of server devices and thus execute this method.

[0055] The following will describe in detail exemplary embodiments of the present disclosure with reference to the accompanying drawings.

[0056] Figure 2 It is a flowchart of the data processing method of the distributed computing system in the exemplary embodiment of the present disclosure.

[0057] Referring to Figure 2 , the data processing method of the distributed computing system may include:

[0058] Step S202: In response to a received processing request including a computing task, perform on-chain processing on the computing task.

[0059] Step S204: Determine the computing attribute information of the computing resources associated with the distributed system.

[0060] Step S206: Allocate the computing task to the computing resources according to the requirement information of the computing task and the computing attribute information.

[0061] Step S208: Determine the computing result of the computing task according to the result feedback by the computing resources.

[0062] In the embodiments of the present disclosure, by performing on-chain processing on the computing task in response to a received processing request including a computing task, determining the computing attribute information of the computing resources associated with the distributed system, then allocating the computing task to the computing resources according to the requirement information of the computing task and the computing attribute information, and finally determining the computing result of the computing task according to the result feedback by the computing resources, that is, using blockchain technology for large model computing task distribution, resource allocation, and reward distribution, thereby achieving efficient, low-cost, and sustainable development of large model training and inference.

[0063] In an exemplary embodiment of the present disclosure, as Figure 9 shown, the users of the distributed computing framework 900 can be divided into three roles, namely the model / / data owner 906, the computing task publisher 902, and the computing resource owner 904. The main tasks of these three in the computing framework are:

[0064] Computing task publisher 902: Responsible for publishing the training and inference tasks of large models. They provide the requirements and benefits for the platform and are the demand side of the distributed computing framework.

[0065] Computing resource owner 904: Individuals or organizations that own idle computing resources (such as gaming or mining GPUs). They provide the computing resources to the platform and participate in the execution of distributed computing tasks to obtain corresponding rewards. These computing resources help reduce the cost of the entire platform and improve computing efficiency.

[0066] Model / data owner 906: Those who provide the data required for large model training, which can be enterprises, research institutions, or individuals. They support the training of large models by providing data, thereby enabling higher-quality and more widely applicable large models.

[0067] Next, each step of the data processing method of the distributed computing system will be described in detail.

[0068] In an exemplary embodiment of the present disclosure, as Figure 3 shown, determining the computing attribute information of the computing resources associated with the distributed system includes:

[0069] Step S302, determining the network status index, computing performance index, and load usage status index of the computing resources.

[0070] In an exemplary embodiment of the present disclosure, when the distributed system is started, information of each computing resource r is obtained. This information mainly includes the computing performance index R(r), the load usage status index L(r), and the network status index N(r), and normalization processing is performed.

[0071] In an exemplary embodiment of the present disclosure, the computing performance index R(r) mainly includes the processor speed (unit: GHz), memory size (unit: GB), hard disk space (unit: TB), GPU capability (unit: TFLOPS), etc., but is not limited thereto. After formulating weights for each index and normalizing, the performance score R(r) can be obtained.

[0072] In an exemplary embodiment of the present disclosure, the current load usage status index L(r) mainly includes CPU usage rate, memory usage rate, hard disk usage rate, network bandwidth usage rate, GPU usage rate, and video memory usage rate, etc., but is not limited thereto. The unit is all percentage. After formulating weights for each index and normalizing, L(r) can be obtained.

[0073] In an exemplary embodiment of the present disclosure, the network state metric N(r) mainly includes latency (unit: ms), bandwidth (unit: Mbps), packet loss rate (unit: %), jitter (unit: ms), etc., but is not limited thereto. After formulating weights for each metric and normalizing them, N(r) can be obtained.

[0074] Step S304: Perform weighted processing on the network state metric, the computing performance metric, and the load usage state metric.

[0075] In an exemplary embodiment of the present disclosure, a computing resource score S(r) is defined for each computing resource r, which can be expressed as:

[0076] S(r) = w1 × R(r) - w2 × L(r) + w3 × N(r).

[0077] Where w1 is the scoring weight of the computing performance metric, w2 is the scoring weight of the load usage state metric, w3 is the scoring weight of the network state metric, and the computing framework uploads the calculated resource score result to the blockchain.

[0078] Step S306: Determine the computing attribute information according to the result of the weighted processing, and perform on-chain processing on the computing attribute information.

[0079] In an exemplary embodiment of the present disclosure, as Figure 4 shown, allocating the computing task to the computing resource according to the requirement information of the computing task and the computing attribute information includes:

[0080] Step S402: Determine the computing complexity of the computing task according to the requirement information.

[0081] Step S404: Determine the priority of the computing task according to the computing complexity and the computing attribute information.

[0082] Step S406: Allocate the computing task to the computing resource according to the priority.

[0083] In an exemplary embodiment of the present disclosure, for each computing task t to be distributed, a function F(r, t) for calculating the task priority is defined, which is expressed as follows:

[0084] F(r, t) = R(r) / (L(r) + 1) - T(t) - N(r).

[0085] Among them, R(r) represents the computing resource capacity, L(r) represents the current load, N(r) represents the network status of the current resource, T(t) represents the complexity of the task, and R(r) / (L(r)+1) represents the effective computing resource capacity. Adding 1 to L(r) is to prevent the denominator from being zero. The lower the load, the higher the effective capacity. For the task complexity T(t), the higher its value, the lower the priority.

[0086] In an exemplary embodiment of the present disclosure, as Figure 5 shown, allocating the computing task to the computing resource according to the demand information of the computing task and the computing attribute information includes:

[0087] Step S502, determining a corresponding computing model according to the demand information of the computing task.

[0088] Step S504, determining whether the volume of the computing model is less than a preset model volume.

[0089] Step S506, allocating the computing task and the computing model to the corresponding computing resource according to the determination result of the volume of the computing model.

[0090] In an exemplary embodiment of the present disclosure, as Figure 6 shown, allocating the computing task and the computing model to the corresponding computing resource according to the determination result of the volume of the computing model includes:

[0091] Step S602, if it is determined that the volume of the computing model is less than the preset computing model volume, then allocating the computing model to the computing resource that can carry the computing task according to the computing attribute information.

[0092] In an exemplary embodiment of the present disclosure, select the task with the highest priority from the list of tasks to be distributed, check whether the selected task has specific hardware or software environment requirements. If so, confirm whether the selected computing resource meets these requirements. If not, add the task back to the list of tasks to be distributed, then return to the fourth step to recalculate the task priority, distribute the selected task to the corresponding computing resource r, update the load information L(r) of the computing resource r, and remove the task from the list of tasks to be distributed.

[0093] In an exemplary embodiment of the present disclosure, as Figure 7 shown, allocating the computing task and the computing model to the corresponding computing resource according to the determination result of the volume of the computing model further includes:

[0094] Step S702, if it is determined that the volume of the computing model is greater than or equal to the preset computing model volume, then split the tensor of the computing model according to the computing attribute information.

[0095] Step S704: evenly distribute the segmented tensors and corresponding preset weights to multiple said computing resources for the multiple said computing resources to perform the computing task in a tensor parallel manner.

[0096] In an exemplary embodiment of the present disclosure, for a scenario where the model volume is small but the inference data volume is large, a data parallel method is used. The computing framework distributes the model and parameters to each computing resource owner, that is, each computing node has the same model and parameters. The computing framework distributes the data according to the device computing power. After receiving the data, each computing node independently completes the training or inference task and uploads the results to the blockchain.

[0097] In an exemplary embodiment of the present disclosure, as Figure 8 shown, splitting the tensors of the computing model according to the computing attribute information includes:

[0098] Step S802: When it is determined that the architecture of the computing model is a transformer architecture, if the preset weight is split by column, the input item of the computing model is not split.

[0099] Step S804: When it is determined that the architecture of the computing model is a transformer architecture, if the preset weight is split by row, the input item of the computing model is split.

[0100] In an exemplary embodiment of the present disclosure, for a scenario where the model volume is large, a tensor parallelism method is used. The computing framework splits a complete tensor into multiple blocks and then evenly distributes them to the computing nodes for computing. For example, in a transformer architecture, it mainly consists of a linear layer and activation functions such as GeLU. When splitting the weights of the model by row or column, the linear matrix operations are as follows: when splitting the weights by column, the input does not need to be split, and the final result is combined through concat; when splitting the weights by row, the input also needs to be split, and the final result is combined through addition.

[0101] In an exemplary embodiment of the present disclosure, as Figure 10As shown, for scenarios where the model volume is large, the computing framework will split a complete tensor into multiple blocks and then evenly distribute them to computing nodes for calculation. For example, in the transformer architecture, it mainly consists of linear layers and activation functions such as GeLU. When splitting the model weights by rows or columns, the linear matrix operations are as follows: when splitting the 1002 weights by columns into A1 and A2, the input X does not need to be split, and the final calculation results Y1 and Y2 are combined through concat to obtain the result Y; when splitting the 1004 weights A into A3 and A4 by rows, the input X also needs to be split into X3 and X4, and the Y3 and Y4 obtained through dot product calculation are finally combined through addition to obtain the result Y.

[0102] In an exemplary embodiment of the present disclosure, assume a function is GeLU(XA)B. Splitting A in the activation function by columns can complete the activation calculation in a single GPU. At this time, the corresponding B can be calculated by rows, and all the above operations can be completed only in their respective GPUs, with fewer communication operations, and finally the results are synchronized and merged.

[0103] In summary, the calculation process of the distributed computing framework of the present disclosure can be described as the following steps:

[0104] Step 1: Data reception. The computing framework receives relevant data from the model data owner and stores it in an integrated distributed storage (such as IPFS). If it is a model training task, the computing framework needs to receive the model structure and training data, etc.; if it is an inference task, the computing framework needs to receive the large model weight file, etc.

[0105] Step 2: Task recording and publishing. The computing framework receives the computing task requirements from the computing task publisher, uploads and records them on the blockchain, and each task generates a unique task ID for task tracking and management. In the training task, the computing task publisher and the model data owner may be the same role; while in the inference task, the computing task publisher may be the user of the large model.

[0106] Step 3: Resource access. The computing framework receives the access requests of idle GPUs (such as gaming GPUs, etc.) from computing resource owners and registers them on the blockchain. The registration information includes the performance and status of the resources, etc.

[0107] Step 4: Task distribution. The computing framework will distribute tasks to suitable computing resources through smart contracts according to the task requirements and the performance of the computing resources. Due to the large scale of the model, the framework will split the model, and each task may be divided into multiple subtasks and performed on different computing resources respectively.

[0108] Step 5: Computation Execution. The devices of the computing resource owners will start to execute the distributed subtasks. After the computation is completed, the results will be sent back to the computing framework. To ensure the reliability of the results returned by the computing owners and the stability of the computing process, the computing framework needs to redundantly distribute the tasks, that is, randomly and repeatedly distribute the same task to multiple computing nodes.

[0109] Step 6: Result Aggregation: The computing framework will collect the computation results of all subtasks for aggregation and processing. After the first node returns the result, it is necessary to wait for the computation results of randomly selected N nodes that execute the same task. When the result of this node is consistent with the results generated by these nodes, the result of the first node is used as the final task result and returned.

[0110] Step 7: Reward Distribution. The computing resource owners who complete the tasks will receive rewards (such as tokens, etc.) according to their contributions, and the distribution of rewards is completed by the smart contract of the computing framework.

[0111] Corresponding to the above method embodiments, the present disclosure also provides a data processing device for a distributed computing system, which can be used to execute the above method embodiments.

[0112] Figure 11 It is a block diagram of a data processing device for a distributed computing system in an exemplary embodiment of the present disclosure.

[0113] Reference Figure 11 , the data processing device 1100 of the distributed computing system may include:

[0114] The on-chain module 1102 is configured to perform on-chain processing on the computing task in response to a received processing request including the computing task.

[0115] The determination module 1104 is configured to determine the computing attribute information of the computing resources associated with the distributed system.

[0116] The allocation module 1106 is configured to allocate the computing task to the computing resources according to the requirement information of the computing task and the computing attribute information.

[0117] The feedback module 1108 is configured to determine the computing result of the computing task according to the result feedback by the computing resources.

[0118] In an exemplary embodiment of the present disclosure, the determination module 1104 is further configured to:

[0119] Determine the network status index, computing performance index, and load usage status index of the computing resources;

[0120] Perform weighted processing on the network status index, the computing performance index, and the load usage status index;

[0121] Determine the calculation attribute information according to the result of the weighted processing, and perform an on-chain process on the calculation attribute information.

[0122] In an exemplary embodiment of the present disclosure, the allocation module 1106 is further configured to:

[0123] Determine the calculation complexity of the calculation task according to the requirement information;

[0124] Determine the priority of the calculation task according to the calculation complexity and the calculation attribute information;

[0125] Allocate the calculation task to the calculation resources according to the priority.

[0126] In an exemplary embodiment of the present disclosure, the allocation module 1106 is further configured to:

[0127] Determine a corresponding calculation model according to the requirement information of the calculation task;

[0128] Judge whether the volume of the calculation model is less than a preset model volume;

[0129] Allocate the calculation task and the calculation model to corresponding calculation resources according to the determination result of the volume of the calculation model.

[0130] In an exemplary embodiment of the present disclosure, the allocation module 1106 is further configured to:

[0131] If it is determined that the volume of the calculation model is less than the preset calculation model volume, allocate the calculation model to the calculation resources that can carry the calculation task according to the calculation attribute information.

[0132] In an exemplary embodiment of the present disclosure, the allocation module 1106 is further configured to:

[0133] If it is determined that the volume of the calculation model is greater than or equal to the preset calculation model volume, split the tensor of the calculation model according to the calculation attribute information;

[0134] Evenly allocate the split tensor and the corresponding preset weights to multiple calculation resources for the multiple calculation resources to calculate the calculation task in a tensor parallel manner.

[0135] In an exemplary embodiment of the present disclosure, the allocation module 1106 is further configured to:

[0136] When it is determined that the architecture of the calculation model is a transformer architecture, if the preset weights are split by columns, the input items of the calculation model are not split;

[0137] When it is determined that the architecture of the computing model is a Transformer architecture, if the preset weights are split by row, the input items of the computing model are split.

[0138] Since the functions of apparatus 1100 have been described in detail in their corresponding method embodiments, they will not be elaborated herein.

[0139] It should be noted that although several modules or units of a device for action execution are mentioned in the above detailed description, such a division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more of the above-mentioned modules or units can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0140] In an exemplary embodiment of the present disclosure, there is also provided an electronic device capable of implementing the above method.

[0141] Those skilled in the art to which the present invention pertains can understand that various aspects of the present invention can be implemented as a system, a method, or a program product. Therefore, various aspects of the present invention can be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or an implementation combining hardware and software aspects, which can be collectively referred to herein as "circuitry", "module", or "system".

[0142] The following refers to Figure 12 to describe the electronic device 1200 according to this embodiment of the present invention. Figure 12 The electronic device 1200 shown is merely an example and should not impose any limitation on the functions and usage scope of the embodiments of the present invention.

[0143] As Figure 12 shown, the electronic device 1200 is presented in the form of a general-purpose computing device. The components of the electronic device 1200 may include, but are not limited to: at least one of the above-mentioned processing units 1210, at least one of the above-mentioned storage units 1220, and a bus 1230 connecting different system components (including the storage unit 1220 and the processing unit 1210).

[0144] Wherein, the storage unit stores program code, and the program code can be executed by the processing unit 1210, so that the processing unit 1210 executes the steps according to various exemplary embodiments of the present invention described in the above "Exemplary Method" section of this specification. For example, the processing unit 1210 can execute the method as shown in the embodiments of the present disclosure.

[0145] The storage unit 1220 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 12201 and / or a cache storage unit 12202, and may further include a read-only storage unit (ROM) 12203.

[0146] The storage unit 1220 may also include a program / utilities 12204 having a set (at least one) of program modules 12205. Such program modules 12205 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment.

[0147] The bus 1230 may represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus structures.

[0148] The electronic device 1200 may also communicate with one or more external devices 1240 (such as a keyboard, a pointing device, a Bluetooth device, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 1200, and / or may communicate with any device that enables the electronic device 1200 to communicate with one or more other computing devices (such as a router, a modem, etc.). Such communication may be carried out through an input / output (I / O) interface 1250. Moreover, the electronic device 1200 may also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 1260. As shown in the figure, the network adapter 1260 communicates with other modules of the electronic device 1200 through the bus 1230. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 1200, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0149] Through the description of the above embodiments, those skilled in the art can easily understand that the example embodiments described herein can be implemented by software, or can be implemented by a combination of software and necessary hardware. Therefore, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

[0150] In an exemplary embodiment of the present disclosure, there is also provided a computer-readable storage medium, on which a program product capable of implementing the above methods of this specification is stored. In some possible implementation manners, various aspects of the present invention can also be implemented in the form of a program product, which includes program code. When the program product runs on a terminal device, the program code is used to cause the terminal device to execute the steps according to various exemplary embodiments of the present invention described in the above "Exemplary Method" section of this specification.

[0151] The program product for implementing the above method according to an embodiment of the present invention may be a portable compact disc read-only memory (CD-ROM) and includes program code, and can run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto. In this document, the readable storage medium may be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, apparatus, or device.

[0152] The program product may adopt any combination of one or more readable media. The readable media may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0153] The computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries the readable program code. Such a propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable signal medium may also be any readable medium other than the readable storage medium, and the readable medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device.

[0154] The program code contained on the readable medium can be transmitted by any suitable medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0155] The program code for performing the operations of the present invention can be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, C++, etc., and also including conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, executed as a stand-alone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user's computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., by using an Internet service provider to connect through the Internet).

[0156] In addition, the above-mentioned drawings are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present invention, rather than for limiting purposes. It is easy to understand that the processes shown in the above-mentioned drawings do not indicate or limit the chronological order of these processes. Additionally, it is also easy to understand that these processes can be executed, for example, synchronously or asynchronously in multiple modules.

[0157] Other embodiments of the present disclosure will be readily apparent to those skilled in the art after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include known common knowledge or conventional technical means in the technical field not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of the present disclosure are pointed out by the claims.

Claims

1. A data processing method for a distributed computing system, characterized in that, it includes: In response to a processing request containing a computing task, perform on-chain processing on the computing task; Determine the computing attribute information of the computing resources associated with the distributed system; Allocate the computing task to the computing resources according to the requirement information of the computing task and the computing attribute information; Determine the computing result of the computing task according to the result feedback by the computing resources.

2. The data processing method for a distributed computing system according to claim 1, characterized in that, Determining the computing attribute information of the computing resources associated with the distributed system includes: Determine the network status index, computing performance index, and load usage status index of the computing resources; Perform weighted processing on the network status index, the computing performance index, and the load usage status index; Determine the computing attribute information according to the result of the weighted processing, and perform on-chain processing on the computing attribute information.

3. The data processing method for a distributed computing system according to claim 1, characterized in that, Allocating the computing task to the computing resources according to the requirement information of the computing task and the computing attribute information includes: Determine the computing complexity of the computing task according to the requirement information; Determine the priority of the computing task according to the computing complexity and the computing attribute information; Allocate the computing task to the computing resources according to the priority.

4. The data processing method for a distributed computing system according to any one of claims 1-3, characterized in that, Allocating the computing task to the computing resources according to the requirement information of the computing task and the computing attribute information includes: Determine the corresponding computing model according to the requirement information of the computing task; Judge whether the volume of the computing model is less than the preset model volume; Allocate the computing task and the computing model to the corresponding computing resources according to the determination result of the volume of the computing model.

5. The data processing method for a distributed computing system according to claim 4, characterized in that, Allocating the computing task and the computing model to the corresponding computing resources according to the determination result of the volume of the computing model includes: If it is determined that the volume of the computing model is less than the preset computing model volume, then allocate the computing model to the computing resources that can carry the computing task according to the computing attribute information.

6. The data processing method for a distributed computing system according to claim 4, characterized in that, Allocating the computing task and the computing model to the corresponding computing resources according to the determination result of the volume of the computing model further includes: If it is determined that the volume of the computing model is greater than or equal to the preset computing model volume, then split the tensor of the computing model according to the computing attribute information; Evenly allocate the split tensors and the corresponding preset weights to multiple computing resources for multiple computing resources to calculate the computing task in a tensor parallel manner.

7. The data processing method for a distributed computing system according to claim 6, characterized in that, Partitioning the tensors of the computing model according to the calculated attribute information includes: When it is determined that the architecture of the computing model is a Transformer architecture, if the preset weights are split by column, the input items of the computing model are not split; When it is determined that the architecture of the computing model is a Transformer architecture, if the preset weights are split by row, the input items of the computing model are split.

8. A data processing device for a distributed computing system Characterized in that It includes: An on-chain module, configured to perform on-chain processing on the computing task in response to a received processing request including the computing task; A determination module, configured to determine the computing attribute information of the computing resources associated with the distributed system; An allocation module, configured to allocate the computing task to the computing resources according to the requirement information of the computing task and the computing attribute information; A feedback module, configured to determine the computing result of the computing task according to the result feedback by the computing resources.

9. An electronic device Characterized in that It includes: A memory; And A processor coupled to the memory, the processor being configured to execute the data processing method of the distributed computing system according to any one of claims 1-7 based on instructions stored in the memory.

10. A computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, it implements the data processing method of the distributed computing system according to any one of claims 1-7.