Log processing method and device of multi-model cooperation system and electronic equipment

Abstract

The invention relates to the technical field of data processing, and provides a log processing method and device for a multi-model cooperation system and electronic equipment, and the method comprises the steps: obtaining output data of target subtasks corresponding to the processing of a plurality of target models in the multi-model cooperation system; wherein the target subtask is a subtask obtained by splitting a target processing task, and the target model is a model for processing the target subtask; according to the calling sequence of the multiple target models and the output data, generating an execution log of the target processing task; and under the condition that the execution result of the target processing task is abnormal, determining an abnormal model in the target model according to the execution log.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the technical field of logs, and more specifically, to a method, apparatus, and electronic device for processing logs of a multi-model collaboration system. Background Art

[0002] A multi-model collaboration system is a system that combines multiple different models and completes complex tasks through collaboration.

[0003] The multi-model collaboration system selects an appropriate model or model combination from the model library according to the type and characteristics of the task to process the task.

[0004] Since the models in the model library involve the collaboration of multiple service providers and multiple models, it is impossible to trace the source of problems when problems occur in the task execution results. Summary of the Invention

[0005] An object of embodiments of the present disclosure is to provide a new technical solution for processing logs of a multi-model collaboration system.

[0006] According to a first aspect of embodiments of the present disclosure, there is provided a method for processing logs of a multi-model collaboration system, including: Obtaining output data of multiple target models in the multi-model collaboration system for processing corresponding target subtasks; wherein, the target subtask is a subtask obtained by splitting a target processing task, and the target model is a model for processing the target subtask; Generating an execution log of the target processing task according to the call order of the multiple target models and the output data; In the case where an exception occurs in the execution result of the target processing task, determining an abnormal model that has an exception among the target models according to the execution log.

[0007] Optionally, the determining an abnormal model that has an exception among the target models according to the execution log includes: Traversing the multiple target models in the call order, determining the input data of the currently traversed target model according to the execution log, and detecting whether the currently traversed target model is an abnormal model that has an exception according to the input data of the currently traversed target model; In the case where the currently traversed target model is an abnormal model that has an exception, ending the traversal; In the case where the currently traversed target model is not an abnormal model that has an exception, continuing to execute the step of traversing the multiple target models in the call order.

[0008] Optionally, detecting whether the target model being currently traversed is an abnormal model with an anomaly includes: Performing compliance detection on the input data of the target model being currently traversed based on a set compliance policy library; and / or, performing semantic analysis on the input data of the target model being currently traversed and performing compliance detection on the semantic analysis result based on the set compliance policy library; Determining whether the target model being currently traversed is an abnormal model with an anomaly according to the compliance detection result of the input data.

[0009] Optionally, determining whether the target model being currently traversed is an abnormal model with an anomaly according to the compliance detection result of the input data includes: When the compliance detection result indicates that the output data of the target model being currently traversed is non-compliant and the input data is compliant, determining that the target model being currently traversed is an abnormal model with an anomaly.

[0010] Optionally, the method further includes: Obtaining at least one set compliance policy; Constructing the compliance policy library according to the at least one compliance policy.

[0011] Optionally, the method further includes: Obtaining intermediate data representing the processing process of the target model for processing the target subtask; Generating an execution log of the target processing task according to the intermediate data.

[0012] Optionally, the method further includes: When an anomaly occurs in the execution result of the target processing task, determining the reason for the anomaly of the abnormal model according to the intermediate data in the execution log.

[0013] Optionally, the method further includes: When the number of anomalies of the abnormal model within a set time period is greater than or equal to a threshold number of times, reducing the weight of the abnormal model in the multi-model collaboration system to reduce the invocation of the abnormal model.

[0014] According to a second aspect of the present disclosure, there is provided a log processing device for a multi-model collaboration system, including: A data acquisition module, configured to acquire output data of multiple target models in the multi-model collaboration system for processing corresponding target subtasks; wherein, the target subtask is a subtask obtained by splitting a target processing task, and the target model is a model for processing the target subtask; A log generation module, configured to generate an execution log of the target processing task according to the call order of the multiple target models and the output data. An exception location module, configured to determine an abnormal model with an exception in the target models according to the execution log in the case where an exception occurs in the execution result of the target processing task.

[0015] According to a third aspect of the present disclosure, there is provided an electronic device, including a processor and a memory, where the memory is configured to store a computer program, and the processor is configured to execute the method according to the first aspect of the present disclosure under the control of the computer program.

[0016] According to a fourth aspect of the present disclosure, there is provided a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the method according to the first aspect of the present disclosure.

[0017] Through this embodiment, according to the call order of the target models of the target subtasks obtained by splitting the target processing task in the multi-model collaboration system and the output data of the target models for processing the corresponding target subtasks, an execution log of the target processing task is generated, which can record the processing flow of each target processing task in the multi-model collaboration system. Also, when an error or exception occurs in the multi-model collaboration system, the execution log will record detailed error information, so as to facilitate exception tracing through the execution log.

[0018] Through the following detailed description of the exemplary embodiments of the present invention with reference to the accompanying drawings, other features and advantages of the present invention will become clear. Description of the Drawings

[0019] The drawings incorporated in the specification and constituting a part of the specification illustrate embodiments of the present invention and, together with the description, are used to explain the principles of the present invention.

[0020] Figure 1 is a block diagram showing the hardware configuration of a log processing system that can implement the embodiments of the present disclosure; Figure 2 is a flowchart of a log processing method for a multi-model collaboration system according to an embodiment of the present disclosure; Figure 3 is a schematic diagram of the model call order according to an embodiment of the present disclosure; Figure 4 is a block diagram of a log processing device for a multi-model collaboration system according to an embodiment of the present disclosure; Figure 5 is a block diagram of an electronic device according to an embodiment of the present disclosure. Detailed Embodiments

[0021] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangements of components and steps, numerical expressions, and numerical values set forth in these embodiments do not limit the scope of the present invention.

[0022] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way a limitation on the present invention or its application or use.

[0023] Technologies, methods, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and devices should be regarded as part of the specification.

[0024] In all the examples shown and discussed herein, any specific values should be construed as merely exemplary and not as a limitation. Thus, other examples of the exemplary embodiments may have different values.

[0025] It should be noted that like reference numerals and letters denote like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.

[0026] <Hardware Configuration> Figure 1 is a block diagram showing the hardware configuration of a log processing system 1000 that can implement the embodiments of the present disclosure.

[0027] The log processing system 1000 can be a laptop computer, a desktop computer, a mobile phone, a tablet computer, etc. As Figure 1 shown, the log processing system 1000 may include a processing device 1100, a storage device 1200, an interface device 1300, a communication device 1400, a display device 1500, an input device 1600, a speaker 1700, a microphone 1800, and so on. Among them, the processing device 1100 may be a processor CPU, a microprocessor MCU, etc. The storage device 1200 includes, for example, a ROM (read-only memory), a RAM (random access memory), a non-volatile memory such as a hard disk, etc. The interface device 1300 includes, for example, a USB interface, a headphone interface, etc. The communication device 1400 can perform wired or wireless communication, and specifically may include Wifi communication, Bluetooth communication, 2G / 3G / 4G / 5G communication, etc. The display device 1500 is, for example, a liquid crystal display screen, a touch display screen, etc. The input device 1600 may include, for example, a touch screen, a keyboard, a body sensor input, etc. Users can input / output voice information through the speaker 1700 and the microphone 1800.

[0028] Figure 1The log processing system shown is merely illustrative and in no way limits the present disclosure, its application, or its use. In the embodiments applied to the present disclosure, the storage device 1200 of the log processing system 1000 is used to store instructions for controlling the processing device 1100 to operate to execute any one of the methods provided in the embodiments of the present disclosure. Those skilled in the art should understand that although multiple devices are shown for the log processing system 1000 in Figure 1 , the present disclosure may only relate to some of the devices, for example, the log processing system 1000 only relates to the processing device 1100 and the storage device 1200. Those skilled in the art can design instructions according to the solutions disclosed in the present disclosure. How the instructions control the processing device to operate is well known in the art, so it will not be described in detail herein.

[0029] <Method Embodiment> The present disclosure provides a log processing method for a multi-model collaboration system. The log processing method for the multi-model collaboration system may be implemented by a log processing system. Specifically, it may be implemented by a log processing system 1000 as shown in Figure 1 .

[0030] Figure 2 FIG. is a flowchart of a log processing method for a multi-model collaboration system according to an embodiment of the present disclosure.

[0031] As shown in Figure 2 , the method includes steps S2100 to S2300 as follows: Step S2100, obtaining output data of multiple target models in the multi-model collaboration system for processing corresponding target subtasks; where the target subtask is a subtask obtained by splitting a target processing task, and the target model is a model for processing the target subtask.

[0032] In this embodiment, the multi-model collaboration system receives a processing request for a target processing task, performs preliminary parsing and preprocessing on the target processing task, and splits the target processing task into at least one target subtask; according to the type and characteristics of the target processing task, selects at least one target model from the model library to process the corresponding target subtask; the task scheduler reasonably arranges the execution order and resource allocation of the target model according to factors such as the availability of the target model, the current load situation, and the priority of the target subtask; the selected target model processes the corresponding target subtask sequentially or in parallel according to the arrangement of the scheduler, and transmits the output result to the result fusion module; the result fusion module fuses the results of multiple target models using a suitable method according to the specific target subtask and model collaboration method, and outputs the final result to the user in a specified format or a suitable manner.

[0033] In this embodiment, the target processing task can be of various types, such as image recognition, text generation, speech translation, etc. The target processing task can be generated based on the original data and task parameters input by the user. Among them, the original data can include data such as images, texts, and audios, and the task parameters can include the specified output format, specific task requirements, etc.

[0034] In this embodiment, multiple different models are stored in the model library, and each model has its specific function and scope of application. For example, for an image recognition task, models good at object detection and models good at image classification may be selected to cooperate together. When the task scheduler arranges the execution order and resource allocation of the target model, if there are multiple similar models available for selection, the scheduler may decide which model to use according to the performance metrics of the model (such as accuracy, recall, inference speed, etc.).

[0035] During the process that the selected target models process the corresponding target subtasks sequentially or in parallel according to the arrangement of the scheduler, the target models may cooperate in various ways. For example, the output of one target model may be used as the input of another target model, or multiple target models may process the same data simultaneously and then merge or fuse their respective results. The target models may perform multiple iterations and optimizations during the execution process, especially when dealing with complex tasks. For example, in a text generation task, one model may first generate a preliminary text, and then another model checks and polishes its grammar. After multiple such iterations, the quality of the text is gradually improved.

[0036] The methods for the result fusion module to fuse the results of multiple target models can include simple average, weighted average, voting, etc. For example, in an image classification task, multiple models classify the image respectively, and then determine the final category by voting. In addition, the result fusion module also evaluates the fused results to determine whether they meet the requirements of the task. The evaluation metrics can be determined according to the nature of the target processing task, such as accuracy, recall, F1 value, BLEU value (for text generation tasks such as machine translation), etc. If the results do not meet the requirements, the system may feedback the evaluation results to the model selection and scheduling module, reselect the model or adjust the parameters of the model, and perform the task processing again until satisfactory results are achieved.

[0037] The results output by the multi-model cooperation system can be in the forms of text, image, audio, etc. For example, in a speech translation task, the final output is the translated speech content. At the same time, the system may also provide some auxiliary information, such as the confidence of the results, relevant explanations, etc., to help users better understand and use the results.

[0038] In some embodiments, when the target model processes the corresponding target subtask to obtain output data, a digital signature of the service provider to which the target model belongs is attached to the output data to ensure the credibility of the output data.

[0039] Step S2200, generate an execution log of the target processing task according to the invocation order of multiple target models and the output data.

[0040] In this embodiment, it may be to record the task identifier of the target processing task, the model identifier and model version of the target model, the invocation order of the target model, and the output data to obtain the execution log of the target processing task.

[0041] In some embodiments, each model in the model library has a service provider identifier identifying its service provider, a model identifier identifying the model, and a model version. When any target model is invoked to process the corresponding target subtask, the service provider identifier, model identifier, model version, output data, and timestamp of the target model can be associated as the execution record of the target processing task corresponding to the target model.

[0042] Further, according to the invocation order of the target models, the execution records of the target processing task corresponding to all target models are integrated, and thus the execution log of the target processing task can be obtained.

[0043] In some embodiments, the way to integrate the execution records of the target processing task corresponding to all target models according to the invocation order of the target models may be to add the model identifier of the previous target model of the target model (the previous target model may be a model in the model library or the initial input) to the execution records of the target processing task corresponding to all target models.

[0044] In some embodiments, the execution record of the target processing task corresponding to the target model can also be generated according to the timestamp when the target model processes the corresponding target subtask.

[0045] Step S2300, when an exception occurs in the execution result of the target processing task, determine the abnormal model that appears abnormal in the target models according to the execution log.

[0046] Through this embodiment, according to the invocation order of the target models that process the target subtasks split from the target processing task in the multi-model collaboration system and the output data of the target models processing the corresponding target subtasks, an execution log of the target processing task is generated, which can record the processing flow of each target processing task in the multi-model collaboration system. Also, when an error or exception occurs in the multi-model collaboration system, detailed error information will be recorded in the execution log, so as to facilitate anomaly tracing through the execution log.

[0047] In this embodiment, it may be determined that the execution result of the target processing task is abnormal in response to a negative operation on the execution result of the target processing task by the user. Alternatively, it may also be to perform a compliance check on the execution result of the target processing task, and determine that the execution result of the target processing task is abnormal when it is detected that the execution result of the target processing task does not comply with the regulations.

[0048] In some embodiments, determining the abnormal model with an abnormality in the target model according to the execution log may include: traversing multiple target models in the call order, determining the input data of the currently traversed target model according to the execution log, and detecting whether the currently traversed target model is an abnormal model with an abnormality according to the input data of the currently traversed target model; when the currently traversed target model is an abnormal model with an abnormality, end the traversal; when the currently traversed target model is not an abnormal model with an abnormality, continue to execute the step of traversing multiple target models in the call order.

[0049] In this embodiment, the input data of the currently traversed target model may include the output data of the corresponding previous target model, or may also include the initial input.

[0050] In this embodiment, if the input of a first target model depends on the outputs of at least two second target models, then, after traversing the first target model and determining that the first target model is not an abnormal model, it may be to sequentially traverse at least two second target models, and the order of traversing at least two second target models may be the order of the corresponding timestamps in the execution log, and this embodiment does not make a limitation in this regard.

[0051] If the inputs of at least two third models depend on the output of a fourth target model, and the input data of the at least two third models is the same, then it may be to traverse any one of the third models, perform an abnormality detection, and use the abnormality result of the traversed third model as the abnormality detection result of the other third models.

[0052] In this embodiment, it may be to traverse multiple target models in the call order from first to last, or it may also be to traverse multiple target models in the call order from last to first, and no limitation is made here.

[0053] In some embodiments, detecting whether the currently traversed target model is an abnormal model with an abnormality according to the input data of the currently traversed target model may include: performing a compliance check on the input data of the currently traversed target model based on the compliance policy library; determining whether the currently traversed target model is an abnormal model with an abnormality according to the compliance check result of the input data.

[0054] In this embodiment, multiple compliance policies can be pre-organized to construct a compliance policy library. Among them, the compliance policies can include laws and regulations, social customs and ethics, and policies with correct facts.

[0055] Based on the compliance policy library, perform compliance detection on the input data of the currently traversed target model, which can be to detect whether the input data can pass all the compliance policies in the compliance policy library. If so, determine that the input data is compliant; if not, determine that the input data is non-compliant.

[0056] Through this embodiment, it is possible to detect whether the input data itself is compliant, and further detect whether the currently traversed target model is an abnormal model with anomalies.

[0057] In some embodiments, according to the input data of the currently traversed target model, detecting whether the currently traversed target model is an abnormal model with anomalies may include: performing semantic analysis on the input data of the currently traversed target model, and performing compliance detection on the semantic analysis result based on the compliance policy library; according to the compliance detection result of the input data, determine whether the currently traversed target model is an abnormal model with anomalies.

[0058] In this embodiment, semantic analysis can be performed on the input data of the currently traversed target model based on a semantic analysis model to obtain a corresponding semantic analysis result, and then the compliance policy library performs compliance detection on the semantic analysis result.

[0059] In the case where the semantic analysis result of the input data is detected to be non-compliant, it can be determined that the compliance detection result is that the input data is non-compliant; in the case where the semantic analysis result of the input data is detected to be compliant, it can be determined that the compliance detection result is that the input data is compliant.

[0060] Through this embodiment, it is possible to further detect whether the semantics of the input data is compliant, and further detect whether the currently traversed target model is an abnormal model with anomalies.

[0061] In some embodiments, according to the compliance detection result of the input data, determining whether the currently traversed target model is an abnormal model with anomalies includes: in the case where the compliance detection result is that the output data of the currently traversed target model is non-compliant and the input data is compliant, determining that the currently traversed target model is an abnormal model with anomalies.

[0062] In this embodiment, the abnormal model may be a target model with compliant input data and non-compliant output data. Since this embodiment traverses multiple target models in the call order, if the input data of the previously traversed target model is non-compliant, which means the output data of the currently traversed target model is non-compliant, then it is only necessary to detect whether the input data of the currently traversed target model is compliant, that is, to detect whether the currently traversed target model is an abnormal model.

[0063] Through this embodiment, the input data of each target model is determined according to the execution log. Based on the input data of each target model, when the execution result of the target processing task is abnormal, the abnormal model among the multiple target models can be located.

[0064] In one embodiment, if the input of a first target model depends on the outputs of at least two second target models, then when the input data (i.e., the output data of the second target model) of any one of the second target models or its semantic analysis result is non-compliant, the second target model can be continuously traversed, and the second target model with compliant input data and its semantic analysis result need not be traversed any further. When the input data or its semantic analysis result of at least two second target models is non-compliant, the at least two second target models can be traversed in sequence.

[0065] If the inputs of at least two third models depend on the output of a fourth target model, then when traversing any one of the third target models and determining that the input data or its semantic analysis result of the third target model is non-compliant, the other third target models need not be traversed any further, and the fourth target model can be continuously traversed. When traversing any one of the third target models and determining that the input data and its semantic analysis result of the third target model are compliant, it can be determined that the third target model is an abnormal model, and the other third target models need not be traversed any further, that is, the traversal ends.

[0066] In an example, the target models may include Model A to Model E, and the call order of Model A to Model E may be as Figure 3 shown. Then, it is possible to detect the input data of Model E, which has an abnormal result. Since the output data of Model E is abnormal, it may be that Model C or Model D provided abnormal input data for Model E, or it may be that Model E itself is abnormal.

[0067] If the input data provided by Model C and Model D to Model E are both normal, then it can be determined that Model E is an abnormal model; if the output data of Model C or Model D is abnormal, it can be detected whether the input data provided by Detection Model B to Model C and Model D is abnormal. If the input data provided by Model B to Model C and Model D is normal, then it can be determined that Model C or Model D with abnormal output data is the abnormal model (specifically, the model with abnormal output data); if the input data provided by Model B to Model C and Model D is abnormal, then it is detected whether the input data provided by Model A to Model B is abnormal. If the input data provided by Model A to Model B is normal, it can be determined that Model B is an abnormal model; if the input data provided by Model A to Model B is abnormal, it can be detected whether the input data of Model A is abnormal. If the input data of Model A is abnormal, then it can be determined that the abnormal execution result of the target processing task is caused by an abnormality in the target processing task itself; if the input data of Model A is normal, then it can be determined that Model A is an abnormal model.

[0068] In this embodiment, by constructing a multi-layer verifiable traceability system, accurate problem positioning in the scenario of multi-model collaboration is achieved, while balancing system performance and security requirements.

[0069] In some embodiments, the method may further include: obtaining intermediate data representing the processing process of the target model for processing the target subtask; and generating an execution log of the target processing task according to the intermediate data.

[0070] Specifically, an execution record of the target processing task corresponding to the target model may be generated according to the intermediate data representing the processing process of the target model for processing the target subtask.

[0071] In this embodiment, the intermediate data may include a retrieval path and / or a thought chain corresponding to the target model.

[0072] Through this embodiment, the obtained execution log can be made more comprehensive, facilitating subsequent traceability and anomaly analysis when the execution result is abnormal.

[0073] In some embodiments, a circular buffer may be used to store the generated execution log, so that when performing anomaly traceability, the model where the anomaly occurs and the cause of the anomaly can be quickly located according to the execution log stored in the circular buffer.

[0074] In some embodiments, it may also be to store only the execution log of the target processing task with an abnormal execution result into the circular buffer to further accelerate the time of anomaly traceability.

[0075] In some embodiments, the method further includes: when an exception occurs in the execution result of a target processing task, detecting the reason for the exception of the exception model according to the intermediate data of the exception model in the execution log.

[0076] In this embodiment, it may be to perform compliance detection on the retrieval path and / or the thinking chain of the exception model, and obtain the reason for the exception of the exception model according to the detection result.

[0077] Further, performing compliance detection on the retrieval path of the exception model may be to detect the content and / or domain name of the retrieval path for compliance detection. When any retrieval path is non-compliant, it can be obtained that the reason for the exception of the exception model includes non-compliance of the retrieval path.

[0078] On this basis, the non-compliant retrieval path can be added to the blacklist, and the service providers of each model in the multi-model collaboration system can be notified to reduce the processing of tasks according to the non-compliant retrieval path.

[0079] Still further, when the compliance detection result of the thinking chain of the exception model is detected to be non-compliant, it can be obtained that the reason for the exception of the exception model includes non-compliance of the thinking chain.

[0080] On this basis, the method may further include: when the number of exceptions of the exception model within a set period is greater than or equal to the number threshold, reducing the weight of the exception model in the multi-model collaboration system to reduce the invocation of the exception model.

[0081] In this embodiment, it may be when the number of exceptions of non-compliance of the thinking chain of the exception model within a set period is greater than or equal to the number threshold, reducing the weight of the exception model in the multi-model collaboration system to reduce the invocation of the exception model.

[0082] Through this embodiment, the accuracy of the task processing result of the multi-model collaboration system can be improved.

[0083] In this embodiment, the set period may be set in advance according to the application scenario or specific requirements. For example, the set period may be the past 3 days.

[0084] Further, the number threshold may be set in advance according to the application scenario or specific requirements. For example, the number threshold may be 10 times.

[0085] In this embodiment, according to the number of exceptions of non-compliance of the thinking chain of each model within a set period, the service stability of the corresponding model is evaluated, and a higher weight is given to the model with better stability. When calling this type of model, the model with a higher score is given priority. In addition, a higher credibility weight can also be given to the output of the model with higher stability.

[0086] In some embodiments, the method may further include: analyzing the execution logs of the target processing tasks with abnormal execution results to determine the association relationships between multiple fields. The multiple fields may include model identifiers, model versions, knowledge bases, etc. When the occurrence times of an association relationship within a set time period are greater than or equal to a threshold number of times, the invocation of this association relationship is reduced.

[0087] For example, for the association relationship corresponding to model A, model version V1.2, and knowledge base ANIMAL, when the occurrence times within a set time period are greater than or equal to the threshold number of times, the invocation of model A of version V1.2 to the knowledge base ANIMAL can be reduced.

[0088] In some embodiments, the compliance policy library may also be updated according to the execution results of the target processing tasks to detect non-compliant data in a timely manner.

[0089] <Device Embodiment> The present disclosure also provides a log processing device for a multi-model collaboration system, as Figure 4 shown. The log processing device 4000 may include a data acquisition module 4100, a log generation module 4200, and an exception location module 4300.

[0090] The data acquisition module 4100 is configured to acquire the output data of multiple target sub-tasks processed by multiple target models in the multi-model collaboration system; wherein, the target sub-tasks are sub-tasks obtained by splitting the target processing task, and the target models are models for processing the target sub-tasks.

[0091] The log generation module 4200 is configured to generate the execution log of the target processing task according to the invocation order of the multiple target models and the output data.

[0092] The exception location module 4300 is configured to determine the abnormal model that appears abnormal among the target models according to the execution log when the execution result of the target processing task appears abnormal.

[0093] In some embodiments, determining the abnormal model that appears abnormal among the target models according to the output data in the execution log includes: Traversing the multiple target models in the invocation order, determining the input data of the currently traversed target model according to the execution log, and detecting whether the currently traversed target model is the abnormal model that appears abnormal according to the input data of the currently traversed target model; When the currently traversed target model is the abnormal model that appears abnormal, end the traversal; In the case that the currently traversed target model is not an abnormal model with an abnormality, continue to execute the step of traversing the multiple target models in the call order.

[0094] In some embodiments, detecting whether the currently traversed target model is an abnormal model with an abnormality according to the input data of the currently traversed target model includes: Performing compliance detection on the input data of the currently traversed target model based on a set compliance policy library; and / or, performing semantic analysis on the input data of the currently traversed target model and performing compliance detection on the semantic analysis result based on the set compliance policy library; Determining whether the currently traversed target model is an abnormal model with an abnormality according to the compliance detection result of the input data.

[0095] In some embodiments, determining whether the currently traversed target model is an abnormal model with an abnormality according to the compliance detection result of the input data includes: In the case that the compliance detection result is that the output data of the currently traversed target model is non-compliant and the input data is compliant, determining that the currently traversed target model is an abnormal model with an abnormality.

[0096] In some embodiments, the log processing device 4000 further includes: A module for obtaining at least one set compliance policy; A module for constructing the compliance policy library according to the at least one compliance policy.

[0097] In some embodiments, the data acquisition module 4100 is further configured to: acquire intermediate data representing the processing process of the target model for processing the target subtask; the log generation module 4200 is further configured to generate an execution log of the target processing task according to the intermediate data.

[0098] In some embodiments, the log processing device 4000 further includes: A module for determining the reason for the abnormality of the abnormal model according to the intermediate data in the execution log in the case that the execution result of the target processing task is abnormal.

[0099] In some embodiments, the log processing device 4000 further includes: A module for reducing the weight of the abnormal model in the multi-model collaboration system to reduce the invocation of the abnormal model in the case that the number of times the abnormal model has an abnormality within a set time period is greater than or equal to a threshold number of times.

[0100] <Embodiment of the electronic device> This embodiment provides an electronic device. On the one hand, the electronic device may include the aforementioned log processing device 4000.

[0101] On the other hand, as Figure 5 shown, the electronic device 5000 may include a processor 5100 and a memory 5200. The memory 5200 is used to store a computer program, and the processor 5100 is used to control the electronic device to execute the methods of any embodiment of the present disclosure under the control of the computer program.

[0102] <Readable Storage Medium Embodiment> This embodiment provides a computer-readable storage medium in which a computer program is stored. When the computer program is executed by a processor, it executes the methods described in any method embodiment of the present disclosure.

[0103] The present invention may be a system, a method, and / or a computer program product. The computer program product may include a computer-readable storage medium having thereon computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0104] A computer-readable storage medium may be a tangible device that can retain and store instructions for use by an instruction execution device. A computer-readable storage medium may be, for example, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (non-exhaustive list) of the computer-readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disc (DVD), a memory stick, a floppy disk, a mechanically encoded device, such as a punched card or raised structures in a groove having instructions stored thereon, and any suitable combination of the foregoing. The computer-readable storage medium used herein is not construed as a transitory signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated through a waveguide or other transmission medium (e.g., an optical pulse through an optical fiber cable), or an electrical signal transmitted through a wire.

[0105] The computer-readable program instructions described herein can be downloaded to various computing / processing devices from a computer-readable storage medium or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing / processing device.

[0106] The computer program instructions for carrying out operations of the present invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state-setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider). In some embodiments, by using the state information of the computer-readable program instructions to customize an electronic circuit, such as a programmable logic circuit, a field-programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can execute the computer-readable program instructions to implement various aspects of the present invention.

[0107] Aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0108] These computer-readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine such that the instructions, when executed by the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in one or more boxes of the flowchart and / or block diagram. These computer-readable program instructions may also be stored in a computer-readable storage medium that causes a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer-readable medium storing the instructions comprises a manufacture including instructions for implementing various aspects of the functions / acts specified in one or more boxes of the flowchart and / or block diagram.

[0109] The computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process such that the instructions executed on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in one or more boxes of the flowchart and / or block diagram.

[0110] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram may represent a module, a segment of code, or a portion of an instruction, which comprises one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions noted in the boxes may occur out of the order noted in the figures. For example, two consecutive boxes may in fact be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending upon the functionality involved. It should also be noted that each box of the block diagrams and / or flowcharts, and combinations of boxes in the block diagrams and / or flowcharts, may be implemented by special purpose hardware-based systems that perform the specified functions or acts, or by combinations of special purpose hardware and computer instructions. As will be apparent to those skilled in the art, implementation by hardware, implementation by software, and implementation by a combination of software and hardware are all equivalent.

[0111] The embodiments of the present invention have been described above. The above description is exemplary and not exhaustive, and is also not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to technologies in the market, or to enable other ordinary skill in the art to understand the embodiments disclosed herein. The scope of the present invention is defined by the appended claims.

Claims

1. A method for processing logs of a multi-model collaboration system, characterized in that, Including: Obtaining output data of multiple target models in a multi-model collaboration system for processing corresponding target subtasks; wherein, the target subtasks are subtasks obtained by splitting a target processing task, and the target models are models for processing the target subtasks; Generating an execution log of the target processing task according to the call order of the multiple target models and the output data; When an exception occurs in the execution result of the target processing task, determining an abnormal model that has an exception among the target models according to the execution log; 2. The method according to claim 1, characterized in that, The determining an abnormal model that has an exception among the target models according to the execution log includes: Traversing the multiple target models in the call order, determining the input data of the currently traversed target model according to the execution log, and detecting whether the currently traversed target model is an abnormal model with an exception according to the input data of the currently traversed target model; When the currently traversed target model is an abnormal model with an exception, ending the traversal; When the currently traversed target model is not an abnormal model with an exception, continuing to execute the step of traversing the multiple target models in the call order; 3. The method according to claim 2, wherein The detecting whether the currently traversed target model is an abnormal model with an exception according to the input data of the currently traversed target model includes: Based on a set compliance policy library, performing compliance detection on the input data of the currently traversed target model; and / or, performing semantic analysis on the input data of the currently traversed target model and performing compliance detection on the semantic analysis result based on the compliance policy library; wherein, the compliance policy library includes at least one compliance policy; Determining whether the currently traversed target model is an abnormal model with an exception according to the compliance detection result of the input data; 4. The method according to claim 3, wherein The determining whether the currently traversed target model is an abnormal model with an exception according to the compliance detection result of the input data includes: When the compliance detection result is that the output data of the currently traversed target model is non-compliant and the input data is compliant, determining that the currently traversed target model is an abnormal model with an exception; 5. The method according to claim 3, wherein The method further includes: Obtaining at least one set compliance policy; Constructing the compliance policy library according to the at least one compliance policy; 6. The method according to claim 1, wherein The method further includes: Obtaining intermediate data representing the processing process of the target model for processing the target subtask; Generating an execution log of the target processing task according to the intermediate data; 7. The method according to claim 6, characterized in that The method further includes: When an exception occurs in the execution result of the target processing task, determining the reason for the exception of the abnormal model according to the intermediate data in the execution log; 8. The method according to claim 1, wherein The method further includes: When the number of times the abnormal model has an exception within a set time period is greater than or equal to a threshold number of times, reducing the weight of the abnormal model in the multi-model collaboration system to reduce the call of the abnormal model; 9. A log processing device for a multi-model collaboration system, characterized in that, Including: A data acquisition module, configured to acquire output data of multiple target models processing corresponding target subtasks in a multi-model collaboration system; wherein, the target subtasks are subtasks obtained by splitting a target processing task, and the target models are models for processing the target subtasks; A log generation module, configured to generate an execution log of the target processing task according to the call order of the multiple target models and the output data; An exception location module, configured to determine an exception model with an exception among the target models according to the execution log when an exception occurs in the execution result of the target processing task.

10. An electronic device, characterized in that, It includes a processor and a memory. The memory is used to store a computer program, and the processor is configured to execute the method according to any one of claims 1 to 8 under the control of the computer program.

Images