Data processing method and apparatus
By determining the target parameters and selecting a target model that meets the conditions from multiple large models, the problem of excessive computational resources and time consumption in the existing technology is solved, and fast and accurate large model selection and processing are achieved.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- LENOVO (BEIJING) LTD
- Filing Date
- 2026-01-20
- Publication Date
- 2026-07-30
AI Technical Summary
When using large models to process input information, existing technologies require complex analysis and processing of the target input information, resulting in excessive consumption of computing resources and time, and making it difficult to quickly select a suitable large model for output.
By obtaining target input information, target parameters are determined, including the first parameter representing the model task type and the second parameter of other dimensions. Thus, a target first model that meets the parameter conditions is selected from multiple large models for output, and a second model is used for preliminary classification to reduce computational complexity.
It enables the rapid selection of a suitable large model for processing without relying on the capabilities of the large model itself, reducing the demand for computing resources and time consumption, and improving processing accuracy.
Smart Images
Figure CN2026073617_30072026_PF_FP_ABST
Abstract
Description
Data processing method and apparatus
[0001] This application claims priority to Chinese Patent Application No. 202510104515.4, filed on January 22, 2025, entitled “Data Processing Method and Apparatus”, the entire contents of which are incorporated herein by reference. Technical Field
[0002] This application relates to, but is not limited to, the field of data processing technology, and in particular to a data processing method and apparatus. Background Technology
[0003] Large models refer to machine learning models with a large number of parameters and complex computational structures. Different large models have different capabilities. Therefore, selecting a large model that matches the input information is crucial to the output results. Summary of the Invention
[0004] This application provides at least one data processing method and apparatus.
[0005] The technical solution of this application embodiment is implemented as follows:
[0006] This application provides a data processing method applied to a target application, including:
[0007] Obtain the target input information;
[0008] The target parameters are determined based on the target input information. The target parameters include a first parameter and at least one second parameter. The first parameter represents the task type of the corresponding model, and the second parameter is different from the first parameter.
[0009] Based on the target parameters, a target first model that meets the parameter conditions is determined from multiple first models. The target first model is used to output based on the target input information.
[0010] Multiple first models constitute a large model, and the first parameters of multiple first models are the same, while at least one second parameter is different.
[0011] This application provides a data processing apparatus, including:
[0012] The acquisition module is used to obtain target input information;
[0013] The first determining module is used to determine target parameters based on target input information; the target parameters include a first parameter and at least one second parameter, the first parameter characterizing the task type of the corresponding model, and the second parameter being different from the first parameter;
[0014] The second determining module is used to determine the target first model that meets the parameter conditions from multiple first models based on the target parameters. The target first model is used to output based on the target input information.
[0015] Multiple first models constitute a large model, and the first parameters of multiple first models are the same, while at least one second parameter is different.
[0016] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of this application. Attached Figure Description
[0017] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application.
[0018] Figure 1 is a schematic diagram of the implementation flow of a data processing method provided in an embodiment of this application;
[0019] Figure 2 is a schematic diagram of an implementation process for determining target parameters provided in an embodiment of this application;
[0020] Figure 3 is a schematic diagram of an implementation process for selecting a target first model from multiple first models according to an embodiment of this application;
[0021] Figure 4 is a schematic diagram of a method for selecting a large model provided in an embodiment of this application;
[0022] Figure 5 is a schematic diagram of the composition structure of a data processing device provided in an embodiment of this application;
[0023] Figure 6 is a schematic diagram of the hardware entity of a computer device provided in an embodiment of this application. Detailed Implementation
[0024] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application are further described in detail below with reference to the accompanying drawings and embodiments. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0025] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.
[0026] It should be noted that the terms "first, second, and third" used in the embodiments of this application are merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.
[0027] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of this application pertain. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.
[0028] In related technologies, when using large models to process input information, there are multiple large models with outstanding capabilities in different dimensions. The large models use their own capabilities to select the appropriate large model to process and output the input information based on the input information.
[0029] Based on the above description, embodiments of this application provide a method for quickly selecting a suitable large model from multiple large models, by measuring at least one parameter of the large model to select a suitable large model to process the input information.
[0030] This application provides a data processing method applied to a target application. Figure 1 is a schematic diagram of the implementation flow of a data processing method provided in this application. As shown in Figure 1, the method includes the following steps S101 to S103:
[0031] Step S101: Obtain target input information.
[0032] The target input information is the input information that meets the input conditions.
[0033] Here, the target application can be a software program used to perform a specific task or provide a specific service. The software program can run on various computing platforms, including but not limited to at least one of desktop computers, laptops, tablets, smartphones, and other smart devices.
[0034] In some implementations, the target application does not have model reasoning capabilities; that is, the target application is not a Large Model (LM) or a Large Language Model (LLM). The target application is a prerequisite step for the target first model reasoning.
[0035] When selecting a suitable large model using a large model or large language model, it depends on the algorithm and logic of the large model itself, as well as the pre-trained model library or knowledge base. Selecting a large model requires complex analysis and processing of the target input information, which consumes more computing resources and time.
[0036] When selecting a suitable large model using the target application, there is no need to call the large model or large language model, which can reduce the inference cost of the large model.
[0037] In some implementations, the input methods for the target input information may include, but are not limited to, users entering information in a search box, users clicking links or options, and information stored in a database.
[0038] The target input information can be at least one of the following: text information, image information, video information, audio information, etc.
[0039] The target input information is the input information that meets the input conditions.
[0040] In some implementations, the input conditions include: the existence of multiple first models that support processing tasks of the task type corresponding to the target input information, and the multiple first models have at least one different second parameter, such as different model accuracies.
[0041] In some implementations, input conditions include: the data format of the target input information meets the input format of the large model. For example, the input format of a large language model may include, but is not limited to, specific text formats, image resolutions, or audio sampling rates.
[0042] In some implementations, the target input information may include input information and / or prompts obtained by analyzing and summarizing the input information.
[0043] Step S102: Determine target parameters based on the target input information; the target parameters include a first parameter and at least one second parameter, the first parameter characterizing the task type of the corresponding model, and the second parameter being different from the first parameter.
[0044] Here, the target parameters characterize the capabilities required for a large model to process the target input information. For example, the first parameter characterizes the types of tasks supported by a large model capable of processing the target input information.
[0045] Task types can include, but are not limited to, translation tasks, text classification tasks, text-to-image conversion tasks, etc. For example, the first parameter representation can support the processing of target input information. Large models can support the processing of translation tasks.
[0046] In some implementations, the target input information is analyzed to match the task type corresponding to the target input information, i.e., the first parameter. For example, if there are multiple task types, the probability of the target input information matching each of the multiple task types is determined based on the target input information, and the task type with the highest probability of matching the target input information is selected from the multiple task types as the task type represented by the first parameter.
[0047] The target input information is analyzed, and the second parameter corresponding to the target input information is matched under the second parameter dimension. For example, there are multiple parameters, including the second parameter, under various second parameter dimensions. Based on the target input information, the probability of the target input information matching each of the multiple parameters is determined, and the parameter with the highest probability of matching the target input information is selected as the second parameter.
[0048] In some implementations, the second parameter includes, but is not limited to, at least one of language type, knowledge domain, model accuracy, security level, etc.
[0049] Step S103: Based on the target parameters, determine the target first model that meets the parameter conditions from multiple first models. The target first model is used to output based on the target input information.
[0050] Here, multiple first models are large models, and these multiple first models have the same first parameter, but at least one different second parameter. All multiple first models support processing tasks of the task type represented by the first parameter.
[0051] The parameter conditions are the conditions that the target first model must meet to select from multiple first models. For example, if the target parameters include the first parameter, since the first parameter represents the task type of the corresponding model, the parameter conditions are the parameter conditions corresponding to the task type dimension; if the target parameters include the second parameter, the parameter conditions are the parameter conditions corresponding to the second parameter dimension.
[0052] If the first model satisfies the parameter conditions corresponding to the task type dimension and the parameter conditions corresponding to the second parameter dimension, then the first model is determined as the target first model.
[0053] In some implementations, based on a first parameter, a plurality of first models that satisfy the parameter conditions corresponding to the task type dimension are determined from a plurality of models; based on a second parameter, a target first model that satisfies the parameter conditions corresponding to the second parameter dimension is determined from a plurality of first models.
[0054] In some implementations, a model that satisfies the parameter conditions corresponding to the task type dimension supports processing tasks of the task type represented by the first parameter; a model that satisfies the parameter conditions corresponding to the second parameter dimension supports processing tasks with the second parameter.
[0055] In some implementations, target parameters are assigned corresponding weights based on their importance to the target input information; and a target first model is selected from multiple first models based on the target parameters and their corresponding weights.
[0056] In this embodiment, target parameters are determined based on target input information that meets the input conditions. The first parameter in the target parameters represents the task type of the corresponding model, and the second parameter is different from the first parameter. A target first model that meets the parameter conditions is determined from multiple first models based on the target parameters. The target first model is used to output based on the target input information. The multiple first models are large models, and the first parameters of the multiple first models are the same, while at least one second parameter is different. Thus, when selecting a target first model, it is only necessary to select the target first model that meets the parameter conditions from multiple first models with the same first parameter, based on the task type supported by the corresponding model represented by the first parameter and the capabilities required by the corresponding model represented by at least one second parameter different from the first parameter. This allows the target first model to better meet the task requirements corresponding to the target input information, thereby improving the accuracy of processing the target input information using the target first model and reducing the inference cost of the large model.
[0057] In some embodiments, the step S102 described above, which involves determining the target parameters based on the target input information, may include the following step S111:
[0058] Step S111: Using the second model, determine the target parameters based on the target input information; the magnitude of the processing parameters of the second model is smaller than the magnitude of the processing parameters of the first model.
[0059] In this embodiment, although the target application does not have model reasoning capabilities, it does have model invocation capabilities, and further, it has the ability to invoke a second model.
[0060] Here, the magnitude of the processing parameters of the second model is on the normal order of magnitude, which is smaller than that of the processing parameters of the first model.
[0061] In some implementations, large models are massive in scale, with the number of processing parameters reaching over 100 million, and they employ deep learning structures such as Transformer, which have significant advantages in processing long sequences and capturing contextual information in language.
[0062] Compared to the large model, the second model processes far fewer than 100 million parameters and has a relatively simple structure. It typically employs simpler neural network structures or statistical methods, such as logistic regression and Naive Bayes.
[0063] In some implementations, large models are capable of handling multiple complex tasks, including various natural language processing tasks such as text generation, machine translation, question answering systems, and text summarization.
[0064] The second model is generally single-task, designed and trained for a specific natural language processing task, such as text classification.
[0065] In some implementations, the large model has powerful text generation capabilities and can automatically generate output text based on the input information; the classifier in the second model classifies the input information based on predefined rules or templates and outputs the corresponding probabilities.
[0066] In some implementations, the text classification capability of the second model is utilized to classify the target input information, obtain the corresponding probability, and determine the corresponding target parameters based on the probability.
[0067] In some implementations, the second model can be deployed in the cloud or locally, and the target application calls the second model from the cloud or local to determine the target parameters based on the target input information.
[0068] In some implementations, the second model is trained using multiple sample inputs and corresponding sample parameters in the second parameter dimension. The sample parameters can be second parameters from historical data corresponding to the sample inputs, or manually determined second parameters corresponding to the sample inputs. Because the second model is smaller, its training time is shorter, reducing training time compared to selecting a suitable large model.
[0069] In this embodiment, a second model is used to determine target parameters based on target input information. The number of parameters processed by the second model is smaller than that processed by the first model. Thus, when the second model is deployed locally, its smaller size reduces memory usage and improves computational efficiency, thereby reducing computational resource requirements. When the second model is deployed in the cloud, the high-speed network connection and data transmission mechanisms of the cloud improve the efficiency of the target application in calling the second model to determine target parameters.
[0070] In some embodiments, FIG2 is a schematic diagram of an implementation process for determining target parameters provided by an embodiment of this application. As shown in FIG2, the determination of target parameters based on the target input information using the second model in step S111 above may include the following steps S121 and S122:
[0071] Step S121: Using the second model, classify the target input information in the task type dimension to determine the first parameter.
[0072] In some implementations, a second model is used to classify the target input information in the task type dimension, determine the probability that the target input information falls into each task type in the task type dimension, and determine the first parameter based on the probability corresponding to each task type.
[0073] Step S122: Using the second model, classify the target input information under the parameter dimension corresponding to the second parameter, and determine the second parameter.
[0074] In some implementations, a second model is used to classify the target input information under the parameter dimensions corresponding to the second parameter, determine the probability that the target input information falls into each parameter of the corresponding parameter dimension under multiple parameter dimensions, and determine the second parameter based on the probability corresponding to each parameter.
[0075] In this embodiment, a second model is used to classify the target input information under the task type dimension to determine the first parameter, and to classify the target input information under the parameter dimension corresponding to the second parameter to determine the second parameter. Thus, by utilizing the classification capability of the second model, the target input information can be classified under the task type dimension and the parameter dimension corresponding to the second parameter, respectively, yielding classification results for different dimensions.
[0076] In some embodiments, FIG3 is a schematic diagram of the implementation process of selecting a target first model from multiple first models according to an embodiment of the present application. As shown in FIG3, the step S103 above, which involves determining the target first model that meets the parameter conditions from multiple first models based on the target parameters, may include the following steps S131 and S132:
[0077] Step S131: Based on the first parameter, determine multiple first models that satisfy the parameter conditions corresponding to the task type dimension.
[0078] In some implementations, the task type dimension can include multiple task types, such as generative tasks, translation tasks, and classification tasks. Generative tasks may include, but are not limited to, at least one of text generation tasks, image generation tasks, video generation tasks, and audio generation tasks. Text generation tasks may include, but are not limited to, at least one of text-to-text tasks and image-to-text tasks. Image generation tasks may include, but are not limited to, at least one of text-to-image tasks and image-to-image tasks.
[0079] When the task type represented by the first parameter is a translation task, and each model supports processing translation tasks, the model that satisfies the parameter conditions corresponding to the task type dimension is determined to be the first model.
[0080] Step S132: Based on the second parameter, determine the target first model from multiple first models that satisfies the parameter conditions corresponding to the dimension of the second parameter.
[0081] In some implementations, the target first model that satisfies the parameter conditions can be the first model among multiple first models that has the highest support capability for tasks with second parameters.
[0082] In some implementations, the target first model that satisfies the parameter conditions can be the first model among multiple first models whose support capability for tasks with second parameters is greater than a preset threshold.
[0083] In this embodiment, multiple first models that satisfy the parameter conditions corresponding to the task type dimension are determined based on a first parameter; a target first model that satisfies the parameter conditions corresponding to the second parameter dimension is determined from the multiple first models based on a second parameter. Thus, by fully considering both the first and second parameter dimensions, a target first model that supports processing tasks with the second parameter is selected from multiple first models that support processing tasks representing the task type of the first parameter.
[0084] In some embodiments, the step S131 above, which involves determining multiple first models that satisfy the parameter conditions corresponding to the task type dimension based on the first parameter, may include the following steps S141 and S142:
[0085] Step S141: Obtain first data corresponding to multiple models under the first parameter; the first data characterizes the model's ability to support the task type corresponding to the first parameter.
[0086] In some implementations, the support capabilities of multiple models for various task types are predetermined, and the support capabilities of the model for the task type represented by the first parameter are determined based on the task type represented by the first parameter, i.e., the first data.
[0087] The first data can be described using corresponding scores. For example, multiple models can be scored in advance based on their ability to support multiple task types to obtain the first data.
[0088] The first data can be described using the corresponding accuracy rate. For example, the first data can be obtained by pre-processing multiple tasks using multiple models.
[0089] Step S142: Based on the first data, select multiple first models from each of the models that satisfy the first target data conditions corresponding to the task type dimension.
[0090] In some implementations, the model is determined to meet the first target data conditions corresponding to the task type dimension based on the first data.
[0091] In some implementations, if the model processes a task of the type represented by the first parameter and the corresponding score is greater than the first score threshold, it is determined that the model meets the first target data condition corresponding to the task type dimension.
[0092] Based on the model's processing of the task type represented by the first parameter, the corresponding score is determined. Multiple models are then ranked, and a first preset number of models that meet the first target data condition corresponding to the task type dimension are determined.
[0093] In some implementations, if the accuracy of the model in processing the task of the task type represented by the first parameter is greater than the first accuracy threshold, it is determined that the model meets the first target data condition corresponding to the task type dimension.
[0094] Based on the accuracy of the model in processing the task type represented by the first parameter, multiple models are ranked, and a first preset number of models that meet the first target data condition corresponding to the task type dimension are determined.
[0095] In some implementations, if it is determined that the model satisfies the first target data condition corresponding to the task type dimension, multiple first models can support the processing of tasks of the task type represented by the first parameter.
[0096] In this embodiment, based on the model's ability to support the task type corresponding to the first parameter, multiple first models that satisfy the first target data conditions corresponding to the task type dimension are selected from among the various models. This allows for consideration from the task type dimension, ensuring that the selected target first model supports processing tasks of the task type represented by the first parameter.
[0097] In some embodiments, the step S132 above, which involves determining a target first model from a plurality of first models that satisfies the parameter conditions corresponding to the second parameter dimension, may include the following steps S151 and S152:
[0098] Step S151: Obtain the second data corresponding to each of the first models under the second parameters; the second data characterizes the support capability of the first model for tasks with the second parameters.
[0099] In some implementations, the ability of the first model to support a task with a second parameter, i.e., the second data, is predetermined.
[0100] The second data can be described using corresponding scores. For example, multiple first models can be scored in advance based on their ability to support tasks with second parameters, thus obtaining the second data.
[0101] The second data can be described using the corresponding accuracy rates. For example, the second data can be obtained by pre-processing the accuracy rates of multiple models on a task with a second parameter.
[0102] Step S152: Based on the second data, select a target first model from multiple first models that satisfies the second target data conditions corresponding to the second parameter dimension.
[0103] In some implementations, the model is determined to meet the first target data condition corresponding to the task type dimension based on the second data.
[0104] In some implementations, if the second data is described using a corresponding score, and the score corresponding to the first model processing the task with the second parameter is greater than a second score threshold, then the first model is determined to satisfy the second target data condition corresponding to the second parameter dimension.
[0105] Based on the score corresponding to the processing of tasks with the second parameter by the first model, the multiple first models are sorted, and the first model ranked first among the multiple first models is determined to satisfy the second target data condition corresponding to the second parameter dimension.
[0106] In some implementations, if the accuracy of the first model in processing a task with the second parameter is greater than the second accuracy threshold, and the second data is described using the corresponding accuracy, then the first model is determined to satisfy the second target data condition corresponding to the second parameter dimension.
[0107] Based on the accuracy of the first model in processing tasks with the second parameter, the multiple first models are sorted, and the first model ranked first among the multiple first models is determined to satisfy the second target data condition corresponding to the second parameter dimension.
[0108] Given that the first model satisfies the second target data conditions corresponding to the second parameter dimension, multiple target first models can support the processing of tasks with the second parameter.
[0109] In this embodiment, based on the first model's ability to support tasks with second parameters, a target first model that satisfies the second target data conditions corresponding to the second parameter dimension is selected from multiple first models. This allows consideration from the second parameter dimension, ensuring that the selected target first model supports processing tasks with second parameters.
[0110] In some embodiments, the second parameter includes a first sub-parameter, which represents the language type of the corresponding model; the second target data condition includes the first sub-data condition.
[0111] The step S152 above, which involves selecting a target first model from multiple first models based on the second data that satisfies the second target data condition corresponding to the second parameter dimension, may include the following step S161:
[0112] Step S161: Based on the second data, select a target first model from multiple first models that satisfies the first sub-data conditions corresponding to the language type dimension.
[0113] Here, the language type can include, but is not limited to, at least one of English, Chinese, French, Arabic, etc.
[0114] In some implementations, a target first model matching the language type of the second data representation is selected from multiple first models based on the first model's ability to support the task of the language type represented by the second parameter. For example, if the language type of the second data representation is English, a target first model matching English is selected from multiple first models based on the first model's ability to support English.
[0115] If the first model's ability to support tasks involving the language type represented by the second parameter is greater than a first preset capability threshold, then the first model is determined to satisfy the first sub-data condition corresponding to the language type dimension. For example, if the language type represented by the second data is English, and the first model's ability to support tasks involving English is greater than the first preset capability threshold, then the first model is determined to satisfy the first sub-data condition corresponding to the language type dimension.
[0116] Based on the ability of the first model to support tasks related to the language type represented by the second parameter, multiple first models are ranked, and the first model ranked highest satisfies the first sub-data condition corresponding to the language type dimension. For example, if the language type represented by the second data is English, based on the ability of the first model to support tasks related to English, multiple first models are ranked, and the first model ranked highest satisfies the first sub-data condition corresponding to the language type dimension.
[0117] In this embodiment, the second parameter includes a first sub-parameter, which represents the language type of the corresponding model; the second target data condition includes the first sub-data condition; based on the second data, a target first model that satisfies the first sub-data condition corresponding to the language type dimension is selected from multiple first models. In this way, the language type of the selected target first model can be considered from the perspective of language type, ensuring that the language type of the selected target first model meets the language type requirements of the target input information.
[0118] In some embodiments, the second parameter includes a second sub-parameter, which represents the knowledge domain of the corresponding model; the second target data condition includes a second sub-data condition.
[0119] The step S152 above, which involves selecting a target first model from multiple first models based on the second data that satisfies the second target data condition corresponding to the second parameter dimension, may include the following step S171:
[0120] Step S171: Based on the second data, select a target first model from multiple first models that satisfies the second sub-data conditions corresponding to the knowledge domain dimension.
[0121] Here, the field of knowledge may include, but is not limited to, at least one of literature, medicine, physics, etc.
[0122] In some implementations, a target first model matching the knowledge domain of the second data representation is selected from multiple first models based on the first model's ability to support tasks in the knowledge domain represented by the second parameter. For example, if the knowledge domain of the second data representation is literature, a target first model matching literature is selected from multiple first models based on the first model's ability to support tasks in literature.
[0123] If the first model's ability to support tasks within the knowledge domain represented by the second parameter is greater than a second preset capability threshold, then the first model is determined to satisfy the second sub-data condition corresponding to the knowledge domain dimension. For example, if the knowledge domain represented by the second data is literature, and the first model's ability to support tasks related to literature is greater than the second preset capability threshold, then the first model is determined to satisfy the second sub-data condition corresponding to the knowledge domain dimension.
[0124] Based on the ability of the first model to support tasks within the knowledge domain represented by the second parameter, multiple first models are ranked, and the first model ranked highest satisfies the second sub-data condition corresponding to the knowledge domain dimension. For example, if the knowledge domain represented by the second data is literature, based on the ability of the first model to support tasks related to literature, multiple first models are ranked, and the first model ranked highest satisfies the second sub-data condition corresponding to the knowledge domain dimension.
[0125] In this embodiment, the second parameter includes a second sub-parameter, which represents the knowledge domain of the corresponding model; the second target data condition includes a second sub-data condition; based on the second data, a target first model that satisfies the second sub-data condition corresponding to the knowledge domain dimension is selected from multiple first models. In this way, the knowledge domain of the selected target first model can be considered from the perspective of the knowledge domain, ensuring that the knowledge domain of the selected target first model meets the knowledge domain requirements of the target input information.
[0126] In some embodiments, the second parameter includes a third sub-parameter, which characterizes the model accuracy of the corresponding model; the second target data condition includes a third sub-data condition.
[0127] The step S152 above, which involves selecting a target first model from multiple first models based on the second data that satisfies the second target data condition corresponding to the second parameter dimension, may include the following step S181:
[0128] Step S181: Based on the second data, select a target first model from multiple first models that satisfies the third sub-data conditions corresponding to the model accuracy dimension.
[0129] In some implementations, the second data characterizes the first model's ability to support the task of model accuracy characterized by the second parameter, and a target first model that matches the model accuracy characterized by the second data is selected from a plurality of first models.
[0130] If the first model's ability to support the task represented by the model accuracy of the second parameter is greater than the third preset capability threshold, then the first model is determined to satisfy the third sub-data condition corresponding to the model accuracy dimension.
[0131] Based on the ability of the first model to support the task of model accuracy represented by the second parameter, multiple first models are sorted, and the first model ranked first among the multiple first models is determined to satisfy the third sub-data condition corresponding to the model accuracy dimension.
[0132] In this embodiment, the second parameter includes a third sub-parameter, which characterizes the model accuracy of the corresponding model; the second target data condition includes a third sub-data condition; based on the second data, a target first model that satisfies the third sub-data condition corresponding to the model accuracy dimension is selected from multiple first models. In this way, considering the dimension of model accuracy, the selected target first model's model accuracy meets the model accuracy requirements of the target input information.
[0133] The following describes the application of the embodiments of this application in a real-world scenario.
[0134] In the process of using large models to process input information, there are multiple large models with outstanding capabilities in different dimensions. In related technologies, users can choose a suitable large model to process and output the input information based on the input information, or use the large model's own capabilities to assist in decision-making.
[0135] Based on the above description, embodiments of this application provide a method for quickly determining a large model suitable for input information. This method selects a large model suitable for executing specific input information by quantitatively evaluating multiple parameters of the large model.
[0136] When selecting a suitable large model based on specific input information, decisions can be made based on the following parameter dimensions:
[0137] Task Type: Based on the input information, determine the corresponding task type; that is, the selected large model must support the task type corresponding to the input information. Task types can include text generation, code generation, text-to-image conversion, image classification, etc.
[0138] Language Type: Based on the input information, determine the corresponding language type; that is, the selected large model must support the language type corresponding to the input information. Language types can include English, Chinese, French, Arabic, etc.
[0139] Knowledge Domain: Based on the input information, determine the corresponding knowledge domain; that is, the selected large model must support the knowledge domain corresponding to the input information. Knowledge domains can include literature, medicine, finance, history, chemistry, etc.
[0140] Model Costs: Determine the official pricing for each model and estimate the cost of using each model. For example, the cost of a large model like GPT4 is relatively high, the cost of a large model like Claude 3.5 is moderate, the cost of a large model like GPT4o-mini is relatively low, and the cost of a large model like AWS Mistral is relatively low.
[0141] The method for quickly determining whether a large model is suitable for input information provided in this application embodiment may include the following steps S601 to S603:
[0142] Step S601: Based on the input information, determine the target parameters corresponding to the input information from multiple parameter dimensions; the target parameters include a first parameter and a second parameter, the first parameter representing the task type of the corresponding model; the second parameter may include language type, knowledge domain, model cost, etc.
[0143] By leveraging the text classification capabilities of neural network models, the input information is classified from multiple parameter dimensions to obtain the target parameters.
[0144] When the parameter dimension includes task type, determine the task type corresponding to the input information. For example, the task type represented by the first parameter can be text generation.
[0145] When the parameter dimension includes language type, the language type corresponding to the input information is determined. For example, the language type represented by the second parameter can be English.
[0146] When the parameter dimension includes a knowledge domain, the knowledge domain corresponding to the input information is determined. For example, the knowledge domain represented by the second parameter can be literature.
[0147] Determine the model cost of multiple large models, taking into account the model cost as a parameter dimension.
[0148] Step S602: Obtain the scores of multiple large models under the target parameters;
[0149] Obtain the score of the large model under the first parameter.
[0150] Obtain the score of the large model under the second parameter. If the second parameter includes language type, obtain the score of the large model under language type. If the second parameter includes knowledge domain, obtain the score of the large model under knowledge domain.
[0151] Step S603: Determine the target model (corresponding to the first target model in the aforementioned embodiment) based on the scores of multiple large models under the target parameters.
[0152] Under the parameter dimension corresponding to the target parameter, weights are assigned to the target parameter according to its importance. Based on the weights of the target parameter and the score of the large model under the target parameter, the score of the large model is determined, and the target model is selected based on the score.
[0153] For a large model m and input information w, the score Score(m,w) of the large model can be calculated using formula (1):
[0154] In formula (1), S(w,d) iα represents the score of the large model under the target parameters corresponding to the input information w; i These are the weights corresponding to the target parameters.
[0155] Figure 4 is a schematic diagram of a method for selecting a large model provided in an embodiment of this application. As shown in Figure 4, taking the parameter dimension as the task type dimension as an example, firstly, based on the multiple language descriptions in the input information, each piece of input information is matched to the corresponding task type. Secondly, based on the support capabilities of multiple large models for the task type corresponding to the input information, a suitable large model is selected. For example, if the language description of the input information is "Create an avatar for my company's social media account. Ensure that its background is professional, the color scheme is consistent, and it matches my company's brand image," the language description is analyzed, and the corresponding task type is matched as text and image conversion. Based on the support capabilities of multiple large models for the text and image conversion task, the first models that support the text and image conversion task among the multiple large models are determined to be the large model Stable Diffusion and the large model CogVideoX.
[0156] Based on the task types supported by the large model under the task type dimension and at least one second parameter, select the target first model from multiple first models.
[0157] When the second parameter includes a knowledge domain, each piece of input information is matched to its corresponding knowledge domain based on multiple linguistic descriptions within the input information. Then, based on the support capabilities of multiple large models for the knowledge domain corresponding to the input information, a suitable target first model is selected from the large models Stable Diffusion and CogVideoX. For example, if the linguistic description of the input information is "Create an avatar for my company's social media accounts. Ensure its background is professional, the color scheme is consistent, and it matches my company's brand image," analyzing the linguistic description reveals that the corresponding knowledge domain is the company operations domain. Based on the support capabilities of the large models Stable Diffusion and CogVideoX for the company operations domain, it is determined that the large model Stable Diffusion performs better in the company operations domain task; therefore, the large model Stable Diffusion is selected to process the input information.
[0158] In this embodiment, a neural network model is used to classify the input information from multiple parameter dimensions to obtain target parameters. Then, a target model is selected based on the scores of multiple large models corresponding to the target parameters. This allows for the automatic selection of a suitable large model for different input information without relying on the large model itself, reducing computational resource consumption by not needing to call the large model during selection.
[0159] Based on the foregoing embodiments, this application provides a data processing device, which includes various modules and units included in each module, and can be implemented by a processor in a computer device; of course, it can also be implemented by specific logic circuits; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.
[0160] Figure 5 is a schematic diagram of the composition structure of a data processing device provided in an embodiment of this application. As shown in Figure 5, the data processing device 500 includes: an acquisition module 501, a first determination module 502, and a second determination module 503, wherein:
[0161] The module 501 is used to obtain target input information; the target input information is input information that meets the input conditions.
[0162] The first determining module 502 is used to determine target parameters based on the target input information; the target parameters include a first parameter and at least one second parameter, wherein the first parameter characterizes the task type of the corresponding model, and the second parameter is different from the first parameter;
[0163] The second determining module 503 is used to determine a target first model that meets the parameter conditions from multiple first models based on the target parameters. The target first model is used to output based on the target input information. The multiple first models are large models, and the first parameters of the multiple first models are the same, while at least one second parameter is different.
[0164] In some embodiments, the first determining module includes a first determining unit, configured to determine target parameters based on the target input information using a second model; the magnitude of the processing parameters of the second model is smaller than the magnitude of the processing parameters of the first model.
[0165] In some embodiments, the first determining unit includes a first classification subunit, configured to classify the target input information in the task type dimension using the second model to determine the first parameter; and a second classification subunit, configured to classify the target input information in the parameter dimension corresponding to the second parameter using the second model to determine the second parameter.
[0166] In some embodiments, the second determining model includes a second determining unit, configured to determine a plurality of first models that satisfy the parameter conditions corresponding to the task type dimension based on the first parameters; and a third determining unit, configured to determine a target first model that satisfies the parameter conditions corresponding to the second parameter dimension from the plurality of first models based on the second parameters.
[0167] In some embodiments, the second determining unit includes a first acquiring subunit, configured to acquire first data corresponding to multiple models under the first parameter; the first data characterizes the model's ability to support the task type corresponding to the first parameter; and a first selecting subunit, configured to select multiple first models from each of the models based on the first data, which satisfy the first target data condition corresponding to the task type dimension.
[0168] In some embodiments, the third determining unit includes a second obtaining subunit, configured to obtain second data corresponding to each of the first models under the second parameter; the second data characterizes the support capability of the first model for a task having the second parameter; and a second selecting subunit, configured to select a target first model from a plurality of first models based on the second data, which satisfies the second target data condition corresponding to the dimension of the second parameter.
[0169] In some embodiments, the second parameter includes a first sub-parameter, which represents the language type of the corresponding model; the second target data condition includes a first sub-data condition; the second selection sub-unit is further configured to select a target first model that satisfies the first sub-data condition corresponding to the language type dimension from a plurality of first models based on the second data.
[0170] In some embodiments, the second parameter includes a second sub-parameter, which represents the knowledge domain of the corresponding model; the second target data condition includes a second sub-data condition; the second selection sub-unit is further configured to select a target first model that satisfies the second sub-data condition corresponding to the knowledge domain dimension from a plurality of first models based on the second data.
[0171] In some embodiments, the second parameter includes a third sub-parameter, which characterizes the model accuracy of the corresponding model; the second target data condition includes a third sub-data condition; the second selection sub-unit is further configured to select a target first model that satisfies the third sub-data condition corresponding to the model accuracy dimension from a plurality of first models based on the second data.
[0172] The descriptions of the apparatus embodiments above are similar to those of the method embodiments above, and have similar beneficial effects. In some embodiments, the functions or modules included in the apparatus provided in this application can be used to perform the methods described in the method embodiments above. For technical details not disclosed in the apparatus embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.
[0173] It should be noted that, in the embodiments of this application, if the above methods are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of software products. These software products are stored in a storage medium and include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware, software, or firmware, or any combination of hardware, software, and firmware.
[0174] This application provides a computer device including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements some or all of the steps in the above-described method.
[0175] This application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements some or all of the steps in the above-described method. The computer-readable storage medium can be transient or non-transient.
[0176] This application provides a computer program including computer-readable code. When the computer-readable code is run in a computer device, the processor in the computer device performs some or all of the steps in the above-described method.
[0177] This application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. When the computer program is read and executed by a computer, it implements some or all of the steps in the above-described method. This computer program product can be implemented specifically through hardware, software, or a combination thereof. In some embodiments, the computer program product is specifically embodied as a computer storage medium; in other embodiments, the computer program product is specifically embodied as a software product, such as a software development kit (SDK), etc.
[0178] It should be noted that the descriptions of the various embodiments above tend to emphasize the differences between them, while their similarities or commonalities can be referred to interchangeably. The descriptions of the above embodiments of the device, storage medium, computer program, and computer program product are similar to the descriptions of the above method embodiments and have similar beneficial effects. For technical details not disclosed in the embodiments of the device, storage medium, computer program, and computer program product of this application, please refer to the descriptions of the method embodiments of this application for understanding.
[0179] It should be noted that Figure 6 is a schematic diagram of a hardware entity of a computer device in an embodiment of this application. As shown in Figure 6, the hardware entity of the computer device 800 includes: a processor 801, a communication interface 802, and a memory 803, wherein:
[0180] Processor 801 typically controls the overall operation of computer device 800.
[0181] The communication interface 802 enables computer devices to communicate with other terminals or servers over a network.
[0182] The memory 803 is configured to store instructions and applications executable by the processor 801, and can also cache data to be processed or already processed (e.g., image data, audio data, voice communication data, and video communication data) in the processor 801 and various modules in the computer device 800. It can be implemented using flash memory or random access memory (RAM). Data transfer between the processor 801, the communication interface 802, and the memory 803 can be performed via bus 804.
[0183] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above steps / processes do not imply a sequential order of execution; the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above embodiments of this application are merely descriptive and do not represent the superiority or inferiority of the embodiments.
[0184] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0185] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.
[0186] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, the functional units in the embodiments of this application may all be integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the integrated unit may be implemented in hardware or in a combination of hardware and software functional units.
[0187] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.
[0188] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence or the part that contributes to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.
[0189] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.
Claims
1. A data processing method, applied to a target application, comprising: Obtain the target input information; Determine the target parameters based on the target input information; The target parameters include a first parameter and at least one second parameter, wherein the first parameter characterizes the task type of the corresponding model, and the second parameter is different from the first parameter; Based on the target parameters, a target first model that satisfies the parameter conditions is determined from a plurality of first models, and the target first model is used to output based on the target input information; Multiple first models are large models, and multiple first models have the same first parameter, but at least one second parameter is different.
2. The method according to claim 1, wherein determining the target parameters based on the target input information comprises: Using the second model, the target parameters are determined based on the target input information; The magnitude of the processing parameters of the second model is smaller than that of the processing parameters of the first model.
3. The method according to claim 2, wherein determining the target parameters based on the target input information using the second model includes: Using the second model, the target input information is classified in the dimension of task type to determine the first parameter; Using the second model, the target input information is classified under the parameter dimension corresponding to the second parameter to determine the second parameter.
4. The method according to any one of claims 1 to 3, wherein determining the target first model satisfying the parameter conditions from a plurality of first models based on the target parameters comprises: Based on the first parameter, determine multiple first models that satisfy the parameter conditions corresponding to the task type dimension; Based on the second parameter, a target first model that satisfies the parameter conditions corresponding to the dimension of the second parameter is determined from multiple first models.
5. The method according to claim 4, wherein determining a plurality of first models that satisfy the parameter conditions corresponding to the task type dimension based on the first parameter includes: Obtain the first data corresponding to multiple models under the first parameter; The first data characterizes the model's ability to support the task type corresponding to the first parameter; Based on the first data, select multiple first models from each of the models that satisfy the first target data conditions corresponding to the task type dimension.
6. The method according to claim 4, wherein determining the target first model that satisfies the parameter conditions corresponding to the second parameter dimension from a plurality of first models based on the second parameter comprises: Obtain the second data corresponding to each of the first models under the second parameters; The second data characterizes the ability of the first model to support tasks with the second parameters; Based on the second data, a target first model that satisfies the second target data condition corresponding to the second parameter dimension is selected from multiple first models.
7. The method according to claim 6, wherein the second parameter includes a first sub-parameter, the first sub-parameter representing the language type of the corresponding model; and the second target data condition includes the first sub-data condition; The step of selecting a target first model from multiple first models based on the second data, which satisfies the second target data condition corresponding to the second parameter dimension, includes: Based on the second data, a target first model that meets the first sub-data conditions corresponding to the language type dimension is selected from multiple first models.
8. The method according to claim 6, wherein the second parameter includes a second sub-parameter, the second sub-parameter representing the knowledge domain of the corresponding model; and the second target data condition includes a second sub-data condition. The step of selecting a target first model from multiple first models based on the second data, which satisfies the second target data condition corresponding to the second parameter dimension, includes: Based on the second data, a target first model that satisfies the second sub-data conditions corresponding to the knowledge domain dimension is selected from multiple first models.
9. The method according to claim 6, wherein the second parameter includes a third sub-parameter, the third sub-parameter representing the model accuracy of the corresponding model; and the second target data condition includes a third sub-data condition; The step of selecting a target first model from multiple first models based on the second data, which satisfies the second target data condition corresponding to the second parameter dimension, includes: Based on the second data, a target first model that satisfies the third sub-data condition corresponding to the model accuracy dimension is selected from multiple first models.
10. A data processing apparatus, the apparatus comprising: The acquisition module is used to obtain target input information; The first determining module is used to determine target parameters based on the target input information; The target parameters include a first parameter and at least one second parameter, wherein the first parameter characterizes the task type of the corresponding model, and the second parameter is different from the first parameter; The second determining module is used to determine a target first model that satisfies the parameter conditions from a plurality of first models based on the target parameters, and the target first model is used to output based on the target input information; Multiple first models are large models, and multiple first models have the same first parameter, but at least one second parameter is different.