Model optimization method and apparatus, device, and storage medium

Abstract

Embodiments of the present disclosure provide a model optimization method and apparatus, a device, and a storage medium. The method comprises: generating configuration information on the basis of scenario description information associated with a first scenario; on the basis of the configuration information, generating a set of interface interaction instructions associated with a development tool, so as to trigger the generation, in the development tool, of a model interaction request corresponding to the first scenario; determining a plurality of candidate response contents generated by a plurality of models on the basis of the model interaction request; determining a target response content from among the plurality of candidate response contents on the basis of first evaluation information of the plurality of candidate response contents; and by means of the target response content, optimizing at least one model associated with the development tool. Therefore, the embodiments of the present disclosure can improve the processing quality of the model in different scenarios.

Description

Methods, apparatus, equipment and storage media for model optimization Technical Field

[0001] The exemplary embodiments disclosed herein generally relate to the field of computers, and particularly to methods, apparatus, devices, and computer-readable storage media for model optimization. Background Technology

[0002] With the development of computer technology, generative models are increasingly being applied in various scenarios. For example, some generative models can provide integrated development environments (IDEs) with capabilities such as automatic code editing, code generation, and code repair. The goal is to improve the code editing efficiency of such generative models. Summary of the Invention

[0003] In a first aspect of this disclosure, a method for model optimization is provided. The method includes: generating configuration information based on scene description information associated with a first scene; generating a set of interface interaction instructions associated with a development tool based on the configuration information to trigger the generation of a model interaction request corresponding to the first scene in the development tool; determining multiple candidate response contents generated by multiple models based on the model interaction request; determining a target response content from the multiple candidate response contents based on first evaluation information of the multiple candidate response contents; and optimizing at least one model associated with the development tool using the target response content.

[0004] In a second aspect of this disclosure, an apparatus for model optimization is provided. The apparatus includes: a configuration generation module configured to generate configuration information based on scene description information associated with a first scene; an instruction generation module configured to generate a set of interface interaction instructions associated with a development tool based on the configuration information, to trigger the generation of a model interaction request corresponding to the first scene in the development tool; a response acquisition module configured to determine multiple candidate response contents generated by multiple models based on the model interaction request; a response determination module configured to determine a target response content from the multiple candidate response contents based on the first evaluation information of the multiple candidate response contents; and a model optimization module configured to optimize the multiple models using the target response content.

[0005] In a third aspect of this disclosure, an electronic device is provided. The device includes at least one processing unit; and at least one memory coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit. When executed by the at least one processing unit, the instructions cause the device to perform the method of the first aspect.

[0006] In a fourth aspect of this disclosure, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program that can be executed by a processor to implement the method of the first aspect.

[0007] In a fifth aspect of this disclosure, a computer program product is provided. The computer program product includes computer-executable instructions that, when executed by a processor, implement the method according to a first aspect of this disclosure.

[0008] It should be understood that the content described in this content section is not intended to limit the key or essential features of the embodiments of this disclosure, nor is it intended to restrict the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0009] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0010] Figure 1 shows a schematic diagram of an example environment in which embodiments of the present disclosure can be implemented;

[0011] Figure 2 illustrates an example system according to some embodiments of the present disclosure;

[0012] Figure 3 illustrates a flowchart of an example process for model optimization according to some embodiments of the present disclosure;

[0013] Figure 4 shows a schematic structural block diagram of an apparatus for model optimization according to certain embodiments of the present disclosure;

[0014] Figure 5 shows a block diagram of an electronic device capable of implementing several embodiments of the present disclosure. Detailed Implementation

[0015] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0016] It should be noted that the headings of any section / subsection provided herein are not limiting. Various embodiments are described throughout this document, and embodiments of any type may be included under any section / subsection. Furthermore, embodiments described in any section / subsection may be combined in any way with any other embodiments described in the same section / subsection and / or different sections / subsections.

[0017] In the description of embodiments of this disclosure, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The term "some embodiments" should be understood as "at least some embodiments". Other explicit and implicit definitions may also be included below. The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0018] The embodiments of this disclosure may involve user data, data acquisition, and / or use. All of these aspects comply with applicable laws, regulations, and relevant provisions. In the embodiments of this disclosure, all data collection, acquisition, processing, manipulation, forwarding, and use are conducted with the user's knowledge and confirmation. Accordingly, in implementing the embodiments of this disclosure, the type, scope of use, and usage scenarios of any data or information that may be involved should be communicated to the user and their authorization obtained in accordance with relevant laws and regulations through appropriate means. The specific methods of notification and / or authorization may vary depending on the actual situation and application scenario, and the scope of this disclosure is not limited in this respect.

[0019] In this specification and the embodiments, any processing of personal information will be carried out only under the premise of legality (such as obtaining the consent of the personal information subject, or being necessary for the performance of a contract), and will only be carried out within the scope stipulated or agreed upon. A user's refusal to process personal information other than that necessary for basic functions will not affect the user's use of basic functions.

[0020] As mentioned above, some generative models are gradually being applied to integrated development environments (IDEs). However, obtaining high-quality training data is difficult for IDEs, and manual annotation is costly and time-consuming. Furthermore, automated methods often fail to generate diverse data that matches the distribution of real-world data, resulting in trained models that may not accurately reflect real-world programming scenarios.

[0021] The embodiments of this disclosure propose a model optimization scheme. According to this scheme, configuration information can be generated based on scene description information associated with a first scene. Furthermore, based on the configuration information, a set of interface interaction instructions associated with a development tool can be generated to trigger the generation of a model interaction request corresponding to the first scene in the development tool.

[0022] Furthermore, multiple candidate response contents generated by multiple models based on model interaction requests can be identified, and a target response content can be determined from these candidate response contents based on first evaluation information. Accordingly, at least one model associated with the development tool can be optimized using the target response content.

[0023] Based on this approach, embodiments of this disclosure construct realistic interaction requests in different scenarios, thereby optimizing the processing quality of the model in different scenarios.

[0024] Example Environment

[0025] Figure 1 illustrates a schematic diagram of an example environment 100 in which embodiments of the present disclosure can be implemented. As shown in Figure 1, the example environment 100 may include an electronic device 110.

[0026] In this example environment 100, electronic device 110 may be, for example, a server, which may support the operation of development tool 120. Development tool 120, also known as an integrated development environment (IDE), provides developers with various tools and functions needed to write, debug, and run code.

[0027] In some scenarios, development tool 120 can be deployed on suitable terminal devices. Examples of terminal devices may include, but are not limited to, any type of mobile terminal, fixed terminal, or portable terminal, including mobile phones, desktop computers, laptop computers, notebook computers, netbook computers, tablet computers, media computers, multimedia tablets, handheld computers, portable gaming terminals, VR / AR devices, personal communication system (PCS) devices, personal navigation devices, personal digital assistants (PDAs), audio / video players, digital cameras / camcorders, positioning devices, television receivers, radio receivers, e-book devices, gaming devices, or any combination of the foregoing, including accessories and peripherals of these devices or any combination thereof.

[0028] Additionally, as shown in Figure 1, development tool 120 can also utilize model 150 to handle user requests within the application. As an example, development tool 120 can support users initiating requests such as code generation, code explanation, and code modification via dialogue.

[0029] As shown in the figure, the development tool 120 can receive natural language text 130 input by the user and can use the model 150 to generate response content 140 for the natural language text 130. For example, the development tool 120 can present the code content generated by the model 150 based on the natural language text 130.

[0030] Electronic device 110 can be, for example, a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and big data and artificial intelligence platforms. Electronic device 110 can also include, for example, computing systems / servers, such as mainframes, edge computing nodes, computing devices in cloud environments, and so on.

[0031] It should be understood that the structure and function of the various elements in environment 100 are described for illustrative purposes only and do not imply any limitation on the scope of this disclosure.

[0032] The following description will continue with reference to the accompanying drawings, which will provide some exemplary embodiments of this disclosure.

[0033] Example System

[0034] Figure 2 illustrates a system 200 for model optimization according to some embodiments of the present disclosure. The system 200 can, for example, be used to optimize a model 150 associated with development tool 120 to improve the model 150's ability to handle user requests in different scenarios.

[0035] As shown in Figure 2, the system 200 may include units for performing different tasks, such as a plan generation unit 210, a prompt word generation unit 220, a response generation unit 250, and a model optimization unit 260.

[0036] Overall, the plan generation unit 210 can be configured to generate an optimization plan based on data from the development tool 120. The prompt generation unit 220 may include a configuration generation unit 230 for generating specific dialogue configuration information. The prompt generation unit 220 may also include an automated dialogue unit 240 for constructing a dialogue request based on the dialogue configuration information. The response generation unit 250 can be configured to use a model to generate a response to the dialogue request and select response content for optimizing the model. The model optimization unit 260 can fine-tune the model using the response content.

[0037] The following section will detail the specific implementation of each unit.

[0038] Plan generation unit

[0039] In some embodiments, the plan generation unit 210 can be used to generate a data production plan for model optimization. Specifically, the plan generation unit 210 can acquire scene statistics associated with the development tool 120 (e.g., IDE).

[0040] As shown in Figure 2, the planning generation unit 210 can, for example, access the original online dataset 212 corresponding to the IDE and perform question-and-answer behavior analysis 214 to generate scene statistics associated with the development tool 120. As an example, the original online dataset 212 can store historical question-and-answer data associated with the development tool 120.

[0041] In some embodiments, the planning generation unit 210 can perform multi-dimensional statistical analysis on the historical question-and-answer data stored in the original online dataset 212, thereby generating classification information associated with multiple scene attributes. As an example, multiple scene attributes are used to describe different aspects of a scene.

[0042] Table 1 shows some example scenario attributes and corresponding classifications of embodiments according to this disclosure.

[0043] Table 1 Example Scene Attributes

[0044] Table 1 serves as an example. Scene attributes can include, for example, cursor behavior to indicate the type of cursor operation. For instance, the cursor operation type could indicate whether a file is active, or whether a code block or line of code is selected.

[0045] Scene attributes may also include, for example, a trigger mode to indicate the interaction mode that triggers the request. As an example, a trigger mode could include triggering a dialogue within a line of code or triggering a dialogue from a dialogue view.

[0046] Scene attributes may also include, for example, the instruction type, indicating the type of instruction that triggered the request. For instance, the instruction type may include a query statement, a quick dialogue template, or a combination of a query statement and a dialogue template.

[0047] Scene attributes may also include, for example, the programming language, to indicate the programming language corresponding to the currently developed code.

[0048] Scene attributes may also include, for example, the dialogue rounds to indicate the dialogue rounds that trigger the request, such as whether it is the first round of dialogue or multiple rounds of dialogue.

[0049] Context attributes may also include referencing information, indicating the information referenced in generating the response. For example, referencing information could indicate what the response to the request references, such as historical dialogues, selected codes, contextual information, questions, error messages, etc.

[0050] Scene attributes may also include scene difficulty, to indicate the difficulty of processing the request, such as high, medium, low, etc.

[0051] Scene attributes may also include, for example, the request type, indicating the category of the requested task. As an example, task categories could include code generation, code editing, code explanation, comment generation, code fixing, code Q&A, etc.

[0052] In some embodiments, the first statistical information corresponding to the first type of scene attribute in the above scene attributes can be determined based on the historical interaction operation rules indicated by the original online dataset 212. As an example, the plan generation unit 210 can use rules to determine the classification statistics of the following attributes: cursor behavior, triggering mode, instruction type, programming language, dialogue turn, etc.

[0053] In some embodiments, second statistical information corresponding to second-type scenario attributes, such as reference information, scenario difficulty, and request type, can also be generated using a model. As an example, the plan generation unit 210 can provide descriptive information associated with historical interactions to a language model to generate the second statistical information. For instance, the language model can generate classifications associated with second-type scenario attributes such as reference information, scenario difficulty, and request type based on the descriptive information.

[0054] Furthermore, the planning generation unit 210 can also utilize the model to process scene statistics to generate a data production plan 216. Specifically, the planning generation unit 210 can, for example, provide scene statistics to the language model. As an example, the scene statistics can describe the classification distribution of the above scene attributes by the development tool 120.

[0055] Furthermore, the planning generation unit 210 can also instruct the language model to generate a data production plan based on scene statistics. The data production plan may include a specified number of scene descriptions associated with the scene.

[0056] As an example, the plan generation unit 210 can indicate the number of data items to be produced in each plan, and the language model can generate corresponding scenario description information for each data item. Such scenario description information can indicate the specific classification of the above scenario attributes.

[0057] Prompt word generation unit

[0058] In some embodiments, the prompt word generation unit 220 can be configured to generate model interaction requests in the development tool 120, thereby constructing corresponding prompt words.

[0059] As shown in Figure 2, the configuration generation unit 230 may include a configuration generator 232, which can generate configuration information 234 based on the scene description information generated by the planning generation unit 210.

[0060] Specifically, the configuration generator 232 can construct prompts based on scene description information and preset prompt templates. As an example, the configuration generator 232 can fill the prompt template with at least some of the scene attributes indicated by the scene description information. For example, the prompt template can be filled with attributes such as programming language, request type, instruction type, reference information, and scene difficulty.

[0061] Accordingly, the configuration generator 232 can provide prompt words to the language model to obtain the question content generated by the language model, also known as the query item.

[0062] As an example, the request type could be code editing, the scenario difficulty could be high, and the reference information could include the selected code. Correspondingly, the question generated by the language model could be "Please refactor the selected code to optimize performance."

[0063] In some embodiments, the dialogue turn indicated by the scene description information may include both the first turn and subsequent turns. If the scene description information indicates a target turn (e.g., greater than 1), the question content generated by the model includes both the question content of the current turn and historical question content corresponding to at least one previous turn. In this way, embodiments of this disclosure can support the simulation of multi-turn dialogues, thereby improving the model's processing capabilities.

[0064] In some embodiments, the configuration generation unit 230 may also determine other configuration items in the configuration information 234 besides the "problem content" based on other scene attributes indicated in the scene description information. As an example, such configuration items may include, but are not limited to: task identifier, task dialogue round, code repository address (URL), programming language, open file, selected code, cursor position, current dialogue round, etc.

[0065] Further, as shown in Figure 2, the automated dialogue unit 240 can filter the generated configuration information 234 using the query filter 242. Specifically, the automated dialogue unit 240 can provide the model with the question content (i.e., the query item) indicated by the configuration information 234 and the associated code snippet to generate evaluation information for the question content. If the evaluation information indicates that the quality of the question content does not meet the conditions, the automated dialogue unit 240 may choose not to use the configuration information.

[0066] Conversely, if the evaluation information meets preset conditions, the automated dialogue unit 240 can generate a set of interface interaction instructions associated with the development tool based on the configuration information. Furthermore, the automated dialogue unit 240 can control the generation of model interaction requests corresponding to the first scenario within the development tool based on the interface interaction script.

[0067] Specifically, the automated dialogue unit 240 can generate a user interface interaction script based on configuration information. The user interface interaction script can include a set of user interface interaction instructions to trigger the user interface interaction process. As an example, the automated dialogue unit 240 can generate a script file that the automated UI tool 244 can execute based on the configuration information.

[0068] As an example, the interface interaction script can instruct the user to first select the line of code specified in the configuration file within the interface, and control the cursor to move to the location indicated in the configuration file. Subsequently, the interface interaction script can instruct the user to open a dialog interaction page, where they can enter the question content specified in the configuration information, thereby triggering a model interaction request.

[0069] Therefore, the automatic dialogue unit 240 can generate multi-turn dialogue prompts (e.g., question content) and corresponding context information corresponding to the model interaction request, thereby simulating real interaction requests in the development tool.

[0070] Response generation unit

[0071] In some embodiments, the response generation unit 250 may be configured to acquire the target response content for the constructed model interaction request.

[0072] As shown in the figure, the response generator 252 can send the constructed model interaction requests to multiple models associated with the development tool, and can obtain multiple candidate response contents generated by multiple models.

[0073] Furthermore, the response generation unit 250 can also rank and filter multiple candidate response contents 254. Specifically, the response generation unit 250 can determine evaluation information for multiple candidate response contents. In some embodiments, this evaluation information can be determined based on rules. Alternatively, such evaluation information can also be generated using a model.

[0074] As an example, the response generation unit 250 can provide multiple candidate response contents to the language model to generate content evaluations (also known as third evaluation information) of the multiple candidate response contents, and can determine the evaluation information of the multiple candidate response contents based on the content evaluations generated by the model.

[0075] For example, the response generation unit 250 can instruct the language model to generate scores for multiple candidate contents as individual evaluations of each candidate content. Furthermore, the response generation unit 250 can also instruct the language model to generate comparative evaluations of the multiple candidate contents.

[0076] In addition to model-generated ratings, response generation unit 250 may also consider rule-based scoring mechanisms to generate rule-based content ratings. As an example, response generation unit 250 may combine model-determined ratings and rule-based content ratings to determine a final rating for multiple candidate content items.

[0077] Furthermore, the response generation unit 250 may, for example, determine the target response content corresponding to the optimal evaluation based on the evaluation information of multiple candidate response contents, thereby constructing multi-turn dialogue SFT (Supervised Fine-Tuning) data 256.

[0078] Model optimization unit

[0079] In some embodiments, the model optimization unit 260 can fine-tune at least one model associated with the development tool based on the constructed multi-turn dialogue SFT data, thereby completing the model optimization.

[0080] As shown in Figure 2, the model optimization unit 260 can perform question-and-answer prompt engineering transformation 262 based on multi-turn dialogue SFT data, and can use the model fine-tuning module 264 to fine-tune, for example, multiple models corresponding to the response generator 252. Additionally or alternatively, the model fine-tuning module 264 can also fine-tune the model actually invoked by the development tool 120.

[0081] Therefore, the embodiments of this disclosure can optimize the model's processing performance in different scenarios. Furthermore, the embodiments of this disclosure can also improve the model's performance when handling complex tasks requiring dialogue history, particularly in intents such as code editing and code generation, where multi-turn dialogue data can significantly improve the model's response accuracy.

[0082] Example process

[0083] Figure 3 shows a flowchart of a model optimization process 300 according to some embodiments of the present disclosure. Process 300 can be implemented in system 200 as shown in Figure 2. Process 300 is described below with reference to Figures 1 and 2.

[0084] In box 310, system 200 generates configuration information based on scene description information associated with the first scene.

[0085] In box 320, system 200 generates a set of interface interaction instructions associated with the development tool based on configuration information, so as to trigger the generation of a model interaction request corresponding to the first scene in the development tool.

[0086] In box 330, system 200 determines multiple candidate response contents generated by multiple models based on model interaction requests.

[0087] In box 340, system 200 determines the target response content from multiple candidate response contents based on the first evaluation information of multiple candidate response contents.

[0088] In box 350, system 200 uses the target response content to optimize at least one model associated with the development tool.

[0089] In some embodiments, the scene description information indicates classification information associated with multiple scene attributes, which are used to describe different aspects of the first scene.

[0090] In some embodiments, multiple scene attributes include at least one of the following: cursor behavior, indicating the type of cursor operation; trigger mode, indicating the interaction mode used to trigger the request; instruction type, indicating the type of instruction used to trigger the request; dialogue turn, indicating the dialogue turn when the request is triggered; reference information, indicating the information referenced for generating the response; scene difficulty, indicating the processing difficulty of the request; and request type, indicating the task category of the request.

[0091] In some embodiments, process 300 further includes: acquiring scene statistics associated with the development tool; and sending the scene statistics to the first model to generate scene description information.

[0092] In some embodiments, scene statistics include: first statistics associated with a first scene attribute, the first statistics being determined based on historical interaction operations; and / or second statistics associated with a second scene attribute, the second statistics being determined by a second model based on descriptive information associated with historical interaction operations.

[0093] In some embodiments, generating configuration information based on scene description information associated with the first scene includes: generating prompt words based on scene description information and a preset prompt word template; providing prompt words to a third model to generate question content; and generating configuration information based on the question content.

[0094] In some embodiments, providing prompt words to a third model to generate question content includes: in response to a dialogue indicating a target round in the scene description information, obtaining question content generated by the third model, wherein the question content also indicates historical question content corresponding to at least one previous round of the target round.

[0095] In some embodiments, the model interaction request indicates the question content and the contextual information associated with the question content.

[0096] In some embodiments, generating a set of interface interaction instructions associated with the development tool based on configuration information includes: providing the fourth model with the problem content indicated by the configuration information and the associated code snippet to generate second evaluation information of the problem content; and in response to the second evaluation information meeting preset conditions, generating a set of interface interaction instructions associated with the development tool based on the configuration information.

[0097] In some embodiments, generating a set of interface interaction instructions associated with the development tool based on configuration information includes: generating an interface interaction script based on configuration information, the interface interaction script including a set of interface interaction instructions for triggering the interface interaction process; and controlling the generation of a model interaction request corresponding to the first scene in the development tool based on the interface interaction script.

[0098] In some embodiments, process 300 further includes: providing a plurality of candidate response contents to a fifth model to generate third evaluation information of the plurality of candidate response contents; and determining first evaluation information of the plurality of candidate response contents based on the third evaluation information.

[0099] In some embodiments, the third evaluation information includes a comparative evaluation of multiple candidate response contents and / or an individual evaluation of candidate response contents.

[0100] Example devices and equipment

[0101] Embodiments of this disclosure also provide corresponding apparatus for implementing the methods or processes described above. Figure 4 shows a schematic structural block diagram of an apparatus 400 for model optimization according to certain embodiments of this disclosure. The apparatus 400 may be implemented as or included in the system 200 discussed above. The various modules / components in the apparatus 400 may be implemented by hardware, software, firmware, or any combination thereof.

[0102] As shown in Figure 4, the device 400 includes a configuration generation module 410 configured to generate configuration information based on scene description information associated with a first scene; an instruction generation module 420 configured to generate a set of interface interaction instructions associated with a development tool based on the configuration information, so as to trigger the generation of a model interaction request corresponding to the first scene in the development tool; a response acquisition module 430 configured to determine multiple candidate response contents generated by multiple models based on the model interaction request; a response determination module 440 configured to determine a target response content from the multiple candidate response contents based on first evaluation information of the multiple candidate response contents; and a model optimization module 450 configured to optimize the multiple models using the target response content.

[0103] In some embodiments, the scene description information indicates classification information associated with multiple scene attributes, which are used to describe different aspects of the first scene.

[0104] In some embodiments, multiple scene attributes include at least one of the following: cursor behavior, indicating the type of cursor operation; trigger mode, indicating the interaction mode used to trigger the request; instruction type, indicating the type of instruction used to trigger the request; dialogue turn, indicating the dialogue turn when the request is triggered; reference information, indicating the information referenced for generating the response; scene difficulty, indicating the processing difficulty of the request; and request type, indicating the task category of the request.

[0105] In some embodiments, the apparatus 400 further includes a scene generation module configured to: acquire scene statistics associated with the development tool; and send the scene statistics to a first model to generate scene description information.

[0106] In some embodiments, scene statistics include: first statistics associated with a first scene attribute, the first statistics being determined based on historical interaction operations; and / or second statistics associated with a second scene attribute, the second statistics being determined by a second model based on descriptive information associated with historical interaction operations.

[0107] In some embodiments, generating configuration information based on scene description information associated with the first scene includes: generating prompt words based on scene description information and a preset prompt word template; providing prompt words to a third model to generate question content; and generating configuration information based on the question content.

[0108] In some embodiments, providing prompt words to a third model to generate question content includes: in response to a dialogue indicating a target round in the scene description information, obtaining question content generated by the third model, wherein the question content also indicates historical question content corresponding to at least one previous round of the target round.

[0109] In some embodiments, the model interaction request indicates the question content and the contextual information associated with the question content.

[0110] In some embodiments, generating a set of interface interaction instructions associated with the development tool based on configuration information includes: providing the fourth model with the problem content indicated by the configuration information and the associated code snippet to generate second evaluation information of the problem content; and in response to the second evaluation information meeting preset conditions, generating a set of interface interaction instructions associated with the development tool based on the configuration information.

[0111] In some embodiments, generating a set of interface interaction instructions associated with the development tool based on configuration information includes: generating an interface interaction script based on configuration information, the interface interaction script including a set of interface interaction instructions for triggering the interface interaction process; and controlling the generation of a model interaction request corresponding to the first scene in the development tool based on the interface interaction script.

[0112] In some embodiments, the apparatus 400 further includes an evaluation determination module configured to: provide multiple candidate response contents to a fifth model to generate third evaluation information of the multiple candidate response contents; and determine first evaluation information of the multiple candidate response contents based on the third evaluation information.

[0113] In some embodiments, the third evaluation information includes a comparative evaluation of multiple candidate response contents and / or an individual evaluation of candidate response contents.

[0114] The units included in device 400 can be implemented in various ways, including software, hardware, firmware, or any combination thereof. In some embodiments, one or more units may be implemented using software and / or firmware, such as machine-executable instructions stored on a storage medium. In addition to or as an alternative to machine-executable instructions, some or all of the units in device 400 may be implemented at least partially by one or more hardware logic components. By way of example and not limitation, exemplary types of hardware logic components that may be used include field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-chips (SoCs), complex programmable logic devices (CPLDs), and so on.

[0115] Figure 5 illustrates a block diagram of an electronic device 500 in which one or more embodiments of the present disclosure may be implemented. It should be understood that the electronic device 500 shown in Figure 5 is merely exemplary and should not constitute any limitation on the functionality and scope of the embodiments described herein. The electronic device 500 shown in Figure 5 can be used to implement the system 200 discussed above.

[0116] As shown in Figure 5, the electronic device 500 is in the form of a general-purpose electronic device. Components of the electronic device 500 may include, but are not limited to, one or more processors or processing units 510, memory 520, storage devices 530, one or more communication units 540, one or more input devices 550, and one or more output devices 560. The processing unit 510 may be a physical or virtual processor and is capable of performing various processes according to programs stored in the memory 520. In a multiprocessor system, multiple processing units execute computer-executable instructions in parallel to improve the parallel processing capability of the electronic device 500.

[0117] Electronic device 500 typically includes multiple computer storage media. Such media can be any accessible media that is accessible to electronic device 500, including but not limited to volatile and non-volatile media, removable and non-removable media. Memory 520 can be volatile memory (e.g., registers, cache, random access memory (RAM)), non-volatile memory (e.g., read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory), or some combination thereof. Storage device 530 can be a removable or non-removable medium and can include machine-readable media, such as flash drives, disks, or any other media that can be used to store information and / or data (e.g., training data for training) and can be accessed within electronic device 500.

[0118] Electronic device 500 may further include additional removable / non-removable, volatile / non-volatile storage media. Although not shown in FIG. 5, disk drives for reading from or writing to removable, non-volatile disks (e.g., "floppy disks") and optical disk drives for reading from or writing to removable, non-volatile optical disks may be provided. In these cases, each drive may be connected to a bus (not shown) via one or more data media interfaces. Memory 520 may include computer program product 525 having one or more program modules configured to perform various methods or actions of various embodiments of the present disclosure.

[0119] Communication unit 540 enables communication with other electronic devices via a communication medium. Additionally, the functionality of components of electronic device 500 can be implemented using a single computing cluster or multiple computing machines capable of communicating via communication connections. Therefore, electronic device 500 can operate in a networked environment using logical connections to one or more other servers, network personal computers (PCs), or another network node.

[0120] Input device 550 can be one or more input devices, such as a mouse, keyboard, trackball, etc. Output device 560 can be one or more output devices, such as a monitor, speaker, printer, etc. Electronic device 500 can also communicate with one or more external devices (not shown) via communication unit 540 as needed. These external devices include storage devices, display devices, etc., and can communicate with one or more devices that enable user interaction with electronic device 500, or with any device that enables electronic device 500 to communicate with one or more other electronic devices (e.g., network card, modem, etc.). Such communication can be performed via input / output (I / O) interface (not shown).

[0121] According to an exemplary implementation of this disclosure, a computer-readable storage medium is provided that stores computer-executable instructions thereon, wherein the computer-executable instructions are executed by a processor to implement the methods described above. According to an exemplary implementation of this disclosure, a computer program product is also provided, which is tangibly stored on a non-transitory computer-readable medium and includes computer-executable instructions, which are executed by a processor to implement the methods described above.

[0122] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, devices, and computer program products implemented according to this disclosure. 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.

[0123] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processing unit of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0124] Computer-readable program instructions can 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, thereby causing the instructions that execute on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0125] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0126] Various implementations of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed implementations. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described implementations. The terminology used herein is chosen to best explain the principles, practical applications, or improvements to technology in the market, or to enable others skilled in the art to understand the various implementations disclosed herein.

Claims

1. A method for model optimization, comprising: Based on the scene description information associated with the first scene, generate configuration information; Based on the configuration information, a set of interface interaction instructions associated with the development tool are generated to trigger the generation of a model interaction request corresponding to the first scenario in the development tool. Determine multiple candidate response contents generated by multiple models based on the model interaction request; Based on the first evaluation information of the multiple candidate response contents, the target response content is determined from the multiple candidate response contents; as well as Utilize the target response content to optimize at least one model associated with the development tool.

2. The method of claim 1, wherein the scene description information indicates classification information associated with a plurality of scene attributes, the plurality of scene attributes being used to describe different aspects of the first scene.

3. The method of claim 2, wherein the plurality of scene attributes includes at least one of the following: Cursor behavior indicates the type of operation performed by the cursor. Trigger mode, indicating the interaction mode used to trigger the request; Command type, indicating the type of command used to trigger the request; Dialogue turn, indicating the dialogue turn that triggered the request; Reference information indicates the information referenced in generating the response; The difficulty of the scenario indicates the difficulty of processing the request; Request type indicates the category of the requested task.

4. The method according to claim 1, further comprising: Obtain scene statistics associated with the development tool; as well as The scene statistics are sent to the first model to generate the scene description information.

5. The method according to claim 4, wherein the scene statistics information includes: First statistical information associated with a first scene attribute, the first statistical information being determined based on historical interaction operations; and / or The second statistical information associated with the second scene attribute is determined by the second model based on descriptive information associated with historical interaction operations.

6. The method according to claim 1, wherein generating configuration information based on scene description information associated with the first scene includes: Based on the scene description information and the preset prompt word template, prompt words are generated; Provide the prompt words to the third model to generate question content; as well as Based on the aforementioned problem content, the configuration information is generated.

7. The method according to claim 6, wherein providing the prompt words to the third model to generate question content includes: In response to the dialogue indicating the target round in the scenario description information, the question content generated by the third model is obtained, and the question content also indicates historical question content corresponding to at least one previous round of the target round.

8. The method of claim 6, wherein the model interaction request indicates the question content and contextual information associated with the question content.

9. The method according to claim 1, wherein generating a set of interface interaction instructions associated with the development tool based on the configuration information includes: The configuration information is provided to the fourth model along with the associated code snippets to generate second evaluation information for the problem content. as well as In response to the second evaluation information meeting the preset conditions, the set of interface interaction instructions associated with the development tool is generated based on the configuration information.

10. The method of claim 1, wherein generating a set of interface interaction instructions associated with the development tool based on the configuration information includes: Based on the configuration information, an interface interaction script is generated, which includes the set of interface interaction instructions for triggering the interface interaction process. as well as Based on the interface interaction script, the development tool is controlled to generate the model interaction request corresponding to the first scenario.

11. The method according to claim 1, further comprising: The multiple candidate response contents are provided to the fifth model to generate third evaluation information for the multiple candidate response contents; as well as Based on the third evaluation information, the first evaluation information of the multiple candidate response contents is determined.

12. The method according to claim 11, wherein the third evaluation information includes a comparative evaluation of the plurality of candidate response contents and / or an individual evaluation of the candidate response contents.

13. An apparatus for model optimization, comprising: The configuration generation module is configured to generate configuration information based on scene description information associated with the first scene; The instruction generation module is configured to generate a set of interface interaction instructions associated with the development tool based on the configuration information, so as to trigger the generation of a model interaction request corresponding to the first scenario in the development tool; The response acquisition module is configured to determine multiple candidate response contents generated by multiple models based on the model interaction request; The response determination module is configured to determine the target response content from the multiple candidate response contents based on the first evaluation information of the multiple candidate response contents; as well as The model optimization module is configured to optimize at least one model associated with the development tool using the target response content.

14. An electronic device comprising: At least one processing unit; as well as At least one memory, coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions causing the electronic device to perform the method according to any one of claims 1 to 12 when executed by the at least one processing unit.

15. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method according to any one of claims 1 to 12.

16. A computer program product comprising computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the method according to any one of claims 1 to 12.

Images