CN118519660A

CN118519660A - Code updating method, code generation method, task processing method and code interpreter based on large model

Info

Publication number: CN118519660A
Application number: CN202410797752.9A
Authority: CN
Inventors: 吴京京; 贺思俊; 陈泽裕; 马艳军
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2024-06-19
Filing date: 2024-06-19
Publication date: 2024-08-20

Abstract

The disclosure provides a code updating method, a code generating method, a task processing method and a code interpreter based on a large model, relates to the technical field of artificial intelligence, and particularly relates to the technical fields of deep learning, natural language processing, software engineering and the large model. The specific implementation scheme is as follows: receiving operation feedback information from a test environment, wherein the operation feedback information is used for representing an operation result of codes in the test environment, and the codes are generated based on user intention by using a large language model; responding to the running feedback information to indicate code running errors, and obtaining session information based on the error state of the code; and updating the code by using the session information to obtain an updated code.

Description

Code updating method, code generating method, task processing method and code interpreter based on large model

Technical Field

The present disclosure relates to the field of artificial intelligence, and more particularly, to the field of deep learning, natural language processing, software engineering, and large model technology, and more particularly, to a code updating method, a code generating method, a task processing method, and a code interpreter based on a large model.

Background

With the continuous development of large models, the application of the large models in the professional field is also more and more widespread, for example, in the field of software engineering, the large code models can generate high-quality codes according to the context and the requirements of users, and thus the requirements on the codes are completed.

Disclosure of Invention

The disclosure provides a code updating method, a code generating method, a task processing method and a code interpreter based on a large model.

According to an aspect of the present disclosure, there is provided a large model-based code updating method, including: receiving operation feedback information from a test environment, wherein the operation feedback information is used for representing an operation result of codes in the test environment, and the codes are generated based on user intention by using a large language model; responding to the operation feedback information to indicate that the code is in error operation, and obtaining session information based on the error state of the code; and updating the code by using the session information to obtain an updated code.

According to another aspect of the present disclosure, there is provided a code generation method including: generating initial code based on intention information included in the code generation request in response to the code generation request; and in the test environment, based on the error state of the currently running code, performing iterative updating on the currently running code to obtain an object code, wherein in the first code updating round, the currently running code is the initial code.

According to another aspect of the present disclosure, there is provided a task processing method, including: responding to a task processing request, and analyzing a task text of a target task included in the task processing request to obtain intention information; generating an initial code based on the intention information; in a test environment, based on the error state of the currently running code, performing iterative updating on the currently running code to obtain an object code, wherein in a first code updating round, the currently running code is the initial code; and running the target code in a production environment to perform the target task.

According to another aspect of the present disclosure, there is provided a code interpreter comprising: a code sandbox configured to provide a test environment, the code sandbox configured to receive code, run the code in the test environment, and generate run feedback information for the code, wherein the code is generated by a large language model; and an error reflection module configured to receive the operation feedback information, respond to the operation feedback information to indicate that the code is in error operation, obtain session information based on the error state of the code, and send the session information to the large language model, wherein the large language model is configured to generate updated code based on the session information and the code.

According to another aspect of the present disclosure, there is provided a large model-based code updating apparatus including: the receiving module is used for receiving operation feedback information from the test environment, wherein the operation feedback information is used for representing an operation result of codes in the test environment, and the codes are generated based on user intention by using a large language model; the first processing module is used for responding to the operation feedback information to indicate that the code is in error operation and obtaining session information based on the error state of the code; and a first updating module, configured to update the code by using the session information, to obtain an updated code.

According to another aspect of the present disclosure, there is provided a code generating apparatus including: a first generation module, configured to generate an initial code based on intention information included in a code generation request in response to the code generation request; and a second updating module, configured to iteratively update, in a test environment, a currently running code based on an error state of the currently running code, to obtain an object code, where in a first code update round, the currently running code is the initial code.

According to another aspect of the present disclosure, there is provided a task processing device including: the analysis module is used for responding to the task processing request and analyzing the task text of the target task included in the task processing request to obtain intention information; the second generation module is used for generating initial codes based on the intention information; a third updating module, configured to iteratively update, in a test environment, a currently running code based on an error state of the currently running code, to obtain an object code, where in a first code update round, the currently running code is the initial code; and a second processing module for running the object code in a production environment to perform the object task.

According to another aspect of the present disclosure, there is provided an electronic device including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method as described above.

According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method as described above.

According to another aspect of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements a method as described above.

It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the disclosure, nor is it intended to be used to limit the scope of the disclosure. Other features of the present disclosure will become apparent from the following specification.

Drawings

The drawings are for a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

fig. 1 schematically illustrates an exemplary system architecture to which the methods and apparatus of embodiments of the present disclosure may be applied, according to embodiments of the present disclosure.

Fig. 2 schematically illustrates a flow chart of a large model based code update method according to an embodiment of the present disclosure.

Fig. 3 schematically illustrates a schematic diagram of a code update flow according to an embodiment of the present disclosure.

Fig. 4 schematically illustrates a flow chart of a large model based code update method according to another embodiment of the present disclosure.

Fig. 5 schematically shows a flowchart of a code generation method according to an embodiment of the present disclosure.

Fig. 6 schematically illustrates a flow chart of a task processing method according to an embodiment of the present disclosure.

Fig. 7A schematically illustrates an architecture diagram of a code interpreter according to an embodiment of the present disclosure.

Fig. 7B schematically illustrates an architecture diagram of a code interpreter according to another embodiment of the present disclosure.

Fig. 8 schematically illustrates a block diagram of a large model-based code updating apparatus according to an embodiment of the present disclosure.

Fig. 9 schematically shows a block diagram of a code generating apparatus according to an embodiment of the present disclosure.

Fig. 10 schematically illustrates a block diagram of a task processing device according to an embodiment of the present disclosure.

FIG. 11 illustrates a schematic block diagram of an example electronic device that may be used to implement embodiments of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and should be considered as merely exemplary. Accordingly, one of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

The code interpreter may be a tool that assists the large language model in code generation. Code interpreters are mainly aimed at solving the problem of difficulty in direct programming and data processing by non-professional programmers. It enables users without a deep programming background to also achieve complex programming tasks by translating natural language requirements into executable code. The code interpreter can effectively reduce the threshold of programming, so that more users can easily perform operations such as data processing, code editing, image conversion and the like, thereby improving the working efficiency and creativity. Code interpreters find wide application in data analysis and visualization, automation and scripting, image and video processing, website and application development, scientific computing and simulation, technical support, and troubleshooting.

However, current code interpreters suffer from certain drawbacks that affect the quality and efficiency of code generation. For example, in the processing of code execution errors, an effective code error correction mechanism is lacking, thereby limiting the improvement of development efficiency. For another example, when facing complex tasks, such as complex merging header quantity query and analysis tasks, code instructions generated by the code interpreter are irregular and cannot meet high-standard application requirements.

In view of this, embodiments of the present disclosure provide a large model-based code updating method, a code generating method, a task processing method, and a code interpreter, the large model-based code updating method including: receiving operation feedback information from a test environment, wherein the operation feedback information is used for representing an operation result of codes in the test environment, and the codes are generated based on user intention by using a large language model; responding to the running feedback information to indicate code running errors, and obtaining session information based on the error state of the code; and updating the code by using the session information to obtain an updated code.

It should be noted that fig. 1 is only an example of a system architecture to which embodiments of the present disclosure may be applied to assist those skilled in the art in understanding the technical content of the present disclosure, but does not mean that embodiments of the present disclosure may not be used in other devices, systems, environments, or scenarios. For example, in another embodiment, an exemplary system architecture to which the method and apparatus of the embodiments of the present disclosure may be applied may include a terminal device, but the terminal device may implement the method and apparatus provided by the embodiments of the present disclosure without interaction with a server.

As shown in fig. 1, a system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and/or wireless communication links, and the like.

The user may interact with the server 105 via the network 104 using the terminal devices 101, 102, 103 to receive or send messages or the like. Various communication client applications may be installed on the terminal devices 101, 102, 103, such as a knowledge reading class application, a web browser application, a search class application, an instant messaging tool, a mailbox client and/or social platform software, etc. (as examples only).

The terminal devices 101, 102, 103 may be a variety of electronic devices having a display screen and supporting web browsing, including but not limited to smartphones, tablets, laptop and desktop computers, and the like.

The server 105 may be a server providing various services, such as a background management server (by way of example only) providing support for content browsed by the user using the terminal devices 101, 102, 103. The background management server may analyze and process the received data such as the user request, and feed back the processing result (e.g., the web page, information, or data obtained or generated according to the user request) to the terminal device.

It should be noted that, the method provided by the embodiments of the present disclosure may be generally performed by the terminal device 101, 102, or 103. Accordingly, the apparatus provided by the embodiments of the present disclosure may also be provided in the terminal device 101, 102, or 103.

Or the methods provided by embodiments of the present disclosure may generally be performed by server 105. Accordingly, the apparatus provided by the embodiments of the present disclosure may be generally disposed in the server 105. The methods provided by the embodiments of the present disclosure may also be performed by a server or cluster of servers other than the server 105 and capable of communicating with the terminal devices 101, 102, 103 and/or the server 105. Accordingly, the apparatus provided by the embodiments of the present disclosure may also be provided in a server or a server cluster different from the server 105 and capable of communicating with the terminal devices 101, 102, 103 and/or the server 105.

It should be understood that the number of terminal devices, networks and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.

In the technical scheme of the disclosure, the related processes of collecting, storing, using, processing, transmitting, providing, disclosing, applying and the like of the personal information of the user all conform to the regulations of related laws and regulations, necessary security measures are adopted, and the public order harmony is not violated.

In the technical scheme of the disclosure, the authorization or consent of the user is obtained before the personal information of the user is obtained or acquired.

As shown in FIG. 2, the method includes operations S210-S230.

In operation S210, operation feedback information from a test environment is received.

In operation S220, session information is obtained based on the error state of the code in response to the operation feedback information indicating that the code is operated in error.

In operation S230, the code is updated with the session information, resulting in an updated code.

The test environment may be an environment for developing and testing code. The selection and configuration of the test environment may be related to the code requirements, for example, the test environment may be loaded with an operating system suitable for the code to run, the test environment may be configured with environment components required for the code to run under the operating system, and the like. The test environment may be provided by a separate test electronics device or the test environment may be provided by a virtual machine, not limited herein.

Code running in the test environment may refer to code that is generated based on user intent using a large language model. The user intent may include an operating system at which the code is run, the type of programming language in which the code is to be generated, the functions that the code is required to implement, input interface information for the code, output interface information for the code, and the like. Various information included by the user intention can be written into a prompt word template respectively to obtain a prompt word text, and the prompt word text is input into the large language model to obtain a code output by the large language model. Alternatively, the code running in the test environment may be code compiled by a developer based on the requirements thereof, which is not limited herein.

The operation feedback information may be used to indicate an operation result of the code in the test environment, for example, to indicate that the code is operating normally in the test environment, to indicate that the code is operating in the test environment with errors, and so on. Optionally, the operation feedback information may further include other information for characterizing the operation state of the code, such as an error code, which is not limited herein.

The error status of a code may be used to represent the complexity of the problem faced by the code, the manner in which the complexity is determined is not limited herein. For example, the complexity of the problem faced by the code may be determined according to the number of error reporting information contained in the operation feedback information, and the greater the number of error reporting information, the more complex the problem faced by the code is represented. For another example, the complexity of the problem faced by the code may be determined according to the number of updating times of the code, the updated code obtained each time may be put into the test environment again for testing, the process from the running error of the code in the test environment to the updating of the code by using the generated session information to obtain the updated code may be regarded as one updating process of the code, and the greater the number of updating times of the code, the less likely the large language model is to solve the problem faced by the code, that is, the more complex the problem faced by the code relative to the large language model.

The large language model may be subordinate to a dialog system, and the dialog information may be represented as text information directly input to the dialog system. In the dialogue system, the dialogue information and the required context information can be spliced to obtain the text which is actually input into the large language model. Alternatively, the session information may be generated based on the error status of the code, simulating a way of manually asking questions.

The code may be updated based on the session information using a large language model, resulting in updated code. The large language model may refer to a large model provided with code generation plug-ins or code generation tools, the selection and configuration of which is not limited herein.

According to the embodiment of the disclosure, the code generated by using the large language model can be tested in a test environment, and under the condition that the code running errors are confirmed, the construction of the session information is performed based on different error states of the code, and the updating of the code is performed based on the session information by using the large language model. When the codes are in different error states, session information with different text scales can be constructed, so that the efficiency and the accuracy of code generation can be considered, and the practicability of code generation is effectively improved.

The method shown in fig. 2 is further described with reference to fig. 3-4 in conjunction with the specific embodiments.

As shown in fig. 3, the code update flow may be performed in an electronic device in which a virtual machine 301 may be configured, the virtual machine 301 may be used to provide a test environment. With the large language model 302, code 303 may be generated based on user intent. Thereafter, the code 303 may be iteratively updated until a target code 304 is obtained that may function properly.

In each code update round, the code 303 may be entered into the virtual machine 301. The virtual machine 301 may run the code 303 and output the result of the running of the code 303 as running feedback information 305. The processor in the electronic device may parse the operation feedback information 305 and, in case it is determined that the operation feedback information 305 indicates that the code is operating normally, the processor may output the code 303 as the target code 304. The object code 304 may be represented as a code that can function properly and that meets the user's intent. The processor may determine an error state 306 of the code 303 in case it is determined that the run feedback information 305 indicates that the code is running in error. Based on the different states characterized by the error state 306, different session information 307 may be generated. The session information 307 may be entered as at least a portion of the input text of the large language model 302, and entered into the large language model 302 to obtain updated code. The updated code may be fed into the virtual machine 301 for running testing as code 303 in the next code update round.

The error state of the code may be defined based on various parameters generated during the execution of the code update procedure, which is not limited herein.

The manner of defining the error state of the code is not limited herein, and for example, a dichotomy may be used to define the error state of the code, and the error state of the code may be divided into a first state and a second state. The first state may represent that the problem faced by the code is a simple problem and the second state may represent that the problem faced by the code is a complex problem.

For example, a counter may be provided, which may be used to count the number of rounds of code updating, the value of which is expressed as the number of times the code has been updated. The error state of the code may be defined based on the updated number of times of the code. For the code of the current code update round, in the case where the updated number of times of the code is less than or equal to the preset value, it may be determined that the error state of the code is represented as the first state. In the case where the updated number of times of the code is greater than the preset value, it is determined that the error state of the code is represented as the second state.

The preset value may be set according to a specific application scenario, for example, may be set to 1,2, etc., which is not limited herein.

The counter may update its count value at the end of each code update round, e.g., the count value of the counter, i.e., the number of updates of the code, may be adjusted based on the self-increasing policy in response to updating the code with false negative information. The counter may be implemented using a count register, or the counter may be represented as a preset count variable, not limited herein. The self-increasing policy may be expressed, for example, as an increment of the count value based on a set step size, which may be, for example, 1.

For another example, a debugging tool may be configured in the test environment, where the debugging tool may display various problems and the number of various problems generated during the code running process, and accordingly, the running feedback information may include information related to the various problems and the number of various problems. The error state of the code may be defined based on the issue category information and the issue number information contained in the operation feedback information. In the case where the problem type information included in the operation feedback information indicates that there is no complex problem and the problem number information indicates that the number of problems is smaller than the preset threshold, it may be determined that the error state of the code is represented as the first state. In case the problem category information included in the operation feedback information indicates that there is a complex problem and/or the problem number information indicates that the number of problems is greater than or equal to a preset threshold, it may be determined that the error state of the code is represented as the second state. The preset threshold may be set according to a specific application scenario, which is not limited herein.

For simpler problems, such as definition errors of variable types, invocation errors of methods or classes, etc., large language models can generally generate codes that can perform better through a limited number of alternations. For more complex problems, such as class or method algorithm logic errors, the large language model generally requires more information and changes to solve the errors. Optionally, session information with different text lengths can be created for different error states of the code, so that the code is adaptively corrected by adopting texts with different scales based on the problems of different complexity, and therefore the efficiency and the accuracy of code generation can be considered.

According to an embodiment of the present disclosure, in a case where an error state of a code is represented as a first state, deriving session information based on the error state of the code may include the following operations:

Determining current error information based on error codes included in the operation feedback information; and obtaining session information based on the current error information and the tool call information.

A debug tool configured in the test environment may translate problems during code execution into error codes, each of which may be used to indicate a type of problem and the location of the problem in the code. For example, the error code may be expressed as: "code: 001; position: line B in file a).

A relationship table may be configured in the electronic device, which may be used to record the association between the codes in the error code and the error information template, which may be a template for describing one type of problem. For example, if the problem represented by the code 001 is a variable call error, the error information template corresponding to the code may be represented as "the problem that there is a variable call error in the X-th line of the file X". Based on the codes in the error codes, the corresponding error information templates can be determined, and then the position information in the error codes is filled into the error information templates, so that the current error information can be obtained.

The tool call information may include declarative text associated with a particular tool by which the large language model may implement the corresponding function by calling the particular tool. For example, the particular tool may include a code generation tool, a user intent resolution tool, and the like. After the large language model acquires the tool call information, a user intention analysis tool and a code generation tool can be respectively loaded, text information input by a user is analyzed by using the user intention analysis tool, an input prompt text of the model is obtained again, and the output characteristics of the large language model to the input prompt text can be processed into a string of executable codes by the code generation tool. By adding tool call information to the session information, code generation accuracy can be increased.

The current error information and tool call information may be respectively filled into a template to obtain session information.

According to an embodiment of the present disclosure, in a case where an error state of a code is represented as a second state, deriving session information based on the error state of the code may include the following operations:

Determining current error information based on error codes included in the operation feedback information, wherein the current error information represents error information of a current code updating round; and obtaining the session information based on the current error information, the historical error information of the code, the historical session information of the code and the meta information of the code.

According to an embodiment of the present disclosure, the history error information may include error information of each of a plurality of history code update rounds, the history session information may include session information of each of the plurality of history code update rounds, and the meta information includes at least one of code generation task information, model information, and tool call information.

The code generation task information may be obtained based on user intention, i.e., the code generation task information may include an operating system when the code is run, a kind of a program language of the code to be generated, a function that needs the code to be implemented, input interface information of the code, output interface information of the code, and the like.

The model information may represent information related to the large language model that the code generation makes use of, including a version number of the large language model, information of a plug-in or tool that needs to be loaded, and the like.

Alternatively, in the case where the tool call information is included in the meta information, the tool call information included in the session information may be deleted when recording the session information of the history update round, so as to reduce the overall text length of the session information.

According to the embodiment of the disclosure, the code generation effect of the current turn can be effectively improved by adding the meta information in the session information.

Alternatively, among the plurality of historical error information, there may be error information of lower importance that cannot assist in solving the problems faced by the code in the current code update round. The plurality of historical error information may be selectively filtered to obtain more efficient error information and session information generated based on the more efficient error information.

According to an embodiment of the present disclosure, obtaining session information based on current error information, historical error information of a code, historical session information of the code, and meta information of the code may include the following operations:

selecting target error information from historical error information of codes based on an information selection strategy; and splicing the current error information, the historical session information and the meta information to obtain session information.

The information selection policy may be expressed as a way to screen more efficient error information from a plurality of historical error information. Alternatively, the manner may include a most relevant selection, a most recent selection, etc., without limitation.

The most relevant selection may be represented by selecting, as the target error information, the error information having the strongest correlation with the error information in the current code update round from among the plurality of historical error information. For example, the similarity between the error information in the current code update round and each of the plurality of historical error information can be calculated by means of keyword matching and vector feature matching, and the historical error information corresponding to the maximum similarity is determined as the target error information. For another example, based on the error code acquired in the current code update round, the error information corresponding to the error code is selected from the plurality of history error information as the target error information.

The latest selection may be represented as selecting the history error information closest in time as the target error information, that is, the error information of the history code update round closest to the current code update round may be selected as the target error information from among the error information of each of the plurality of history code update rounds.

According to the embodiment of the disclosure, the historical error information is screened, so that the interference of noise information can be reduced, and the accuracy of the generated code is improved.

After the session information is obtained, the code may be updated with the session information by means of a large language model.

According to an embodiment of the present disclosure, updating a code with session information to obtain an updated code may include the following operations:

Filling the session information and codes into a prompt template to obtain a prompt text; and inputting the prompt text into the large language model, so that the large language model takes the session information as prompt information, and generating new codes based on the codes to obtain updated codes.

For example, the hint template may be expressed as: "currently there is the following information: session information: message; code: code. Please correct the code according to the provided session information. The message field in the hint template may be replaced with the session information and the code field in the hint template may be replaced with a code to obtain the hint text.

As shown in FIG. 4, the method includes operations S401-S410.

In operation S401, code is run in a test environment.

In operation S402, it is determined whether the code is operating normally. In the case where it is determined that the code is operating normally, operation S403 is performed. In the case where it is determined that the code is running in error, operation S404 is performed.

In operation S403, it is determined that the code is an object code conforming to the user' S intention.

In operation S404, current error information is acquired from feedback information of the test environment.

In operation S405, it is determined whether the updated number of times of the code is greater than a preset value. In case it is determined that the updated number of times of the code is greater than the preset value, operation S406 is performed. In case it is determined that the updated number of times of the code is less than or equal to the preset value, operation S408 is performed.

In operation S406, target error information is determined from the history error information of each of the plurality of history code update rounds.

In operation S407, the current error information, the target error information, the history session information, and the meta information are spliced to obtain session information. After operation S407 is completed, operation S409 may be performed.

In operation S408, the current error information is spliced with the tool call information to obtain session information.

In operation S409, the code is updated with the session information, resulting in an updated code.

In operation S410, error information and session information of the current code update round are recorded, and the updated number of times of the code is incrementally processed. After operation S410 is completed, the updated code may be used as the code of the next code update round, and operation S401 may be performed.

As shown in FIG. 5, the method includes operations S510-S520.

In operation S510, in response to the code generation request, an initial code is generated based on intention information included in the code generation request.

In operation S510, in the test environment, the currently running code is iteratively updated based on the error status of the currently running code to obtain the target code.

According to an embodiment of the present disclosure, the currently running code may include the initial code, i.e., in the first code update round, the currently running code is the initial code.

According to embodiments of the present disclosure, the iterative updating of the currently running code may be implemented using the large model-based code updating method as described above, and will not be described herein.

As shown in FIG. 6, the method includes operations S610-S640.

In operation S610, in response to the task processing request, task text of a target task included in the task processing request is parsed to obtain intention information.

In operation S620, an initial code is generated based on the intention information.

In operation S630, the currently running code is iteratively updated in the test environment based on the error status of the currently running code to obtain the target code.

In operation S640, object code is run in the production environment to execute the object task.

The large model-based code updating method, the code generating method, and the task processing method described above may be implemented using a code interpreter. The architecture of the code interpreter will be described below taking the implementation of a large model-based code updating method using the code interpreter as an example.

As shown in fig. 7A, the architecture of the code interpreter may be divided into three layers from top to bottom. The first layer may be a code sandbox 710, the second layer may be an error checking module 720, and the third layer may be a large language model 730.

In accordance with an embodiment of the present disclosure, the code sandbox 710 may be configured to provide a test environment. The code sandbox 710 may be configured to receive code, run the code in a test environment, and generate run feedback information for the code. Code running in the code sandbox 710 may be generated by the large language model 730.

According to embodiments of the present disclosure, the error mitigation module 720 may be configured to receive the run feedback information, respond to the run feedback information to indicate that the code is running in error, derive session information based on the error status of the code, and send the session information to the large language model 730. The large language model 730 may be configured to generate updated code based on the session information and the code. The updated code may again be sent to the code sandbox 710 for testing.

Optionally, the code interpreter may also include other functional modules so that the code interpreter may adaptively adjust the size of the generated session information.

As shown in FIG. 7B, an error management module 740 may also be included in the first layer of the code interpreter and a memory management module 750 may also be included in the second layer.

According to embodiments of the present disclosure, the error management module 740 may be configured to record a plurality of error codes, and error information corresponding to the error codes.

According to an embodiment of the present disclosure, the error checking module 720 may be configured to obtain, through the error management module 740, current error information corresponding to an error code included in the operation feedback information, and obtain session information based on the current error information and tool call information, in a case where an error state of the code is represented as a first state. The first state may indicate that the number of updates of the code is less than or equal to a preset value.

According to embodiments of the present disclosure, the memory management module 750 may be configured to record meta information of the codes and record error information and session information in a history code update round, resulting in history error information and history session information.

According to an embodiment of the present disclosure, the error checking module 720 may be configured to obtain current error information corresponding to an error code included in the operation feedback information through the error management module 740, obtain meta information, historical error information, and historical session information through the memory management module 750, and obtain session information based on the current error information, meta information, historical error information, and historical session information in case that an error state of the code is represented as a second state. The second state indicates that the number of updated times of the code is greater than a preset value.

According to the embodiment of the disclosure, by using the code interpreter, the efficiency of task processing can be effectively improved when tasks such as code generation, code correction and the like are executed.

As shown in fig. 8, the large model-based code updating apparatus 800 includes a receiving module 810, a first processing module 820, and a first updating module 830.

The receiving module 810 is configured to receive operation feedback information from the test environment, where the operation feedback information is used to represent an operation result of the code in the test environment, and the code is generated based on the user intention using the large language model.

The first processing module 820 is configured to obtain session information based on an error state of the code in response to the operation feedback information indicating that the code is in error.

The first updating module 830 is configured to update the code with the session information to obtain an updated code.

According to an embodiment of the present disclosure, the error state of the code is determined based on the updated number of times of the code.

According to an embodiment of the present disclosure, the large model based code updating apparatus 800 further includes a first determination module and a second determination module.

And the first determining module is used for determining that the error state of the code is represented as a first state under the condition that the updated times of the code are smaller than or equal to a preset value.

And the second determining module is used for determining that the error state of the code is represented as a second state when the updated times of the code are larger than a preset value.

According to an embodiment of the present disclosure, the first processing module 820 includes a first processing sub-module and a second processing sub-module.

And the first processing sub-module is used for determining current error information based on the error code included in the operation feedback information when the error state of the code is represented as a first state.

And the second processing sub-module is used for obtaining session information based on the current error information and the tool calling information.

According to an embodiment of the present disclosure, the first processing module 820 includes a third processing sub-module and a fourth processing sub-module.

And a third processing sub-module for determining current error information based on the error code included in the operation feedback information, the current error information representing error information of a current code update round, in the case that the error state of the code is represented as the second state.

And a fourth processing sub-module, configured to obtain session information based on current error information, historical error information of the code, historical session information of the code, and meta information of the code, where the historical error information includes error information of each of a plurality of historical code update rounds, the historical session information includes session information of each of the plurality of historical code update rounds, and the meta information includes at least one of code generation task information, model information, and tool call information.

According to an embodiment of the present disclosure, the fourth processing sub-module includes a first processing unit and a second processing unit.

The first processing unit is used for selecting and obtaining target error information from historical error information of the codes based on the information selection strategy.

And the second processing unit is used for splicing the current error information, the target error information, the historical session information and the meta information to obtain the session information.

According to an embodiment of the present disclosure, the first processing unit comprises a first processing subunit and a second processing subunit.

And the first processing subunit is used for selecting error information of the historical code updating round closest to the current code updating round from error information of each of the plurality of historical code updating rounds as target error information.

And a second processing subunit, configured to select, from the error information of each of the plurality of history code update rounds, the error information corresponding to the error code as the target error information.

According to an embodiment of the present disclosure, the large model based code updating apparatus 800 further comprises an adjustment module.

And the adjusting module is used for adjusting the updated times of the code based on the self-increasing strategy in response to updating the code by using the error reflection information.

According to an embodiment of the present disclosure, the first update module 830 includes a first update sub-module and a second update sub-module.

And the first updating sub-module is used for filling the session information and codes into the prompt template to obtain the prompt text.

And the second updating sub-module is used for inputting the prompt text into the large language model so that the large language model takes the session information as the prompt information, and generating new codes based on the codes to obtain updated codes.

According to an embodiment of the present disclosure, the large model based code updating apparatus 800 further comprises a third determination module.

And the third determining module is used for determining the code to be the target code conforming to the user intention in response to the running feedback information indicating that the code runs normally.

As shown in fig. 9, the code generating apparatus 900 includes a first generating module 910 and a second updating module 920.

The first generation module 910 is configured to generate, in response to the code generation request, an initial code based on intention information included in the code generation request.

And a second updating module 920, configured to iteratively update the currently running code based on the error status of the currently running code in the test environment to obtain the target code, where in the first code updating round, the currently running code is the initial code.

As shown in fig. 10, the task processing device 1000 can include a parsing module 1010, a second generation module 1020, a third update module 1030, and a second processing module 1040.

And the parsing module 1010 is configured to parse task text of a target task included in the task processing request in response to the task processing request, so as to obtain intention information.

And a second generation module 1020 for generating an initial code based on the intention information.

And a third updating module 1030, configured to iteratively update the currently running code based on the error status of the currently running code in the test environment to obtain the target code, where in the first code update round, the currently running code is the initial code.

A second processing module 1040 is configured to run object code in the production environment to perform the object task.

According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executable by the at least one processor to enable the at least one processor to perform the method as described above.

According to an embodiment of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method as described above.

According to an embodiment of the present disclosure, a computer program product comprising a computer program which, when executed by a processor, implements a method as described above.

FIG. 11 illustrates a schematic block diagram of an example electronic device that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 11, the electronic device 1100 includes a computing unit 1101 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 1102 or a computer program loaded from a storage unit 1108 into a Random Access Memory (RAM) 1103. In the RAM 1103, various programs and data required for the operation of the device 1100 can also be stored. The computing unit 1101, ROM 1102, and RAM 1103 are connected to each other by a bus 1104. An input/output (I/O) interface 1105 is also connected to bus 1104.

Various components in device 1100 are connected to an input/output (I/O) interface 1105, including: an input unit 1106 such as a keyboard, a mouse, etc.; an output unit 1107 such as various types of displays, speakers, and the like; a storage unit 1108, such as a magnetic disk, optical disk, etc.; and a communication unit 1109 such as a network card, modem, wireless communication transceiver, or the like. The communication unit 1109 allows the device 1100 to exchange information/data with other devices through a computer network such as the internet and/or various telecommunication networks.

The computing unit 1101 may be a variety of general purpose and/or special purpose processing components having processing and computing capabilities. Some examples of computing unit 1101 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1101 performs the respective methods and processes described above, such as a code update method, a code generation method, and a task processing method. For example, in some embodiments, the code update method, code generation method, and task processing method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as storage unit 1108. In some embodiments, some or all of the computer programs may be loaded and/or installed onto device 1100 via ROM 1102 and/or communication unit 1109. When the computer program is loaded into the RAM 1103 and executed by the computing unit 1101, one or more steps of the code updating method, the code generating method, and the task processing method described above may be performed. Alternatively, in other embodiments, the computing unit 1101 may be configured to perform the code update method, the code generation method, and the task processing method in any other suitable manner (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus such that the program code, when executed by the processor or controller, causes the functions/operations specified in the flowchart and/or block diagram to be implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the internet.

The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server may be a cloud server, a server of a distributed system, or a server incorporating a blockchain.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps recited in the present disclosure may be performed in parallel or sequentially or in a different order, provided that the desired results of the technical solutions of the present disclosure are achieved, and are not limited herein.

The above detailed description should not be taken as limiting the scope of the present disclosure. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present disclosure are intended to be included within the scope of the present disclosure.

Claims

1. A large model-based code updating method, comprising:

Receiving operation feedback information from a test environment, wherein the operation feedback information is used for representing an operation result of codes in the test environment, and the codes are generated based on user intention by using a large language model;

responding to the operation feedback information to indicate that the code is in error operation, and obtaining session information based on the error state of the code; and

And updating the code by utilizing the session information to obtain an updated code.

2. The method of claim 1, wherein the error state of the code is determined based on an updated number of times of the code;

The method further comprises the steps of:

determining that an error state of the code is represented as a first state in the case that the updated number of times of the code is less than or equal to a preset value; and

And determining that the error state of the code is represented as a second state in the case that the updated number of times of the code is greater than the preset value.

3. The method of claim 2, wherein the deriving session information based on the error state of the code comprises:

Determining current error information based on an error code included in the operation feedback information in a case where an error state of the code is represented as the first state; and

And obtaining the session information based on the current error information and the tool calling information.

4. The method of claim 2, wherein the deriving session information based on the error state of the code comprises:

Determining current error information based on an error code included in the operation feedback information, the current error information representing error information of a current code update round, in a case where an error state of the code is represented as the second state; and

Obtaining the session information based on the current error information, the historical error information of the code, the historical session information of the code and the meta information of the code;

The historical error information comprises error information of each of a plurality of historical code updating rounds, the historical session information comprises session information of each of the plurality of historical code updating rounds, and the meta information comprises at least one of code generating task information, model information and tool calling information.

5. The method of claim 4, wherein the deriving the session information based on the current error information, the historical error information of the code, the historical session information of the code, and meta information of the code comprises:

selecting target error information from the historical error information of the codes based on an information selection strategy; and

And splicing the current error information, the target error information, the historical session information and the meta information to obtain the session information.

6. The method of claim 5, wherein the selecting, based on an information selection policy, target error information from historical error information of the code comprises:

selecting error information of a history code update round closest to the current code update round from error information of each of the plurality of history code update rounds as target error information; or alternatively

And selecting error information corresponding to the error code from error information of each of the plurality of historical code updating rounds as the target error information.

7. The method of claim 2, further comprising:

and adjusting the updated times of the code based on the self-increasing strategy in response to updating the code by using the error reflection information.

8. The method of claim 1, wherein updating the code with the session information results in an updated code, comprising:

Filling the session information and the codes into a prompt template to obtain a prompt text; and

And inputting the prompt text into a large language model so that the large language model takes the session information as prompt information, and generating new codes based on the codes to obtain the updated codes.

9. The method of claim 1, further comprising:

And determining the code to be an object code conforming to the user intention in response to the operation feedback information indicating that the code is operating normally.

10. A code generation method, comprising:

generating initial code based on intention information included in a code generation request in response to the code generation request; and

In a test environment, based on the error state of the currently running code, performing iterative updating on the currently running code to obtain an object code, wherein in a first code updating round, the currently running code is the initial code.

11. A task processing method, comprising:

responding to a task processing request, and analyzing a task text of a target task included in the task processing request to obtain intention information;

Generating an initial code based on the intention information;

In a test environment, based on the error state of a currently operated code, performing iterative updating on the currently operated code to obtain an object code, wherein in a first code updating round, the currently operated code is the initial code; and

The target code is run in a production environment to perform the target task.

12. A code interpreter, comprising:

A code sandbox configured to provide a test environment, the code sandbox configured to receive code, run the code in the test environment, and generate run feedback information for the code, wherein the code is generated by a large language model;

and the error thinking back module is configured to receive the operation feedback information, respond to the operation feedback information to represent that the code is in operation error, obtain session information based on the error state of the code, and send the session information to the large language model, wherein the large language model is configured to generate updated code based on the session information and the code.

13. The code interpreter of claim 1, further comprising:

An error management module configured to record a plurality of error codes, and error information corresponding to the error codes;

the error thinking module is configured to obtain current error information corresponding to the error code included in the operation feedback information through the error management module when the error state of the code is represented as a first state, and obtain the session information based on the current error information and tool call information;

Wherein the first state indicates that the updated times of the code is less than or equal to a preset value.

14. The code interpreter of claim 1, further comprising:

The memory management module is configured to record meta information of the codes, record error information and session information in updating turns of the historical codes, and obtain historical error information and historical session information;

The error thinking module is configured to obtain, through the error management module, current error information corresponding to an error code included in the operation feedback information when an error state of the code is represented as a second state, obtain, through the memory management module, the meta information, the historical error information, and the historical session information, and obtain the session information based on the current error information, the meta information, the historical error information, and the historical session information;

wherein the second state indicates that the updated number of times of the code is greater than a preset value.

15. A large model-based code updating apparatus comprising:

The system comprises a receiving module, a processing module and a processing module, wherein the receiving module is used for receiving operation feedback information from a test environment, the operation feedback information is used for representing an operation result of codes in the test environment, and the codes are generated based on user intention by using a large language model;

The first processing module is used for responding to the running feedback information to indicate that the code runs in error and obtaining session information based on the error state of the code; and

And the first updating module is used for updating the codes by utilizing the session information to obtain updated codes.

16. A code generating apparatus comprising:

a first generation module, configured to generate an initial code based on intention information included in a code generation request in response to the code generation request; and

And the second updating module is used for carrying out iterative updating on the currently operated code based on the error state of the currently operated code in the test environment to obtain the target code, wherein in the first code updating round, the currently operated code is the initial code.

17. A task processing device comprising:

the analysis module is used for responding to the task processing request and analyzing the task text of the target task included in the task processing request to obtain intention information;

The second generation module is used for generating initial codes based on the intention information;

A third updating module, configured to iteratively update, in a test environment, a currently running code based on an error state of the currently running code, to obtain a target code, where in a first code update round, the currently running code is the initial code; and

And the second processing module is used for running the target code in the production environment to execute the target task.

18. An electronic device, comprising:

at least one processor; and

A memory communicatively coupled to the at least one processor; wherein,

The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-11.

19. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-11.

20. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any of claims 1-11.