CN118428858B - Warehouse management method, device, equipment and medium based on large language model - Google Patents

Abstract

The application discloses a warehouse management method, a device, equipment and a medium based on a large language model, which relate to the field of large models and comprise the following steps: creating a warehouse database based on preset warehouse information and preset cargo information; the initial state of the warehouse database is empty; constructing an initial large language model, and training the initial large language model based on a preset corpus to obtain a target large language model; and generating an operation code of the warehouse database based on a preset code generation method, and accessing the target large language model into the warehouse database so as to carry out warehouse management on the warehouse database by using the operation code based on the target large language model. By constructing a large language model and accessing a database, related personnel can automatically solve related requirements of users by using natural language to communicate with the system without a special operation interface, and the warehousing system improves the warehousing management efficiency and the user experience.

Description

A warehouse management method, device, equipment and medium based on large language model

Technical Field

The present invention relates to the field of large models, and in particular to a warehouse management method, device, equipment and medium based on a large language model.

Background Art

At present, most of the solutions for warehouse logistics management systems in the industry are still proposed by the demand side (warehouse managers), and software developers add functions to the system to meet the needs. However, this method is not scalable when conducting warehouse management, so the functions are limited to manual settings and the system cannot be integrated with other systems (such as finance, customer service, etc.).

The warehousing and logistics management methods based on AI (Artificial Intelligence) that have emerged in recent years only involve large models for processing logistics data, and cannot process human natural language, and thus cannot achieve the problem of human-system interaction. Therefore, how to improve the user interaction experience in the warehousing management process is a problem to be solved in this field.

Summary of the invention

In view of this, the purpose of the present invention is to provide a warehouse management method, device, equipment and storage medium based on a large language model. By constructing a large language model and accessing a database, there is no need for a special operation interface. Relevant personnel communicate with the system using natural language, and the warehouse system can automatically solve the relevant needs of users, thereby improving the efficiency of warehouse management and the user experience. The specific scheme is as follows:

In a first aspect, the present application provides a warehouse management method based on a large language model, comprising:

Creating a storage database based on preset warehouse information and preset cargo information; the initial state of the storage database is empty;

Constructing an initial large language model, and training the initial large language model based on a preset corpus to obtain a target large language model;

An operation code of the storage database is generated based on a preset code generation method, and the target large language model is connected to the storage database, so as to perform storage management on the storage database using the operation code based on the target large language model.

Optionally, the operation code is a code constructed based on structured query language.

Optionally, the generating the operation code of the warehouse database based on a preset code generating method includes:

Building an initial code generation model based on the target large language module;

Obtaining a target code pair input by a user, and training the initial code generation model based on the target code pair, so as to generate a corresponding operation code based on the user instruction according to the obtained target code model;

The target code pair includes a code input by a user and a corresponding action.

Optionally, constructing an initial code generation model based on the target large language module includes:

Adding a code encoder, a code decoder and a dimensionality reduction layer to the hidden layer of the target large language model;

Accordingly, the step of obtaining the target code pair input by the user and training the initial code generation model based on the target code pair includes:

Inputting the action of the target code pair into the target large language model to obtain features corresponding to the action, and reducing the dimensions of the features through the dimension reduction layer to obtain reduced-dimensional features;

Inputting the code of the target code pair into the code encoder to obtain the code corresponding to the code;

Determine a first loss corresponding to the target code based on the dimensionally reduced features and the encoding;

Inputting the code into the code decoder to obtain a corresponding logit value, and determining a second loss between the logit value and the code based on a mean square error function;

The total loss of the first loss and the second loss is determined based on preset hyperparameters, and the initial code generation model is trained to obtain the target code model based on the total loss.

Optionally, after obtaining the target code pair input by the user, the method further includes:

Data enhancement is performed on the target code pair to train the initial code generation model using the enhanced target code pair.

Optionally, the performing storage management on the storage database by using the operation code based on the target large language model includes:

Determine the features corresponding to the user instruction, and determine whether the user instruction is an access instruction to the warehouse database based on the features corresponding to the user instruction through a preset binary classification unit in the target large language model, so as to perform warehouse management on the warehouse database using the operation code according to the judgment result.

Optionally, performing warehouse management on the warehouse database according to the judgment result includes:

If the user instruction is the access instruction of the storage database, the dimension reduction layer is used to determine the dimension-reduced features corresponding to the user instruction;

Using the code decoder to obtain a structured query code of the dimension-reduced feature corresponding to the user instruction, and accessing the storage database based on the structured query code to obtain an access result;

The access result is input into the target large language model to obtain corresponding post-access features, and the features corresponding to the user instructions are updated based on the post-access features, and the updated features are input into the target large language model.

In a second aspect, the present application provides a warehouse management device based on a large language model, comprising:

A database construction module, used to create a storage database based on preset warehouse information and preset goods information; the initial state of the storage database is empty;

A model training module, used to construct an initial large language model and train the initial large language model based on a preset corpus to obtain a target large language model;

The warehouse management module is used to generate an operation code for the warehouse database based on a preset code generation method, and connect the target large language model to the warehouse database, so as to perform warehouse management on the warehouse database based on the target large language model using the operation code.

In a third aspect, the present application provides an electronic device comprising a processor and a memory; wherein the memory is used to store a computer program, and the computer program is loaded and executed by the processor to implement the aforementioned warehouse management method based on a large language model.

In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the aforementioned warehouse management method based on a large language model.

In this application, a storage database is first created based on preset warehouse information and preset cargo information, and an initial large language model is constructed. The initial large language model is trained based on a preset corpus to obtain a target large language model. Then, an operation code for the storage database is generated based on a preset code generation method, and the target large language model is connected to the storage database, so that the storage database can be managed based on the target large language model using the operation code. In this way, this application constructs a large language model, and trains the model using a massive corpus based on a large model training method, so that it has basic language understanding and reasoning capabilities. Then, the large language model is connected to the database. No special operation interface is required. Relevant personnel communicate with the system using natural language, and the storage system can automatically solve the user's related needs, eliminating the cost required for program developers to modify the code for specific functions, thereby improving the efficiency of warehouse management and the user experience.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

FIG1 is a flow chart of a warehouse management method based on a large language model provided by the present application;

FIG2 is a flowchart of a specific warehouse management method based on a large language model provided by the present application;

FIG3 is a schematic diagram of a code comparison method provided by the present application;

FIG4 is a schematic diagram of the structure of a warehouse management device based on a large language model provided by the present application;

FIG5 is a structural diagram of an electronic device provided in this application.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

At present, most of the solutions for warehouse logistics management systems in the industry are still proposed by the demand side, and software developers add functions to the system to meet the needs. However, this method is not scalable when conducting warehouse management, and is limited to manual settings. It is impossible to integrate the system with other systems. This application constructs a large language model for training, so that it has basic language understanding and reasoning capabilities, and then connects the large language model to the database. Without the need for a special operating interface, relevant personnel can communicate with the system using natural language, and the warehousing system can automatically solve the user's related needs.

As shown in FIG1 , an embodiment of the present invention discloses a warehouse management method based on a large language model, including:

Step S11: Create a storage database based on preset warehouse information and preset cargo information; the initial state of the storage database is empty.

In this embodiment, it is necessary to first construct a warehouse database using a large language model according to the pre-set warehouse information and goods information, or a general warehouse database template, so as to create and modify warehouse, goods and other information through the database structure. It can be understood that the initial state of the warehouse database is empty, and the database will be updated when adding a warehouse or modifying goods in the warehouse, and the database can be specifically divided into warehouse information, item information in the warehouse, warehouse transaction information, etc.

Step S12: construct an initial large language model, and train the initial large language model based on a preset corpus to obtain a target large language model.

In this embodiment, an initial large language model for warehouse database management can be constructed, and the large language model can be trained with a pre-set corpus to obtain a target large language model, so that after the large language model is trained with a massive corpus based on the large model training method, it has basic language understanding and reasoning capabilities. It should also be pointed out that in this embodiment, a massive corpus can be used to evaluate the model accuracy during the training process. When the evaluation index reaches the target value, it is considered that the model has sufficient understanding and reasoning capabilities, and the training stops. By pre-training and fine-tuning the large language model with large-scale corpora such as Wikipedia, news articles, and social media, the large language model can adapt to different application scenarios and data sets, achieve better performance and effects, and help to build a warehouse management system in combination with the warehouse database in this embodiment.

Step S13: Generate an operation code for the warehousing database based on a preset code generation method, and connect the target large language model to the warehousing database, so as to perform warehousing management on the warehousing database based on the target large language model using the operation code.

In this embodiment, the operation code of the warehousing database is constructed based on the preset code generation method and structured query language, and the target large language model is connected to the warehousing database, so as to operate the warehousing database based on the target large language model using the operation code of the accessible database to perform warehousing management. It is understandable that if only relying on the large language model to solve warehousing and logistics tasks, the short-term memory ability of the large language model is unreliable. On the one hand, the amount of data remembered is limited, and on the other hand, information that has not been used for a long time will be "forgotten". Therefore, in this embodiment, the storage of data is made into a separate module as a knowledge base of the large model. By storing the warehousing data in a separate database and connecting the model to the database, the long-distance forgetting problem of the model is solved.

In this embodiment, a warehousing database is first created based on preset warehouse information and preset goods information, and an initial large language model is constructed. The initial large language model is trained based on a preset corpus to obtain a target large language model. Then, an operation code for the warehousing database is generated based on a preset code generation method, and the target large language model is connected to the warehousing database, so that the warehousing database is managed using the operation code based on the target large language model. In this way, a warehousing logistics system is built based on a large language model, and the model is trained using a massive corpus based on a large model training method so that it has basic language understanding and reasoning capabilities. Then, the large language model is connected to the database. Without a special operation interface, relevant personnel communicate with the system using natural language, and the warehousing system can automatically solve the user's related needs, eliminating the cost required for program developers to modify the code for specific functions, thereby improving the efficiency of warehousing management and the user experience.

Based on the previous embodiment, it can be known that the present application can build a warehouse logistics system based on a large language model, and connect the large language model to the database to avoid the cost of modifying the code for specific functions. Next, the present embodiment will describe in detail the process of warehouse management based on a large language model. As shown in Figure 2, the present application embodiment discloses a specific warehouse management method based on a large language model, including:

Step S21, constructing an initial code generation model based on the target large language module, obtaining a target code pair input by a user, and training the initial code generation model based on the target code pair, so as to generate corresponding operation codes based on the user instructions according to the obtained target code model.

In this embodiment, when generating the operation code of the warehouse database, a "few sample semi-supervised natural language-code contrast learning" method is proposed, which allows the language model to generate the code to access the database to operate the database, thereby realizing the spontaneous access of the large language model to the database. It should be pointed out that in this embodiment, the database is usually accessed by a SQL (Structured Query Language) code. For example, to access the names of all goods in a warehouse A, the code "SELECT name FROM inventory WHERE warehouse='A'" can be used. It can be understood that although the current large model can generate code after training, in terms of accuracy, the code generated by the small model is often more accurate or contains less messy information. Therefore, in this embodiment, in order to allow the large language model to spontaneously access the database, a "few sample semi-supervised natural language-code contrast learning" method as shown in Figure 3 is proposed. First, an initial code generation model is constructed based on the target large language module, and a target code pair including a code and a corresponding action input by the user is obtained, and the initial code generation model is trained based on the target code pair to generate the corresponding operation code based on the user instruction according to the obtained target code model.

Specifically, when constructing the initial code generation model based on the target large language module, firstly, several hidden layers are added on the basis of the target large language model, including a code encoder, a code decoder, and a dimension reduction layer for language feature representation. Then, the action of the target code pair is input into the target large language model to obtain the feature corresponding to the action, and the feature is reduced in dimension through the dimension reduction layer to obtain the reduced feature. In this process, before the target code pair is input into the large language model, the target code pair can be data enhanced to train the initial code generation model with the enhanced target code pair. For example, given a series of (action, code) pairs, such as ("query all goods under warehouse A", "SELECT name FROM inventoryWHERE warehouse='A'"), these (action, code) pairs are enhanced, such as changing 'A' to 'B', to obtain more samples and improve the effect of model training. Then, the code of the target code pair is input into the code encoder to obtain the encoding corresponding to the code, and the corresponding feature representation is obtained by inputting the action into the language model. ; and input the corresponding code into the code encoder to obtain the corresponding code ; After dimensionality reduction, the reduced dimension features are obtained. Then, based on contrastive learning and amplification, the first loss corresponding to the target code can be determined according to the features and encoding after dimensionality reduction. ; Then the above code Input to the code decoder to get the corresponding logit value , and determine the above logit value and actual code based on the mean-square error (MSE) function The second loss between ; Then you can use the preset hyperparameters Determine the total loss of the first loss and the second loss , and based on the total loss, the initial code generation model is trained to obtain the target code model.

In this embodiment, the large language model will parse the user's input and obtain an intermediate link, namely "feature representation", denoted as In addition, we can build a small model of "input-encoder-decoder-output" for generating code. The code is passed through the encoder to obtain the feature representation of the code, which is recorded as . The code itself is restored through the decoder. After dimensionality reduction, we get Feature matrices of the same dimension . Take the loss function , loss function The MSE loss between the logit of the decoder generated code and the actual, It should be noted that when training based on this training target, only the encoder, decoder and dimensionality reduction are trained, and the large model is not trained.

Step S22, determine the features corresponding to the user instruction, and determine whether the user instruction is an access instruction to a warehouse database based on the features corresponding to the user instruction by a preset binary classification unit in the target large language model, so as to perform warehouse management on the warehouse database using the operation code according to the judgment result.

In this embodiment, a binary classification head can be connected to the large model ,predict Indicates whether the operation accesses the database. Specifically, this can be done by determining the features corresponding to the user instructions and judging whether the user instructions are access instructions to the storage database based on the features corresponding to the user instructions by the preset binary classification unit in the target large language model, so as to perform storage management on the storage database using the operation code according to the judgment result.

If the user instruction is an access instruction to the storage database, the dimension reduction layer is used to determine the dimension reduction features corresponding to the user instruction, and the code decoder is used to obtain the structured query code of the dimension reduction features corresponding to the user instruction, and the storage database is accessed based on the structured query code to obtain the access result, and then the access result is input into the target large language model to obtain the corresponding post-access features, and the features corresponding to the user instruction are updated based on the post-access features, so as to input the updated features into the target large language model. When the prediction result is true, calculate And pass it into the decoder to get the SQL code to access the database, and input the result into the language model to get the feature representation of the data , and order , the superimposed new Used for subsequent actions in the language model.

In this embodiment, an initial code generation model is constructed based on the target large language module, a target code pair input by the user is obtained, and the initial code generation model is trained based on the target code pair, so as to generate corresponding operation codes based on the user instructions according to the obtained target code model, and then determine the features corresponding to the user instructions, and use the preset binary classification unit in the target large language model to determine whether the user instructions are access instructions to the warehouse database based on the features corresponding to the user instructions, so as to use the operation codes to perform warehouse management on the warehouse database according to the judgment results. In this way, the few-sample self-supervised natural language-code comparison learning method proposed in this embodiment can enable the language model to spontaneously generate SQL code to access the database while ensuring the structural integrity of the database, and build a warehousing and logistics system based on the large language model. By applying the system based on the large model to warehousing and logistics management, on the basis of the generalized large model pre-trained with a massive corpus, the basic functions of warehousing and logistics are fine-tuned to improve the accuracy of answering questions or processing transactions. Compared with traditional warehousing system development, there is no need for complicated work such as complicated code writing, debugging, and testing. Instead, you only need to "tell" the model the required new functions, and the model can implement them, eliminating the cost required for program developers to modify the code for specific functions.

As shown in FIG4 , the embodiment of the present application further discloses a warehouse management device based on a large language model, including:

A database construction module 11 is used to create a storage database based on preset warehouse information and preset goods information; the initial state of the storage database is empty;

A model training module 12 is used to construct an initial large language model and train the initial large language model based on a preset corpus to obtain a target large language model;

The warehouse management module 13 is used to generate an operation code for the warehouse database based on a preset code generation method, and connect the target large language model to the warehouse database, so as to perform warehouse management on the warehouse database based on the target large language model using the operation code.

This embodiment first creates a warehousing database based on preset warehouse information and preset cargo information, and constructs an initial large language model, and trains the initial large language model based on a preset corpus to obtain a target large language model, and then generates an operation code for the warehousing database based on a preset code generation method, and connects the target large language model to the warehousing database, so as to perform warehousing management on the warehousing database based on the target large language model using the operation code. Through the above technical solution, a large language model can be constructed and trained using a massive corpus based on a large model training method, so that the model has basic language understanding and reasoning capabilities, and then the large language model is connected to the database. Without the need for a special operation interface, relevant personnel communicate with the system using natural language, and the warehousing system can automatically solve the user's related needs, eliminating the cost required for program developers to modify the code for specific functions, and improving the efficiency of warehousing management and user experience.

In some specific embodiments, the warehouse management module 13 specifically includes:

A model generation submodule, used for constructing an initial code generation model based on the target large language module;

The model training submodule is used to obtain the target code pair input by the user, and train the initial code generation model based on the target code pair, so as to generate corresponding operation codes based on user instructions according to the obtained target code model; wherein the target code pair includes the code input by the user and the corresponding action.

In some specific embodiments, the model generation submodule specifically includes:

A model building unit, used for adding a code encoder, a code decoder and a dimensionality reduction layer in the hidden layer of the target large language model;

Correspondingly, the model training submodule specifically includes:

A feature dimension reduction unit, used for inputting the action of the target code pair into the target large language model, obtaining features corresponding to the action, and reducing the features through the dimension reduction layer to obtain reduced-dimensional features;

A code encoding unit, used for inputting the code of the target code pair into the code encoder to obtain a code corresponding to the code;

A first loss determining unit, configured to determine a first loss corresponding to the target code based on the dimensionally reduced feature and the encoding;

A second loss determining unit, configured to input the code into the code decoder to obtain a corresponding logit value, and determine a second loss between the logit value and the code based on a mean square error function;

A total loss determination unit is used to determine the total loss of the first loss and the second loss based on preset hyperparameters, and to train the initial code generation model to obtain the target code model based on the total loss.

In some specific embodiments, the warehouse management module 13 further includes:

A data enhancement unit is used to perform data enhancement on the target code pair so as to train the initial code generation model using the enhanced target code pair.

In some specific embodiments, the warehouse management module 13 specifically includes:

An instruction judgment unit is used to determine the features corresponding to the user instruction, and judge whether the user instruction is an access instruction of the warehouse database based on the features corresponding to the user instruction through a preset binary classification unit in the target large language model, so as to perform warehouse management on the warehouse database using the operation code according to the judgment result.

In some specific embodiments, the warehouse management module 13 specifically includes:

a feature determination unit, configured to determine, if the user instruction is the access instruction of the storage database, a dimensionally reduced feature corresponding to the user instruction by using the dimension reduction layer;

A database access unit, used to obtain a structured query code of the dimension-reduced feature corresponding to the user instruction by using the code decoder, and access the storage database based on the structured query code to obtain an access result;

A feature updating unit is used to input the access result into the target large language model to obtain corresponding post-access features, update the features corresponding to the user instructions based on the post-access features, and input the updated features into the target large language model.

Furthermore, an embodiment of the present application also discloses an electronic device. FIG5 is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram cannot be regarded as any limitation on the scope of use of the present application.

FIG5 is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input/output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the warehouse management method based on a large language model disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

The operating system 221 is used to manage and control the hardware devices and computer program 222 on the electronic device 20, and can be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the warehouse management method based on the large language model executed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include computer programs that can be used to complete other specific tasks.

Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the warehouse management method based on a large language model disclosed above is implemented. For the specific steps of the method, reference may be made to the corresponding contents disclosed in the aforementioned embodiments, and no further description will be given here.

In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

The technical solution provided by the present application is introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technical personnel in this field, according to the idea of the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims (8)

1. A warehouse management method based on a large language model, characterized by comprising: Creating a storage database based on preset warehouse information and preset cargo information; the initial state of the storage database is empty; Constructing an initial large language model, and training the initial large language model based on a preset corpus to obtain a target large language model; Generate an operation code for the storage database based on a preset code generation method, and connect the target large language model to the storage database, so as to perform storage management on the storage database using the operation code based on the target large language model; The step of generating the operation code of the warehouse database based on a preset code generation method includes: Building an initial code generation model based on the target large language model; Obtaining a target code pair input by a user, and training the initial code generation model based on the target code pair, so as to generate a corresponding operation code based on the user instruction according to the obtained target code model; Wherein, the target code pair includes a code input by a user and a corresponding action; Furthermore, constructing an initial code generation model based on the target large language model includes: Adding a code encoder, a code decoder and a dimensionality reduction layer to the hidden layer of the target large language model; Accordingly, the step of obtaining the target code pair input by the user and training the initial code generation model based on the target code pair includes: Inputting the action of the target code pair into the target large language model to obtain features corresponding to the action, and reducing the dimensions of the features through the dimension reduction layer to obtain reduced-dimensional features; Inputting the code of the target code pair into the code encoder to obtain the code corresponding to the code; Determine a first loss corresponding to the target code based on the dimensionally reduced features and the encoding; Inputting the code into the code decoder to obtain a corresponding logit value, and determining a second loss between the logit value and the code based on a mean square error function; The total loss of the first loss and the second loss is determined based on preset hyperparameters, and the initial code generation model is trained to obtain the target code model based on the total loss. 2. According to the warehouse management method based on a large language model in claim 1, it is characterized in that the operation code is a code constructed based on structured query language. 3. The warehouse management method based on a large language model according to claim 1, characterized in that after obtaining the target code pair input by the user, it also includes: Data enhancement is performed on the target code pair to train the initial code generation model using the enhanced target code pair. 4. The warehouse management method based on a large language model according to claim 1, characterized in that when the warehouse management of the warehouse database is performed based on the target large language model using the operation code, it includes: Determine the features corresponding to the user instruction, and determine whether the user instruction is an access instruction to the warehouse database based on the features corresponding to the user instruction through a preset binary classification unit in the target large language model, so as to perform warehouse management on the warehouse database using the operation code according to the judgment result. 5. The warehouse management method based on a large language model according to claim 4 is characterized in that the warehouse management of the warehouse database according to the judgment result comprises: If the user instruction is the access instruction of the storage database, the dimension reduction layer is used to determine the dimension-reduced features corresponding to the user instruction; Using the code decoder to obtain a structured query code of the dimension-reduced feature corresponding to the user instruction, and accessing the storage database based on the structured query code to obtain an access result; The access result is input into the target large language model to obtain corresponding post-access features, and the features corresponding to the user instructions are updated based on the post-access features, and the updated features are input into the target large language model. 6. A warehouse management device based on a large language model, characterized by comprising: A database construction module, used to create a storage database based on preset warehouse information and preset goods information; the initial state of the storage database is empty; A model training module, used to construct an initial large language model and train the initial large language model based on a preset corpus to obtain a target large language model; A warehouse management module, used to generate an operation code for the warehouse database based on a preset code generation method, and connect the target large language model to the warehouse database, so as to perform warehouse management on the warehouse database using the operation code based on the target large language model; Wherein, the warehouse management module specifically includes: A model generation submodule, used to construct an initial code generation model based on the target large language model; A model training submodule, used to obtain a target code pair input by a user, and train the initial code generation model based on the target code pair, so as to generate a corresponding operation code based on the user instruction according to the obtained target code model; wherein the target code pair includes a code input by the user and a corresponding action; Furthermore, the model generation submodule specifically includes: A model building unit, used for adding a code encoder, a code decoder and a dimensionality reduction layer in the hidden layer of the target large language model; Correspondingly, the model training submodule specifically includes: A feature dimension reduction unit, used for inputting the action of the target code pair into the target large language model, obtaining features corresponding to the action, and reducing the features through the dimension reduction layer to obtain reduced-dimensional features; A code encoding unit, used for inputting the code of the target code pair into the code encoder to obtain a code corresponding to the code; A first loss determining unit, configured to determine a first loss corresponding to the target code based on the dimensionally reduced feature and the encoding; A second loss determining unit, configured to input the code into the code decoder to obtain a corresponding logit value, and determine a second loss between the logit value and the code based on a mean square error function; A total loss determination unit is used to determine the total loss of the first loss and the second loss based on preset hyperparameters, and to train the initial code generation model to obtain the target code model based on the total loss. 7. An electronic device, characterized in that the electronic device comprises a processor and a memory; wherein the memory is used to store a computer program, and the computer program is loaded and executed by the processor to implement the warehouse management method based on a large language model as described in any one of claims 1 to 5. 8. A computer-readable storage medium, characterized in that it is used to store a computer program, and when the computer program is executed by a processor, it implements the warehouse management method based on a large language model as described in any one of claims 1 to 5.