A method and device for intelligent localization of file transfer.

Abstract

The application discloses an implementation method and device for realizing localization of file transmission intelligent support, and relates to the technical field of artificial intelligence. The method comprises the following steps: determining the network scale of an AI model according to the hardware constraints of a to-be-deployed system; selecting an AI model smaller than the network scale, and preparing a specific data set related to file transmission; training the AI model by taking the specific data set related to file transmission as a data sample, to obtain a basic model; fine-tuning a discriminant model and a plurality of expert models from the basic model through fine-tuning technology, the discriminant model being used for matching an input question to a corresponding expert model, and the expert model being used for obtaining a result based on the input question; and publishing the discriminant model, the plurality of expert models and file transmission software to the to-be-deployed system. The embodiments provided by the application are applied to servers with hardware condition restrictions and network access restrictions, realize a faster reasoning speed, and improve the intelligence of file transmission.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to a method for implementing intelligent localization support for file transfer, a device for implementing intelligent localization support for file transfer, a computer-readable storage medium, and a computer program product. Background Technology

[0002] The file transfer system is a system used within banks for transferring files of various sizes without online transactions. It has been in use for over ten years and has many users. For example, version installation packages, large log files, and system-generated report data can all be transferred through the file transfer system. As the number of external users of NFT file transfer increases, the NFT file transfer project team faces more and more complex tasks and more and more problems from users, including basic usage, problem solving, and even emergency support for production issues.

[0003] In recent years, artificial intelligence, exemplified by ChatGPT, has made groundbreaking progress in natural language processing, with computers gradually approaching human-level understanding of natural language. Applying advanced AI to solve current complex support tasks will bring three benefits. First, it can significantly improve customer satisfaction. Problems encountered during NFT usage greatly impact the customer experience, and if these problems are not resolved promptly, they can substantially affect the user company's production, causing production disruptions. Second, it can significantly improve the efficiency of project teams and even the entire company. Third, it allows limited human resources to be used for more important R&D breakthroughs and innovations, shifting the business model from the traditional labor-intensive model of "selling manpower" to "technological innovation" through software licensing, accelerating the export of capabilities and efficiency, thus creating a positive cycle.

[0004] However, migrating current large language model training methods to a small scenario like NFT file transfer is often not worthwhile. There are three main reasons for this: First, the training cost of large oracle models remains prohibitively high. OpenAI's ChatGPT reportedly cost hundreds of millions of RMB to train. While domestic efforts are currently attempting to train a new large language model for under a million RMB, the cost-benefit ratio is still very high, making it difficult to implement in specific vertical scenarios. Second, inference costs are very high. Taking Alibaba Cloud and Tencent Cloud as examples, which I have researched, supporting inference on a large model with 13 billion parameters (direct inference without quantization or other compression methods, as these often lead to decreased inference quality) typically requires at least 24GB of Nvidia GPU memory. The hourly inference cost is mainly hardware, with servers equipped with GPUs costing between 20 and 30 RMB. Such conditions and costs are often not worthwhile for ordinary software support. Therefore, even if fine-tuning based on open-source large models saves on training costs, the excessively high inference costs often make it uneconomical for ordinary small projects. Third, given the current situation of banks and other users, network access is often restricted due to security and other reasons, making the availability of localized and offline services an important demand. Summary of the Invention

[0005] The purpose of this application is to provide a method and apparatus for implementing intelligent localization support for file transfer, which utilizes a small AI model and fine-tuning technology to generate a file transfer model that meets localization and hardware constraints, thereby at least solving some of the technical problems in the background art.

[0006] To achieve the above objectives, the first aspect of this application provides a method for intelligently supporting localization in file transfer, the method comprising:

[0007] The network size of the AI ​​model is determined based on the hardware constraints of the system to be deployed. An AI model smaller than the network size is selected, and a specific dataset related to file transfer is prepared. The specific dataset related to file transfer includes at least one of the following: peripheral software documentation, question-and-answer data, and common error solution data related to the file transfer software; data of the file transfer software itself, including: software documentation, error descriptions, solutions, and daily support question-and-answer data; multi-turn dialogue reasoning process data of the file transfer software; real dialogue data of the file transfer software in daily problem solving; the AI ​​model is trained using the specific dataset related to file transfer as data samples to obtain a basic model; the basic model is layered according to classification and processing functions, and a discriminant model is fine-tuned from the classification function of the basic model using fine-tuning techniques, and several expert models are fine-tuned from the processing function of the basic model using fine-tuning techniques. The discriminant model is used to match the input question to the corresponding expert model, and the expert model is used to obtain the result based on the input question; the discriminant model, several expert models, and the file transfer software are published to the system to be deployed.

[0008] Optionally, the network size is determined based on the hardware constraints of the system to be deployed, including: determining the computing power configuration and response speed based on the CPU of the system to be deployed; determining the software byte size based on the storage of the system to be deployed; determining the network framework and training time based on whether the system to be deployed has a GPU and GPU characteristics; and determining whether to use local inference based on whether the system to be deployed can access the network.

[0009] Optionally, the plurality of expert models are configured according to different types of databases and / or operating systems; the classification results of the discriminant model are configured according to the plurality of expert models.

[0010] Optionally, a discriminant model can be fine-tuned from the classification function in the base model using fine-tuning techniques, including: determining the classification result of the classification function according to the different types of the database and / or operating system; labeling the specific dataset related to file transfer based on the classification result; and fine-tuning the discriminant model using the labeled specific dataset related to file transfer and a supervised learning method.

[0011] Optionally, before training the AI ​​model using a specific dataset related to file transfer, the following preprocessing steps are further included: generating a proprietary data dictionary based on the specific dataset related to file transfer using text segmentation and tokenization tools; converting the specific dataset related to file transfer into numeric vectors according to the proprietary data dictionary; standardizing the numeric vectors into real number vectors; and dividing the real number vectors into a training set and a validation set, wherein the training set and the validation set are used for training the AI ​​model and for post-training validation, respectively.

[0012] Optionally, the file transfer software includes a weight file consisting of a set of weights for network parameters. By loading the weight file and taking user input as input, the software iteratively calculates the model to predict future outputs in order to achieve inference functionality.

[0013] This application also provides an implementation device for localized intelligent support for file transfer. The device includes: a scale determination module for determining the network scale based on the hardware constraints of the system to be deployed; and a data preparation module for selecting an AI model smaller than the network scale and preparing a specific dataset related to file transfer. The specific dataset related to file transfer includes at least one of the following: peripheral software documentation, question-and-answer data, and common error solution data related to the file transfer software; data from the file transfer software itself, including: software documentation, error descriptions, solutions, and daily support question-and-answer data; multi-turn dialogue reasoning process data from the file transfer software; and daily problem-solving data from the file transfer software. The system comprises: a real-world dialogue data set; a basic model module for training the AI ​​model using a specific dataset related to file transfer as data samples to obtain a basic model; a model fine-tuning module for stratifying the basic model according to classification and processing functions, fine-tuning a discriminant model from the classification function of the basic model using fine-tuning techniques, and fine-tuning several expert models from the processing function of the basic model using fine-tuning techniques, wherein the discriminant model is used to match the input question to the corresponding expert model, and the expert model is used to obtain a result based on the input question; and a model deployment module for deploying the discriminant model, several expert models, and file transfer software to the system to be deployed.

[0014] Optionally, the network size is determined based on the hardware constraints of the system to be deployed, including: determining the computing power configuration and response speed based on the CPU of the system to be deployed; determining the software byte size based on the storage of the system to be deployed; determining the network framework and training time based on whether the system to be deployed has a GPU and GPU characteristics; and determining whether to use local inference based on whether the system to be deployed can access the network.

[0015] Optionally, the plurality of expert models are configured according to different types of databases and / or operating systems; the classification results of the discriminant model are configured according to the plurality of expert models.

[0016] Optionally, a discriminant model can be fine-tuned from the classification function in the base model using fine-tuning techniques, including: determining the classification result of the classification function according to the different types of the database and / or operating system; labeling the specific dataset related to file transfer based on the classification result; and fine-tuning the discriminant model using the labeled specific dataset related to file transfer and a supervised learning method.

[0017] Optionally, before training the AI ​​model using a specific dataset related to file transfer, the following preprocessing steps are further included: generating a proprietary data dictionary based on the specific dataset related to file transfer using text segmentation and tokenization tools; converting the specific dataset related to file transfer into numeric vectors according to the proprietary data dictionary; standardizing the numeric vectors into real number vectors; and dividing the real number vectors into a training set and a validation set, wherein the training set and the validation set are used for training the AI ​​model and for post-training validation, respectively.

[0018] Optionally, the file transfer software includes a weight file consisting of a set of weights for network parameters. By loading the weight file and taking user input as input, the software iteratively calculates the model to predict future outputs in order to achieve inference functionality.

[0019] In a third aspect of this application, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the aforementioned file transfer intelligent support localization implementation method.

[0020] In a fourth aspect of this application, a computer-readable storage medium is also provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the steps of the aforementioned implementation method for intelligent support of localization of file transfer.

[0021] A fifth aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method for intelligent support of localization in file transfer.

[0022] The above technical solution has at least the following beneficial effects:

[0023] This application can achieve a relatively fast inference speed on servers with limited hardware configurations while meeting the requirements for question-answering support, using pure CPU inference. It can solve a considerable problem of meaningless repetitive work and fully meet the needs of intelligent question-answering support scenarios for file transfer.

[0024] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0025] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings:

[0026] Figure 1 This illustration schematically shows the steps of a method for implementing intelligent localization support for file transfer according to an embodiment of this application;

[0027] Figure 2 This illustration schematically shows an application environment diagram of the implementation method of intelligent localization support for file transfer according to an embodiment of this application;

[0028] Figure 3 This schematic diagram illustrates the structure of an implementation apparatus for intelligent localization of file transfer according to an embodiment of this application;

[0029] Figure 4 The diagram schematically illustrates the internal structure of a computer device according to an embodiment of this application. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0031] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0032] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0033] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0034] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0035] Figure 1 The illustration schematically depicts the steps of a method for implementing intelligent localization support for file transfer according to an embodiment of this application. For example... Figure 1 As shown, a method for intelligently supporting localization in file transfer is described, the method comprising:

[0036] S01. Determine the network size based on the hardware constraints of the system to be deployed;

[0037] S02. Select an AI model smaller than the network size and prepare a specific dataset related to file transfer; the specific dataset related to file transfer includes at least one of the following: peripheral software documentation, question and answer data, and common error solution data involved in the file transfer software; file transfer software data itself, including: software documentation, error descriptions, solutions, and daily support question and answer data; multi-turn dialogue reasoning process data of the file transfer software; real dialogue data of the file transfer software in daily problem solving.

[0038] S03. Train the AI ​​model using the specific dataset related to the file transfer as data samples to obtain the basic model;

[0039] S04. The basic model is layered according to the classification function and the processing function. A discriminant model is fine-tuned from the classification function of the basic model through fine-tuning technology. Several expert models are fine-tuned from the processing function of the basic model through fine-tuning technology. The discriminant model is used to match the input question to the corresponding expert model. The expert model is used to obtain the result based on the input question.

[0040] S05. Publish the discrimination model, several expert models, and file transfer software to the system to be deployed.

[0041] In this implementation S03, the specific dataset related to file transfer includes at least one of the following data: (1) peripheral software documents, Q&A data, and common error solution data involved in the file transfer software. For example, NFT involves Oracle data, MySQL database, Golden db database, Linux system, Kylin system, etc., and these peripheral data and common error solution data need to be collected. (2) File transfer software data itself, including: software documents, error descriptions, solutions, and daily support Q&A data; (3) Multi-turn dialogue reasoning process data of the file transfer software; complex problem solving often requires multiple rounds of dialogue to collect enough information before it can be solved, and these data are also very important. (4) Real dialogue data of daily problem solving in the file transfer software.

[0042] In this implementation S04, fine-tuning refers to making small adjustments to a pre-trained model to improve its accuracy and performance. In machine learning, model fine-tuning is a common technique that helps developers optimize models to adapt to new datasets and tasks. This includes various methods such as standard fine-tuning, partial layer fine-tuning, and adapter fine-tuning. Partial layer fine-tuning involves tweaking only a portion of the pre-trained model (usually the top or last few layers), while keeping most of the weights at the lower layers unchanged. This method aims to utilize the common feature representations at the lower layers while allowing the higher layers of the model to better adapt to new tasks.

[0043] Through the above implementation methods, a smaller AI model is used, combined with specific software support scenarios, to achieve file transfer under hardware constraints. Furthermore, this implementation method fine-tunes the basic model to develop a discriminative model and an expert model, accelerating inference speed and meeting the requirements of constrained scenarios.

[0044] Figure 2 This illustration schematically depicts an application environment diagram of the file transfer intelligent support for localization implementation method according to an embodiment of this application. For example... Figure 2 As shown, the file transfer intelligent localization implementation method provided in this application can be applied to, for example... Figure 2 In the application environment shown, terminal 102 communicates with server 104 via a network. Terminal 102 is used by a user, typically the initiator of business requests, and can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. Server 104 can be implemented as a standalone server or a server cluster consisting of multiple servers, running a file transfer application. This file transfer application is deployed or implemented based on the aforementioned steps S01 to S04.

[0045] In some optional embodiments of this application, the network size is determined based on the hardware constraints of the system to be deployed, including: determining the computing power configuration and response speed based on the CPU of the system to be deployed; determining the software size in bytes based on the storage of the system to be deployed; determining the network framework and training time based on whether the system to be deployed has a GPU and its characteristics; and determining whether to use localized inference based on whether the system to be deployed can access the network. For example, CPUs are more cost-effective and readily available than high-end GPUs or dedicated AI accelerators. For many applications, the inference performance provided by the CPU is sufficient, and the CPU's computing power determines the computing power configuration and the application's response speed, both during the training and inference phases. The storage of the system to be deployed is not only used to store the software itself but also needs to store cached data required for software operation; the software size in bytes is typically chosen to be within several hundred megabytes. GPUs are one of the most important components for training deep learning models. In the absence of a GPU, the response latency caused by the long inference time needs to be considered. When selecting a GPU, its compatibility with the selected deep learning framework needs to be considered. Adding network restrictions is typically to ensure the security of local data and offline use. When the system to be deployed has network restrictions, rapid model deployment based on localized inference is required. Network size is measured in terms of the number of parameters; large models often have hundreds of millions of parameters. The network size determined in this implementation is mostly in the millions of parameters, in order to achieve deployment within the hardware constraints that allow for inference speed. The aforementioned hardware constraints include, for example: a standard dual-core CPU, 8GB of server memory, no GPU, and no network access requirements. This implementation can achieve an inference speed of 30 Chinese characters per second using pure CPU inference under the aforementioned hardware constraints.

[0046] In some optional embodiments of this application, the plurality of expert models are configured according to different types of databases and / or operating systems; the classification results of the discriminant model are configured according to the plurality of expert models. For example, the discriminant model is typically set to be very small, and a supervised learning method is used to train the discriminant model to quickly identify which problem domain the user's input question belongs to. For example, if a user asks: "What could be the cause of an Oracle ora-0001 error?", the discriminant model quickly (usually less than 0.1 seconds) identifies that the Oracle actual answer model can answer the question and outputs it to the Oracle expert model for processing. The expert models are configured according to the different domains involved in the software itself. Taking NFT file transfer as an example, NFT involves Oracle data, MySQL databases, GoldenDB databases, Linux systems, Kylin systems, etc. Therefore, the expert models will be fine-tuned to include NFT expert models, Oracle expert models, MySQL expert models, etc. In this embodiment, the advantage of using such a two-layer model structure is that although a question-and-answer session requires calling two models, both models are very small, especially the discriminant model, which can quickly locate the question to a professional domain, thus facilitating rapid reasoning in constrained scenarios.

[0047] In some optional embodiments of this application, a discriminative model is fine-tuned from the classification function in the base model using fine-tuning techniques. This includes: determining the classification result of the classification function based on different types of the database and / or operating system; labeling the specific dataset related to file transfer based on the classification result; and fine-tuning the discriminative model using the labeled specific dataset related to file transfer and a supervised learning method. This embodiment provides a method for generating the discriminative model. Since the discriminative model is very small and its output is relatively fixed, a supervised learning method can be used. Correspondingly, it is necessary to set corresponding labels for the specific dataset related to file transfer to mark the classification result and generate training data accordingly. The classification result output by the discriminative model is used to determine which expert model among several expert models should process the input problem.

[0048] In some optional embodiments of the present application, before training the AI model with a specific data set related to file transfer, the following preprocessing steps are further included: generating a proprietary data dictionary based on the specific data set related to file transfer processed by a text tokenization and tagging tool; converting the specific data set related to file transfer into digital vectors according to the proprietary data dictionary; normalizing the digital vectors into real number vectors; dividing the real number vectors into a training set and a validation set, and the training set and the validation set are respectively used for training and post-training verification of the AI model. Exemplarily, based on a text tokenization and tagging tool, such as the Google open-source sentencepiece software and other software with similar functions, a proprietary data dictionary is generated based on the specific data set related to file transfer, and the proprietary data dictionary has a fixed size. According to this proprietary data dictionary, training data is converted into digital vectors. For example, the character "hello" is converted into a vector [1, 2], and the integer vector is converted into a real number vector through Embedding, and then all vectors are divided into a training set and a validation set, so as to train a basic model.

[0049] In some optional embodiments of the present application, the step of publishing the discriminant model, several expert models, and file transfer software to the to-be-deployed system in S05 includes: publishing the fine-tuned classification model, expert models, and software together. The fine-tuned classification model and expert models are usually relatively small, for example, within several hundred megabytes. After training the model, use a GC-free language such as pure C or Rust to develop the inference function, so as to achieve independence from any third-party libraries, thereby achieving an inference speed of 30 Chinese characters per second in a common server with 2 cores and 8G in a scenario without a GPU, fully meeting the requirements of the intelligent question-and-answer support scenario for file transfer (also applicable to other extensive similar systems). Users can directly communicate through a command terminal, just like a smart assistant, to solve problems that ordinary users care about.

[0050] In some optional embodiments of the present application, the step of publishing the discriminant model, several expert models, and file transfer software to the to-be-deployed system in S04 can also be implemented by the following steps: The file transfer software includes a weight file composed of a set of network parameters. By loading the weight file and taking the user input as the input, the inference function is realized by circularly calculating the prediction of the model for the subsequent output. This embodiment realizes another optimized inference scheme by publishing the weight file, thereby better achieving the goal of local CPU-free inference for intelligent question and answer.

[0051] As can be seen from the above implementation methods, the embodiments of this application combine specific software support domain scenarios, prepare high-quality datasets, and train specific data dictionaries to achieve automatic model finding of optimal network parameters. Similarly, this application proposes to fine-tune the discriminant model and expert model on the basic model, further accelerating inference and meeting the requirements of constrained scenarios, such as no GPU or no network connection.

[0052] Based on the same inventive concept, this application also provides a device for implementing intelligent localization support for file transfer. Figure 3 A schematic diagram illustrating the structure of an implementation apparatus for intelligent localization support of file transfer according to an embodiment of this application is shown. Figure 3 As shown, a device for implementing localized intelligent file transfer support includes: a scale determination module for determining the network scale based on the hardware constraints of the system to be deployed; and a data preparation module for selecting an AI model smaller than the network scale and preparing a specific dataset related to file transfer. The specific dataset related to file transfer includes at least one of the following: peripheral software documentation, Q&A data, and common error solution data related to the file transfer software; data from the file transfer software itself, including: software documentation, error descriptions, solutions, and daily support Q&A data; multi-turn dialogue reasoning process data from the file transfer software; and real-world problem-solving data from the file transfer software. The system includes: a dialogue data module; a basic model module for training the AI ​​model using a specific dataset related to file transfer as data samples to obtain a basic model; a model fine-tuning module for stratifying the basic model according to classification and processing functions, fine-tuning a discriminant model from the classification function of the basic model using fine-tuning techniques, and fine-tuning several expert models from the processing function of the basic model using fine-tuning techniques, wherein the discriminant model is used to match the input question to the corresponding expert model, and the expert model is used to obtain the result based on the input question; and a model deployment module for deploying the discriminant model, several expert models, and file transfer software to the system to be deployed.

[0053] In some optional embodiments provided in this application, the network size is determined based on the hardware constraints of the system to be deployed, including: determining the computing power configuration and response speed based on the CPU of the system to be deployed; determining the software byte size based on the storage of the system to be deployed; determining the network framework and training time based on whether the system to be deployed has a GPU and GPU characteristics; and determining whether to use local inference based on whether the system to be deployed can access the network.

[0054] In some optional embodiments provided in this application, the plurality of expert models are configured according to different types of databases and / or operating systems; the classification results of the discriminant model are configured according to the plurality of expert models.

[0055] In some optional embodiments provided in this application, a discriminant model is fine-tuned from the classification function in the base model using fine-tuning techniques, including: determining the classification result of the classification function according to the different types of the database and / or operating system; labeling the specific dataset related to file transfer based on the classification result; and fine-tuning the discriminant model using the labeled specific dataset related to file transfer and a supervised learning method.

[0056] In some optional embodiments provided in this application, before training the AI ​​model using a specific dataset related to file transfer, the following preprocessing steps are further included: generating a proprietary data dictionary based on the specific dataset related to file transfer using text segmentation and tokenization tools; converting the specific dataset related to file transfer into numeric vectors according to the proprietary data dictionary; standardizing the numeric vectors into real number vectors; and dividing the real number vectors into a training set and a validation set, wherein the training set and the validation set are used for training the AI ​​model and for post-training validation, respectively.

[0057] In some optional embodiments provided in this application, the file transfer software includes a weight file consisting of a set of weights for network parameters. By loading the weight file and taking user input as input, the software performs inference by iteratively calculating the model to predict future outputs.

[0058] The specific limitations of each functional module in the aforementioned file transfer intelligent localization support implementation device can be found in the limitations of the file transfer intelligent localization support implementation method described above, and will not be repeated here. Each module in the aforementioned device or system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module. This device, which supports intelligent localization of file transfer, also has the advantage of a high degree of intelligence in file transfer under hardware constraints.

[0059] This application provides a storage medium storing a program that, when executed by a processor, implements the above-described method for intelligent localization of file transfer.

[0060] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown. Figure 4A schematic diagram of the internal structure of a computer device according to an embodiment of this application is shown. The computer device includes a processor A01, a network interface A02, a memory (not shown), and a database (not shown) connected via a system bus. The processor A01 provides computing and control capabilities. The memory includes internal memory A03 and a non-volatile storage medium A04. The non-volatile storage medium A04 stores an operating system B01, a computer program B02, and a database (not shown). The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 stored in the non-volatile storage medium A04. The network interface A02 is used for communication with an external terminal via a network connection. When the computer program B02 is executed by the processor A01, it implements a method for intelligently supporting localized file transfer.

[0061] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0062] In one embodiment, the file transfer intelligent localization support implementation device provided in this application can be implemented as a computer program, and the computer program can be implemented in, for example... Figure 4 The computer device shown operates on the computer. The computer device's memory can store various program modules that constitute the implementation apparatus for intelligent localization support of file transfer. The computer program composed of these program modules causes the processor to execute the steps in the implementation methods for intelligent localization support of file transfer described in the various embodiments of this application.

[0063] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware implementation, a completely software implementation, or an implementation combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0064] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0065] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0066] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0067] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0068] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0069] Computer-readable media include both permanent and non-permanent, removable and non-removable media, which can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0070] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0071] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A method for implementing file transfer intelligent support localization, characterized in that, The method comprises: determining the network scale of the AI model according to the hardware constraints of the system to be deployed; selecting an AI model smaller than the network scale and preparing a file transfer related specific dataset; the file transfer related specific dataset comprises at least one of the following data: peripheral software documentation, question and answer data and error common solution data involved in the file transfer software; file transfer software itself data, including: software documentation, error description, solution and daily support question and answer data; file transfer software multi-round dialogue reasoning process data; file transfer software daily problem solving real dialogue data; training the AI model with the file transfer related specific dataset as a data sample to obtain a basic model; layering the basic model according to classification functions and processing functions, fine-tuning a discriminant model from the classification functions in the basic model through fine-tuning technology, and fine-tuning a plurality of expert models from the processing functions in the basic model through fine-tuning technology, the discriminant model being used to match an input question to a corresponding expert model, and the expert model being used to obtain a result based on the input question; publishing the discriminant model, the plurality of expert models and the file transfer software to the system to be deployed.

2. The method of claim 1, wherein, Determining the network scale of the AI model according to the hardware constraints of the system to be deployed comprises: determining the computing power configuration and response speed according to the CPU of the system to be deployed; determining the software byte size according to the storage of the system to be deployed; determining the network framework and training time consumption according to whether the system to be deployed has a GPU and the GPU characteristics; determining whether to adopt localized inference according to whether the system to be deployed can access a network.

3. The method of claim 1, wherein: the plurality of expert models are correspondingly set according to different types of databases and / or operating systems; the classification results of the discriminant model are set according to the plurality of expert models.

4. The method of claim 3, wherein, Fine-tuning a discriminant model from the classification functions in the basic model through fine-tuning technology comprises: determining the classification results of the classification functions according to different types of databases and / or operating systems; labeling the file transfer related specific dataset based on the classification results; fine-tuning the discriminant model using the labeled file transfer related specific dataset and a supervised learning method.

5. The method of claim 1, wherein, Before training the AI model using the file transfer related specific dataset, the following preprocessing steps are further included: generating a special data dictionary based on the file transfer related specific dataset based on text segmentation and tokenization tools; converting the file transfer related specific dataset into a digital vector according to the special data dictionary; standardizing the digital vector into a real number vector; dividing the real number vector into a training set and a validation set, which are respectively used for training and verifying the AI model after training.

6. The method of claim 1, wherein, The file transfer software comprises a weight file formed by a set of network parameter weights, and the inference function is realized by loading the weight file and taking user input as input, and through a loop calculation model to predict future output.

7. An apparatus for implementing intelligent support localization of file transfer, characterized by comprising: The device comprises: a scale determination module configured to determine a network scale of an AI model according to hardware constraints of a system to be deployed; a data preparation module configured to select an AI model smaller than the network scale and prepare a file transfer related specific dataset; the file transfer related specific dataset includes at least one of the following data: peripheral software documents involved in file transfer software, question and answer data, and error common solution data; file transfer software itself data, including: software documents, error descriptions, solutions, and daily support question and answer data; file transfer software multi-round dialogue reasoning process data; and real dialogue data for daily problem solving of file transfer software; a base model module configured to train the AI model with the file transfer related specific dataset as a data sample to obtain a base model; a model fine-tuning module configured to layer the base model according to classification functions and processing functions, fine-tune a discriminant model from the classification functions in the base model through fine-tuning technology, and fine-tune a plurality of expert models from the processing functions in the base model through fine-tuning technology; the discriminant model is used to match an input question to a corresponding expert model, and the expert model is used to obtain a result based on the input question; and a model publishing module configured to publish the discriminant model, the plurality of expert models, and file transfer software to the system to be deployed.

8. The apparatus of claim 7, wherein, determining a network scale of an AI model according to hardware constraints of a system to be deployed, including: determining computing power configuration and response speed according to CPU of the system to be deployed; determining software byte size according to storage of the system to be deployed; determining network framework and training time consumption according to whether the system to be deployed has a GPU and GPU characteristics; determining whether to adopt localized inference according to whether the system to be deployed can access a network.

9. The apparatus of claim 7, wherein: the plurality of expert models are correspondingly set according to different types of databases and / or operating systems; classification results of the discriminant model are set according to the plurality of expert models.

10. The apparatus of claim 9, wherein, fine-tuning a discriminant model from classification functions in the base model through fine-tuning technology, including: determining classification results of the classification functions according to different types of the databases and / or operating systems; labeling the file transfer related specific dataset based on the classification results; fine-tuning the discriminant model using the labeled file transfer related specific dataset and a supervised learning method.

11. The apparatus of claim 7, wherein, Before training the AI model using the file transfer related specific dataset, the following preprocessing steps are further included: generating a special data dictionary based on the file transfer related specific dataset based on text segmentation and tokenization tools; converting the file transfer related specific dataset into a digital vector according to the special data dictionary; standardizing the digital vector into a real number vector; dividing the real number vector into a training set and a validation set, which are respectively used for training and verifying the AI model after training.

12. The apparatus of claim 7, wherein, The file transmission software comprises a weight file formed by a set of weights of network parameters, and the weight file is loaded and user input is taken as input, and prediction of later output is calculated through a loop calculation model to realize an inference function.

13. A computer-readable storage medium having stored therein instructions, the medium being configured such that, when the instructions are executed by a processor, the processor is caused to carry out the method of any one of claims 1 to 12. When it is run on a computer, it makes the computer execute the steps of the implementation method of the localization of the file transmission intelligence support according to any one of claims 1 to 6.

14. A computer program product comprising a computer program, characterized in that, The computer program, when executed by a processor, implements the steps of the implementation method of the localization of the file transmission intelligence support according to any one of claims 1 to 6.

Images