An artificial intelligence-based training method, apparatus, device, and medium

Abstract

This application provides an artificial intelligence-based training method, apparatus, device, and medium. The method includes: receiving user questions about the learning content through a human-computer interaction interface; preprocessing the question information to convert it into structured data; constructing a target large model based on a LoRA layer; after analysis and processing of the structured data, the extracted target information is fed into the target large model, which then generates answer information; finally, these answer information are displayed through the human-computer interaction interface. This application can optimize model performance through dynamic LoRA layers, achieving an efficient and personalized learning experience, while supporting content expansion and model updates, significantly improving training effectiveness.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a training method, apparatus, equipment and medium based on artificial intelligence. Background Technology

[0002] With the rapid development of artificial intelligence technology, its application in education is becoming increasingly widespread, especially in online training and distance education. Traditional training methods often rely on fixed teaching content and standardized teaching methods, making it difficult to meet the personalized needs of different learners. Meanwhile, facing a vast amount of learning resources and complex learning situations, how to efficiently extract valuable information and provide targeted feedback has become a key issue in improving training effectiveness.

[0003] In recent years, the rise of deep learning technology has provided new possibilities for the innovation of training methods. By building large-scale neural network models, deep learning can capture complex features in data and achieve accurate classification, prediction, and generation tasks. However, these models typically have a large number of parameters and computational requirements, placing high demands on hardware resources. In addition, the training and optimization process of these models is often time-consuming and prone to overfitting. Summary of the Invention

[0004] In view of this, embodiments of this application provide an artificial intelligence-based training method, apparatus, device, and medium that can optimize model performance through dynamic LoRA layers, achieve an efficient and personalized learning experience, and support content expansion and model updates, thereby significantly improving training effectiveness.

[0005] The technical solution of this application embodiment is implemented as follows:

[0006] In a first aspect, embodiments of this application provide an artificial intelligence-based training method that displays learning content through a human-computer interaction interface, the method comprising:

[0007] In response to a question request regarding the content to be learned, the system receives question information and preprocesses the question information to obtain structured data.

[0008] The structured data is analyzed and processed to obtain target information, and the target information is sent to the target large model; wherein, the target large model is built based on the LoRA layer, and the LoRA layer is used to capture changes in model parameters;

[0009] Obtain the answer information generated by the target large model and display the answer information on the human-computer interaction interface.

[0010] Secondly, embodiments of this application also provide an artificial intelligence-based training device, the device comprising:

[0011] The preprocessing module is used to respond to question requests for the content to be learned, receive question information, and preprocess the question information to obtain structured data.

[0012] An analysis module is used to analyze and process the structured data to obtain target information, and send the target information to the target large model; wherein, the target large model is built based on the LoRA layer, and the LoRA layer is used to capture changes in model parameters;

[0013] The acquisition module is used to acquire the answer information generated by the target large model and display the answer information on the human-computer interaction interface.

[0014] Thirdly, embodiments of this application also provide an electronic device, including: a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the artificial intelligence-based training method described in any of the first aspects.

[0015] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the artificial intelligence-based training method described in any one of the first aspects.

[0016] The embodiments of this application have the following beneficial effects:

[0017] (1) Improve training efficiency and personalized experience: By receiving and processing user questions in real time through the human-computer interaction interface, targeted answers can be generated quickly, significantly improving the immediacy and interactivity of training. The target large model built using the LoRA layer can efficiently capture changes in model parameters, thereby achieving accurate capture and response to user learning needs and providing a more personalized learning experience.

[0018] (2) Optimizing Model Performance and Resource Utilization: The application of dynamic LoRA layers enables large target models to maintain high performance while significantly reducing computational resources and storage requirements. By dynamically adjusting the number of LoRA layers, learning rate, and parameter values ​​of the low-rank matrix, the model can continuously optimize during training to reach its optimal state. The embodiments of this application avoid the overfitting and resource waste problems that may occur during traditional model training, thus improving resource utilization efficiency.

[0019] (3) Enhanced problem-solving capabilities: Preprocessing of problem information, including text conversion, cleaning, word segmentation, and vectorization, effectively extracts structured data, providing an accurate foundation for subsequent analysis and processing. Through entity recognition and intent analysis, it is possible to gain a deeper understanding of the essence and intent of user questions, thereby generating more accurate and useful answer information.

[0020] (4) Enhancing Learning Effectiveness and Feedback Mechanism: Recording and analyzing users' learning progress and behavior data enables objective evaluation of learning effectiveness, providing users with customized training suggestions, and further improving learning efficiency and outcomes. This application's embodiments not only focus on the user's current learning status but also dynamically adjust training content and methods based on the user's learning progress and feedback, achieving a continuously optimized learning path.

[0021] (5) Support for content expansion and model updates: Responding to user requests for expanded learning content, the system can continuously update and enrich training content, maintaining the timeliness and comprehensiveness of training. Based on the expanded content, the system updates and trains the target large model, ensuring that the model always keeps pace with the latest learning needs, providing users with a superior learning experience.

[0022] In summary, the embodiments of this application demonstrate significant beneficial effects in improving training efficiency, optimizing model performance, enhancing problem-solving capabilities, improving learning outcomes and feedback mechanisms, and supporting content expansion and model updates. They have important practical application value and promising prospects for promotion. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating steps S101-S103 provided in the embodiments of this application;

[0025] Figure 2 This is a schematic diagram illustrating the training principle of the target large model provided in the embodiments of this application;

[0026] Figure 3 This is a flowchart illustrating steps S301-S303 provided in the embodiments of this application;

[0027] Figure 4 This is a flowchart illustrating steps S401-S402 provided in the embodiments of this application;

[0028] Figure 5This is a schematic diagram of the analysis and processing principle provided in the embodiments of this application;

[0029] Figure 6 This is a flowchart illustrating steps S601-S603 provided in the embodiments of this application;

[0030] Figure 7 This is a schematic diagram of an artificial intelligence-based training system provided in an embodiment of this application;

[0031] Figure 8 This is a schematic diagram of the structure of the artificial intelligence-based training device provided in the embodiments of this application;

[0032] Figure 9 This is a schematic diagram of the composition structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0033] 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 accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0034] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0035] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0036] In the following description, the terms "first, second, third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0037] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application and is not intended to limit this application.

[0039] See Figure 1 , Figure 1 This is a flowchart illustrating steps S101-S103 of the artificial intelligence-based training method provided in this application embodiment, which will be combined with... Figure 1 Steps S101-S103 shown will be explained.

[0040] In step S101, in response to a question request for the content to be learned, question information is received and the question information is preprocessed to obtain structured data.

[0041] Here, when a user asks questions about the learning content, the system receives these questions through a human-computer interaction interface. The received questions are preprocessed to extract key information and construct structured data. This helps the system understand the user's questions more accurately.

[0042] In step S102, the structured data is analyzed and processed to obtain target information, and the target information is sent to the target large model; wherein, the target large model is built based on the LoRA layer, and the LoRA layer is used to capture changes in model parameters.

[0043] Here, appropriate methods are used to analyze and process the structured data, extracting key information relevant to the target. Based on the analysis results, target information is determined, which is then used to send requests to the target large model. The target large model is built based on a LoRA (Low-Rank Adaptation) layer. The role of the LoRA layer is to capture changes in model parameters, enabling the model to adapt to different tasks and datasets without having to train the entire model from scratch. The target information obtained from the analysis is sent to the target large model to request it to generate response information.

[0044] In step S103, the answer information generated by the target large model is obtained and displayed on the human-computer interaction interface.

[0045] Here, after receiving the target information, the target model generates corresponding response information based on its own training data and logic. The system obtains this response information through a human-computer interaction interface and clearly displays it to the user to answer their questions.

[0046] In some embodiments, the target large model includes a dynamic LoRA layer, which is based on a dynamic low-rank matrix; the target large model is constructed in the following manner:

[0047] Collect the learning content and question-and-answer feedback data as training data;

[0048] Add the aforementioned dynamic LoRA layer to the base model and set the parameter values ​​for the number of LoRA layers, learning rate, and low-rank matrix;

[0049] The base model is trained based on the training data, and during the training process, the loss value is reduced by dynamically adjusting the number of LoRA layers, the learning rate, and the parameter values ​​of the low-rank matrix until the base model reaches its optimal state.

[0050] The base model that has reached its optimal state is taken as the target large model.

[0051] For an example, please see Figure 2 , Figure 2 This is a schematic diagram illustrating the training principle of the target large model provided in the embodiments of this application, such as... Figure 2 As shown, the first step is to organize relevant training materials, learning libraries, and question-and-answer feedback data, transforming them into a format suitable for fine-tuning the model. This includes constructing question-and-answer pairs and collecting user evaluations and feedback. This data will serve as the foundation for fine-tuning training, ensuring the model can learn relevant domain expertise and user habits.

[0052] Next, a LoRA layer class is defined, which contains a low-rank matrix as its core component. The low-rank matrix is ​​used to capture changes in model parameters, thereby enabling fine-tuning of the model. By adjusting the parameters of the low-rank matrix, the model can be adapted to a specific task in a relevant domain without changing most of the parameters of the base model. When the model encounters a loss value that is difficult to reduce during training, new LoRA layers can be added to increase the model's complexity, thereby improving the model's performance.

[0053] Within the LoRA layer, the size (number of rows and columns) of the low-rank matrix can be dynamically adjusted based on the loss value during training. By adjusting the size of the low-rank matrix, the model's performance can be further fine-tuned to better adapt to the data and tasks in the relevant domain.

[0054] The learning rate is dynamically adjusted based on the loss value to ensure that the model can steadily reduce the loss value during training. When the loss value no longer decreases significantly, the learning rate is reduced to avoid overfitting or oscillations in the later stages of training.

[0055] Specifically, the ChatGLM3-6B model can be fine-tuned using the prepared dataset. During training, the number of LoRA layers, the parameters of the low-rank matrix AB, and the learning rate are dynamically adjusted based on changes in the loss value. The specific process is as follows:

[0056] Initially set the number of LoRA layers (default is 1), the maximum number of layers (default is 20), the number of layers added per layer (1), the initial learning rate (0.05), and the parameter values ​​of the initial low-rank matrix AB (rank_min=10, rank_max=100).

[0057] After each training epoch, it is determined whether the current loss value is smaller than the loss value of the previous epoch. If the loss value does not improve for three consecutive epochs, fine-tuning training ends. If the current loss value is smaller than the loss value of the previous epoch, parameter fine-tuning is performed: the number of LoRA layers is increased, the parameters of the low-rank matrix AB are adjusted (rank_min is halved, rank_max is increased by half), and the learning rate is halved, and then training for the next epoch begins.

[0058] Repeat the above process until the training is complete.

[0059] Compared to existing LoRA techniques, the above approach utilizes dynamic LoRA layers and a dynamic LoRA low-rank matrix. In the early stages of training, the model uses fewer LoRA layers, gradually increasing the number of LoRA layers as the training logic becomes more complex and the loss value decreases. This dynamic layering method allows for faster learning of more knowledge while reducing the number of learning and training iterations. Furthermore, dynamically adjusting the number of LoRA layers, low-rank matrix parameters, learning rate, and optimizer based on the loss value further improves training efficiency and question-answering accuracy.

[0060] The above embodiments enable the rapid learning of complex language features in relevant fields, reducing the number of training iterations and improving fine-tuning efficiency and question-answering accuracy. This method is not only applicable to relevant fields but can also be extended to artificial intelligence applications in other areas.

[0061] In some embodiments, refer to Figure 3 , Figure 3 which is a schematic flowchart of steps S301 - S303 provided by an embodiment of the present application. The problem information includes text data and / or voice data. The preprocessing of the problem information to obtain structured data can be achieved through steps S301 - S303, and will be described in combination with each step.

[0062] In step S301, the problem information is subjected to conversion processing to obtain text information.

[0063] In step S302, the text information is subjected to cleaning processing to remove irrelevant characters, stop words, and noise in the text information.

[0064] In step S303, the cleaned text information is subjected to word segmentation processing to obtain at least one token, and the at least one token is converted into a text vector, and the text vector is used as the structured data.

[0065] Here, for voice data, we use a speech recognition engine (such as an ASR system) to convert it into text information. For text data, it can be directly used.

[0066] Special symbols, numbers (unless directly related to the problem), HTML tags, and other non - text contents in the text are deleted to ensure the purity of the text information. Stop words are words that frequently appear in a language but contribute little to the meaning of the text (such as "de", "le", etc.). By removing these words, we can reduce data sparsity and improve the efficiency of subsequent processing. The text information may also include spelling mistakes, repeated words, meaningless phrases, etc. By removing these noises, the accuracy and relevance of the text information can be further improved.

[0067] The cleaned text information is cut into smaller units, namely tokens. Converting the tokenized tokens into vector representations can be achieved through word embedding techniques (such as Word2Vec, BERT, etc.), which can map words to vectors in a high - dimensional space, thereby capturing the semantic relationships between words. Then the obtained text vectors are output as structured data for subsequent model processing and analysis.

[0068] The above - mentioned method can convert the original problem information (whether text or voice) into a structured data form, that is, text vectors. These vectors not only remove irrelevant information and noise but also retain the core semantic information of the text, providing a solid foundation for subsequent natural language processing tasks (such as text classification, sentiment analysis, intelligent question - answering, etc.).

[0069] In some embodiments, refer to Figure 4 , Figure 4 This is a flowchart illustrating steps S401-S402 provided in the embodiments of this application. The analysis and processing of the structured data to obtain target information can be achieved through steps S401-S402, which will be explained in conjunction with each step.

[0070] In step S401, entity information in the structured data is identified by an entity recognition model, and input intent is determined by an intent analysis model.

[0071] In step S402, the entity information and the input intent are integrated to obtain the target information.

[0072] For an example, please see Figure 5 , Figure 5 This is a schematic diagram of the analysis and processing principle provided in the embodiments of this application, such as... Figure 5 As shown, this application provides multiple models for semantic analysis. Specifically, the multiple models may include an entity recognition model and an intent analysis model.

[0073] Entity recognition models utilize architectures such as Conditional Random Fields (CRF), Long Short-Term Memory (LSTM), or Transformer to identify entities in structured data (e.g., text vectors). The task of entity recognition models is to identify entities with specific meanings from text, such as names of people, places, times, and organizations. This entity information is fundamental to understanding text content and extracting key information.

[0074] For intent analysis models, deep learning, machine learning, or rule-based methods can be used to determine the intent of structured data. The task of an intent analysis model is to understand the user's input intent, that is, what task the user wants to complete or what information they want to obtain.

[0075] After identifying entity information and determining the input intent, these two pieces of information are integrated and processed, including associating entity information with intent to form a complete and meaningful information structure. Through this integration process, the target information—the specific content the user wants to obtain or the task they want to complete—can be extracted. This information may be output in a structured format (such as JSON, XML, etc.) to facilitate subsequent database queries, API calls, or user interface displays.

[0076] The methods described above enable in-depth analysis of structured data, extracting the entity information and input intent that users want to obtain, and then integrating them into target information. This target information is crucial for applications such as intelligent question answering, natural language understanding, and dialogue systems. It not only understands user intent and needs but also provides the key information required to achieve those needs.

[0077] In some embodiments, see Figure 6 , Figure 6 This is a flowchart illustrating steps S601-S603 provided in the embodiments of this application. The method further includes steps S601-S603, which will be explained in conjunction with each step.

[0078] In step S601, learning progress information for the content to be learned is obtained and learning behavior data is recorded.

[0079] In step S602, the learning effect is determined based on the learning progress information and the learning behavior data.

[0080] In step S603, training suggestions for the learning effect are fed back through the target large model.

[0081] Here, the system obtains information on a user's learning progress for specific learning content (such as courses, chapters, and knowledge points) through user interaction, completion of learning tasks, and recording of learning time. In addition to learning progress, the system also records various behavioral data of the user during the learning process, such as learning duration, Q&A information, number of views, and note-taking. This data helps to comprehensively evaluate the user's learning status and habits.

[0082] The system utilizes learning progress information and learning behavior data to evaluate the user's learning performance through algorithms or models (such as machine learning models, deep learning models, etc.). Based on the evaluation results, the system generates a learning performance report, which includes information such as the user's strengths, weaknesses, areas requiring focused attention, and potential learning bottlenecks.

[0083] The system uses large-scale target models (such as large language models, knowledge graph models, etc.) to analyze learning performance reports and generate training recommendations tailored to each user's individual circumstances. These recommendations include suggested learning resources, learning paths, practice questions, and review strategies.

[0084] Finally, the system will provide the user with training suggestions to help them better plan their learning and improve their learning efficiency. The feedback can take the form of text, charts, voice prompts, etc., depending on user preferences and system functionality.

[0085] The methods described above comprehensively track users' learning progress, record learning behavior data, evaluate learning effectiveness based on this data, and ultimately generate personalized training recommendations. This approach not only helps improve users' learning experience and efficiency but also provides educators and training institutions with valuable teaching feedback and improvement suggestions.

[0086] In some embodiments, the method further includes:

[0087] In response to an expansion request for the content to be learned, the content to be learned is expanded.

[0088] The target large model is updated and trained based on the expanded learning content.

[0089] Here, users or system administrators can submit requests to expand upon the learning content. These requests may stem from a user's need for a deeper understanding of a particular knowledge point, an interest in exploring related fields, or suggestions for supplementing and improving the learning content. The system analyzes the received expansion requests to determine the specific content, scope, and difficulty of the expansion. Based on the analysis results, the system expands the learning content, including adding new knowledge points, updating existing content, and introducing relevant cases and exercises. The expanded content should ensure continuity with the original learning content and maintain logical coherence and consistency.

[0090] After expanding the content to be learned, the system needs to prepare new training data. This data includes expanded descriptions of knowledge points, relevant cases, and practice questions. Simultaneously, the system must ensure the accuracy and completeness of the training data to avoid negatively impacting model training. Then, the updated training data is used to train the target large model.

[0091] The embodiments of this application will be described in full below. Please refer to [link / reference]. Figure 7 , Figure 7 This is a schematic diagram of an artificial intelligence-based training system provided in an embodiment of this application, such as... Figure 7 As shown, training materials are the starting point of the process, containing all original data related to learning or training. The Training Learning module is responsible for collecting and organizing the "training materials" and providing them to subsequent modules. It is the foundation of the entire process, ensuring the accuracy and completeness of the materials. The Progress Monitoring module is responsible for monitoring the progress of the entire process, ensuring that each step is completed on time, specifically involving the management of time, resources, and workload. The Adjust Training Content and Suggestions module evaluates the materials provided by the "Training Learning Module" and proposes modifications based on the actual situation. The Large Model Fine-tuning module (i.e., the target large model) fine-tunes the relevant models based on feedback from the "Adjust Training Content and Suggestions" module. The Interactive Page is a user interface that allows users to interact with the system. Users can ask questions, view feedback, or participate in discussions through this page. The Question Asking module is responsible for collecting questions asked by users through the "Interactive Page" and passing them to the subsequent "Answering Module." The Answering Module generates an answer to the question in the "Question Asking Module," based on model predictions in the "Large Model Fine-tuning Module." The Analysis module is used to further analyze and evaluate the answers provided by the "Answering Module" to ensure their accuracy and effectiveness. The Q&A history and feedback section records all questions and answers, as well as user feedback.

[0092] In summary, the embodiments of this application have the following beneficial effects:

[0093] (1) Improve training efficiency and personalized experience: By receiving and processing user questions in real time through the human-computer interaction interface, targeted answers can be generated quickly, significantly improving the immediacy and interactivity of training. The target large model built using the LoRA layer can efficiently capture changes in model parameters, thereby achieving accurate capture and response to user learning needs and providing a more personalized learning experience.

[0094] (2) Optimizing Model Performance and Resource Utilization: The application of dynamic LoRA layers enables large target models to maintain high performance while significantly reducing computational resources and storage requirements. By dynamically adjusting the number of LoRA layers, learning rate, and parameter values ​​of the low-rank matrix, the model can continuously optimize during training to reach its optimal state. The embodiments of this application avoid the overfitting and resource waste problems that may occur during traditional model training, thus improving resource utilization efficiency.

[0095] (3) Enhanced problem-solving capabilities: Preprocessing of problem information, including text conversion, cleaning, word segmentation, and vectorization, effectively extracts structured data, providing an accurate foundation for subsequent analysis and processing. Through entity recognition and intent analysis, it is possible to gain a deeper understanding of the essence and intent of user questions, thereby generating more accurate and useful answer information.

[0096] (4) Enhancing Learning Effectiveness and Feedback Mechanism: Recording and analyzing users' learning progress and behavior data enables objective evaluation of learning effectiveness, providing users with customized training suggestions, and further improving learning efficiency and outcomes. This application's embodiments not only focus on the user's current learning status but also dynamically adjust training content and methods based on the user's learning progress and feedback, achieving a continuously optimized learning path.

[0097] (5) Support for content expansion and model updates: Responding to user requests for expanded learning content, the system can continuously update and enrich training content, maintaining the timeliness and comprehensiveness of training. Based on the expanded content, the system updates and trains the target large model, ensuring that the model always keeps pace with the latest learning needs, providing users with a superior learning experience.

[0098] In summary, the embodiments of this application demonstrate significant beneficial effects in improving training efficiency, optimizing model performance, enhancing problem-solving capabilities, improving learning outcomes and feedback mechanisms, and supporting content expansion and model updates. They have important practical application value and promising prospects for promotion.

[0099] Based on the same inventive concept, this application also provides an artificial intelligence-based training device corresponding to the artificial intelligence-based training method in the first embodiment. Since the principle of the device in this application is similar to that of the above-mentioned artificial intelligence-based training method, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0100] like Figure 8 As shown, Figure 8 This is a schematic diagram of the structure of the artificial intelligence-based training device 800 provided in this application embodiment. The artificial intelligence-based training device 800 includes:

[0101] The preprocessing module 801 is used to respond to a question request for the content to be learned, receive question information, and preprocess the question information to obtain structured data;

[0102] Analysis module 802 is used to analyze and process the structured data to obtain target information, and send the target information to the target large model; wherein, the target large model is built based on the LoRA layer, and the LoRA layer is used to capture changes in model parameters;

[0103] The acquisition module 803 is used to acquire the answer information generated by the target large model and display the answer information on the human-computer interaction interface.

[0104] Those skilled in the art should understand that Figure 8 The functions of each unit in the AI-based training device 800 shown can be understood by referring to the relevant description of the AI-based training method described above. Figure 8 The functions of each unit in the AI-based training device 800 shown can be implemented through a program running on a processor or through specific logic circuits.

[0105] In one possible implementation, the target large model includes a dynamic LoRA layer, which is based on a dynamic low-rank matrix; the target large model is constructed in the following manner:

[0106] Collect the learning content and question-and-answer feedback data as training data;

[0107] Add the aforementioned dynamic LoRA layer to the base model and set the parameter values ​​for the number of LoRA layers, learning rate, and low-rank matrix;

[0108] The base model is trained based on the training data, and during the training process, the loss value is reduced by dynamically adjusting the number of LoRA layers, the learning rate, and the parameter values ​​of the low-rank matrix until the base model reaches its optimal state.

[0109] The base model that has reached its optimal state is taken as the target large model.

[0110] In one possible implementation, reducing the loss value during training by dynamically adjusting the number of LoRA layers, the learning rate, and the parameter values ​​of the low-rank matrix includes:

[0111] During training, the changes in the loss value are monitored. If the loss value does not improve within a specified number of consecutive epochs, the number of LoRA layers, the learning rate, and the parameter values ​​of the low-rank matrix are adjusted using a preset adjustment strategy. The adjustment strategy includes: determining whether the loss value obtained in the previous epoch is greater than the loss value obtained in the current training; if not, proceeding to the next epoch, and ending training if the loss value does not change within a specified number of consecutive epochs; if so, increasing the number of LoRA layers by 1, halving the minimum rank of the low-rank matrix, increasing the maximum rank of the low-rank matrix by half, and decreasing the learning rate by half before continuing training to obtain the loss value after training.

[0112] In one possible implementation, the problem information includes text data and / or voice data. The preprocessing module 801 preprocesses the problem information to obtain structured data, including:

[0113] The problem information is converted into text information;

[0114] The text information is cleaned to remove irrelevant characters, stop words, and noise.

[0115] The cleaned text information is segmented to obtain at least one word element, and the at least one word element is converted into a text vector, which is then used as the structured data.

[0116] In one possible implementation, the analysis module 802 analyzes and processes the structured data to obtain target information, including:

[0117] The entity recognition model identifies entity information in the structured data, and the intent analysis model determines the input intent.

[0118] The entity information and the input intent are integrated and processed to obtain the target information.

[0119] In one possible implementation, the acquisition module 803 further includes:

[0120] Obtain learning progress information for the content to be learned and record learning behavior data;

[0121] The learning effect is determined based on the learning progress information and the learning behavior data;

[0122] The target large model provides feedback on training suggestions regarding the learning outcomes.

[0123] In one possible implementation, the acquisition module 803 further includes:

[0124] In response to an expansion request for the content to be learned, the content to be learned is expanded.

[0125] The target large model is updated and trained based on the expanded learning content.

[0126] The aforementioned AI-based training device has the following beneficial effects:

[0127] (1) Improve training efficiency and personalized experience: By receiving and processing user questions in real time through the human-computer interaction interface, targeted answers can be generated quickly, significantly improving the immediacy and interactivity of training. The target large model built using the LoRA layer can efficiently capture changes in model parameters, thereby achieving accurate capture and response to user learning needs and providing a more personalized learning experience.

[0128] (2) Optimizing Model Performance and Resource Utilization: The application of dynamic LoRA layers enables large target models to maintain high performance while significantly reducing computational resources and storage requirements. By dynamically adjusting the number of LoRA layers, learning rate, and parameter values ​​of the low-rank matrix, the model can continuously optimize during training to reach its optimal state. The embodiments of this application avoid the overfitting and resource waste problems that may occur during traditional model training, thus improving resource utilization efficiency.

[0129] (3) Enhanced problem-solving capabilities: Preprocessing of problem information, including text conversion, cleaning, word segmentation, and vectorization, effectively extracts structured data, providing an accurate foundation for subsequent analysis and processing. Through entity recognition and intent analysis, it is possible to gain a deeper understanding of the essence and intent of user questions, thereby generating more accurate and useful answer information.

[0130] (4) Enhancing Learning Effectiveness and Feedback Mechanism: Recording and analyzing users' learning progress and behavior data enables objective evaluation of learning effectiveness, providing users with customized training suggestions, and further improving learning efficiency and outcomes. This application's embodiments not only focus on the user's current learning status but also dynamically adjust training content and methods based on the user's learning progress and feedback, achieving a continuously optimized learning path.

[0131] (5) Support for content expansion and model updates: Responding to user requests for expanded learning content, the system can continuously update and enrich training content, maintaining the timeliness and comprehensiveness of training. Based on the expanded content, the system updates and trains the target large model, ensuring that the model always keeps pace with the latest learning needs, providing users with a superior learning experience.

[0132] In summary, the embodiments of this application demonstrate significant beneficial effects in improving training efficiency, optimizing model performance, enhancing problem-solving capabilities, improving learning outcomes and feedback mechanisms, and supporting content expansion and model updates. They have important practical application value and promising prospects for promotion.

[0133] like Figure 9 As shown, Figure 9 This is a schematic diagram of the composition structure of the electronic device 900 provided in the embodiments of this application. The electronic device 900 includes:

[0134] The device 900 includes a processor 901, a storage medium 902, and a bus 903. The storage medium 902 stores machine-readable instructions executable by the processor 901. When the electronic device 900 is running, the processor 901 communicates with the storage medium 902 via the bus 903. The processor 901 executes the machine-readable instructions to perform the steps of the artificial intelligence-based training method described in the embodiments of this application.

[0135] In practical applications, the various components in the electronic device 900 are coupled together via a bus 903. It is understood that the bus 903 is used to achieve communication between these components. In addition to a data bus, the bus 903 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 9 The general designated all buses as Bus 903.

[0136] The above-mentioned electronic devices have the following beneficial effects:

[0137] (1) Improve training efficiency and personalized experience: By receiving and processing user questions in real time through the human-computer interaction interface, targeted answers can be generated quickly, significantly improving the immediacy and interactivity of training. The target large model built using the LoRA layer can efficiently capture changes in model parameters, thereby achieving accurate capture and response to user learning needs and providing a more personalized learning experience.

[0138] (2) Optimizing Model Performance and Resource Utilization: The application of dynamic LoRA layers enables large target models to maintain high performance while significantly reducing computational resources and storage requirements. By dynamically adjusting the number of LoRA layers, learning rate, and parameter values ​​of the low-rank matrix, the model can continuously optimize during training to reach its optimal state. The embodiments of this application avoid the overfitting and resource waste problems that may occur during traditional model training, thus improving resource utilization efficiency.

[0139] (3) Enhanced problem-solving capabilities: Preprocessing of problem information, including text conversion, cleaning, word segmentation, and vectorization, effectively extracts structured data, providing an accurate foundation for subsequent analysis and processing. Through entity recognition and intent analysis, it is possible to gain a deeper understanding of the essence and intent of user questions, thereby generating more accurate and useful answer information.

[0140] (4) Enhancing Learning Effectiveness and Feedback Mechanism: Recording and analyzing users' learning progress and behavior data enables objective evaluation of learning effectiveness, providing users with customized training suggestions, and further improving learning efficiency and outcomes. This application's embodiments not only focus on the user's current learning status but also dynamically adjust training content and methods based on the user's learning progress and feedback, achieving a continuously optimized learning path.

[0141] (5) Support for content expansion and model updates: Responding to user requests for expanded learning content, the system can continuously update and enrich training content, maintaining the timeliness and comprehensiveness of training. Based on the expanded content, the system updates and trains the target large model, ensuring that the model always keeps pace with the latest learning needs, providing users with a superior learning experience.

[0142] In summary, the embodiments of this application demonstrate significant beneficial effects in improving training efficiency, optimizing model performance, enhancing problem-solving capabilities, improving learning outcomes and feedback mechanisms, and supporting content expansion and model updates. They have important practical application value and promising prospects for promotion.

[0143] This application also provides a computer-readable storage medium storing executable instructions that, when executed by at least one processor 901, implement the artificial intelligence-based training method described in this application.

[0144] In some embodiments, the storage medium may be a magnetic random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CDROM), etc.; or it may be a device that includes one or any combination of the above-mentioned memories.

[0145] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0146] As an example, executable instructions may, but do not necessarily, correspond to files in the file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0147] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0148] The aforementioned computer-readable storage media have the following beneficial effects:

[0149] (1) Improve training efficiency and personalized experience: By receiving and processing user questions in real time through the human-computer interaction interface, targeted answers can be generated quickly, significantly improving the immediacy and interactivity of training. The target large model built using the LoRA layer can efficiently capture changes in model parameters, thereby achieving accurate capture and response to user learning needs and providing a more personalized learning experience.

[0150] (2) Optimizing Model Performance and Resource Utilization: The application of dynamic LoRA layers enables large target models to maintain high performance while significantly reducing computational resources and storage requirements. By dynamically adjusting the number of LoRA layers, learning rate, and parameter values ​​of the low-rank matrix, the model can continuously optimize during training to reach its optimal state. The embodiments of this application avoid the overfitting and resource waste problems that may occur during traditional model training, thus improving resource utilization efficiency.

[0151] (3) Enhanced problem-solving capabilities: Preprocessing of problem information, including text conversion, cleaning, word segmentation, and vectorization, effectively extracts structured data, providing an accurate foundation for subsequent analysis and processing. Through entity recognition and intent analysis, it is possible to gain a deeper understanding of the essence and intent of user questions, thereby generating more accurate and useful answer information.

[0152] (4) Enhancing Learning Effectiveness and Feedback Mechanism: Recording and analyzing users' learning progress and behavior data enables objective evaluation of learning effectiveness, providing users with customized training suggestions, and further improving learning efficiency and outcomes. This application's embodiments not only focus on the user's current learning status but also dynamically adjust training content and methods based on the user's learning progress and feedback, achieving a continuously optimized learning path.

[0153] (5) Support for content expansion and model updates: Responding to user requests for expanded learning content, the system can continuously update and enrich training content, maintaining the timeliness and comprehensiveness of training. Based on the expanded content, the system updates and trains the target large model, ensuring that the model always keeps pace with the latest learning needs, providing users with a superior learning experience.

[0154] In summary, the embodiments of this application demonstrate significant beneficial effects in improving training efficiency, optimizing model performance, enhancing problem-solving capabilities, improving learning outcomes and feedback mechanisms, and supporting content expansion and model updates. They have important practical application value and promising prospects for promotion.

[0155] In the several embodiments provided in this application, it should be understood that the disclosed methods and electronic devices can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components may be combined, or integrated into another system, or some features may be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed may be through some interfaces, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0156] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0157] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0158] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a platform server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0159] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An artificial intelligence-based training method, characterized by, Displaying to-be-learned content through a human-computer interaction interface, the method comprising: In response to a question request for the to-be-learned content, receiving question information and preprocessing the question information to obtain structured data; Analyzing and processing the structured data to obtain target information and sending the target information to a target large model; wherein the target large model is constructed based on a LoRA layer, and the LoRA layer is used to capture changes in model parameters; Obtaining answer information generated by the target large model and displaying the answer information on the human-computer interaction interface.

2. The method of claim 1, wherein, The target large model comprises a dynamic LoRA layer, and the dynamic LoRA layer takes a dynamic low-rank matrix as the core; the target large model is constructed in the following way: Collecting the to-be-learned content and question and answer feedback results data as training data; Adding the dynamic LoRA layer to a base model and setting the number of LoRA layers, the learning rate and the parameter values of the low-rank matrix; Training the base model based on the training data, and in the process of training, reducing the loss value by dynamically adjusting the number of LoRA layers, the learning rate and the parameter values of the low-rank matrix until the base model reaches an optimal state; Taking the base model that has reached the optimal state as the target large model.

3. The method of claim 2, wherein, In the process of training, the loss value is reduced by dynamically adjusting the number of LoRA layers, the learning rate and the parameter values of the low-rank matrix, comprising: Monitoring the change of the loss value in the process of training, if the loss value does not improve in a continuous specified number of training processes, adjusting the number of LoRA layers, the learning rate and the parameter values of the low-rank matrix through a preset adjustment strategy; wherein the adjustment strategy comprises: judging whether the loss value obtained in the last training process is greater than the loss value of this training, if not, performing the next training process, and ending the training if the loss value does not change in a continuous specified number of training processes; if it is satisfied, the number of LoRA layers is increased by 1, the minimum value of the rank of the low-rank matrix is halved, the maximum value of the rank of the low-rank matrix is increased by half, and the learning rate is reduced by half, and then the training is continued to obtain the loss value after training.

4. The method of claim 1, wherein, The question information comprises text data and / or voice data, and the preprocessing of the question information to obtain structured data comprises: Converting the question information to obtain text information; Cleaning the text information to remove irrelevant characters, stop words and noise in the text information; Segmenting the cleaned text information to obtain at least one word unit, and converting the at least one word unit into a text vector, and taking the text vector as the structured data.

5. The method of claim 1, wherein, The analysis and processing of the structured data to obtain target information comprises: Identifying entity information in the structured data through an entity recognition model, and determining an input intent through an intent analysis model; Integrating the entity information and the input intent to obtain the target information.

6. The method of claim 1, wherein, The method further comprises: Obtaining learning progress information for the content to be learned and recording learning behavior data; Determining learning effectiveness based on the learning progress information and the learning behavior data; Providing training suggestion information for the learning effectiveness through the target large model.

7. The method of claim 1, wherein, The method further comprises: Expanding the content to be learned in response to an expansion request for the content to be learned; Updating and training the target large model based on the expanded content to be learned.

8. An artificial intelligence-based training device, characterized by comprising: Displaying the content to be learned through a human-computer interaction interface, the device comprising: A preprocessing module for receiving question information in response to a question request for the content to be learned and preprocessing the question information to obtain structured data; An analysis module for analyzing and processing the structured data to obtain target information and sending the target information to a target large model; wherein the target large model is constructed based on a LoRA layer, and the LoRA layer is used to capture changes in model parameters; An acquisition module for acquiring answer information generated by the target large model and displaying the answer information on the human-computer interaction interface.

9. An electronic device, comprising: Comprising: A processor, a storage medium, and a bus, the storage medium storing machine-readable instructions executable by the processor, the processor and the storage medium communicating through the bus when the electronic device is running, the processor executing the machine-readable instructions to perform the artificial intelligence-based training method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which is executed by the processor to perform the artificial intelligence-based training method of any one of claims 1 to 7.

Images