Virtual practical training method and device based on large language model, equipment and medium
By dividing the virtual training process into multiple steps and using a large language model to match the virtual training information, the problem that the virtual training platform cannot provide real-time tutoring and supervision is solved, and the training effect is improved.
Patent Information
- Application Number
- CN202410995758.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-08
AI Technical Summary
The existing virtual training platform cannot provide real-time tutoring and supervising students' experimental process, resulting in low training results.
Using a virtual training method based on a large language model, the virtual training process is divided into multiple training steps. Through the AI teaching assistant virtual role model, the virtual training information and knowledge graph are monitored and matched to the virtual training information and knowledge graph, and the tutoring information is provided in real time to ensure that students' operations are in line with each other. Preset training content.
Real-time supervision and guidance of the virtual training process is realized, the training effect is improved, and students are ensured to operate correctly in every step and achieve preset training content.
Smart Images

Figure CN118918744A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of virtual reality technology, and in particular to a virtual training method, device, equipment and medium based on a large language model. Background Art
[0002] Currently, there are online training platforms based on HTML5, WebGL and VR virtual reality technology. Through this platform, students can repeatedly practice some experiments in real life (such as experiments with expensive equipment, potentially dangerous experiments, long cycles or difficult to reproduce processes) on this training platform.
[0003] However, although this type of experiment solves the problem of physical objects, it still cannot solve the problem of practical training guidance. For example, during the practice of each experiment, a teacher cannot check the experimental progress and the degree of knowledge mastery of each student in real time for 30 students doing experiments in a class. In the end, they still have to look at the experimental results and scores, resulting in a lack of guidance and supervision of the experimental process, which reduces the effectiveness of practical training. Summary of the invention
[0004] In view of this, an embodiment of the present invention provides a virtual training method based on a large language model to solve the technical problem in the prior art that the training effect is low due to the lack of guidance and supervision of the experimental process. The method includes:
[0005] Acquire a corresponding virtual training file according to the user's selection, wherein the virtual training process of the virtual training file is divided into multiple different training steps according to the knowledge system;
[0006] Reading a corresponding knowledge graph according to the virtual training file, wherein different virtual training files correspond to respective knowledge graphs, and the knowledge graph is obtained by inputting all knowledge systems of the virtual training file into a large language model according to knowledge information corresponding to the training steps;
[0007] In the process of performing each of the training steps of the virtual training, for each of the training steps, monitoring and acquiring the virtual training information of the training step, wherein the virtual training information includes the operation of the virtual model and the state of the virtual model;
[0008] The AI teaching assistant virtual role model is used to match the virtual training information with the knowledge graph according to the knowledge system corresponding to the training step to determine whether the virtual training information is consistent with the knowledge information of the training step; if not, the virtual training tutoring information of the training step is output through the AI teaching assistant virtual role model.
[0009] The embodiment of the present invention also provides a virtual training device based on a large language model to solve the technical problem in the prior art that the training effect is low due to the lack of guidance and supervision of the experimental process. The device includes:
[0010] A file acquisition module, used to acquire a corresponding virtual training file according to a user's selection, wherein the virtual training process of the virtual training file is divided into a plurality of different training steps according to the knowledge system;
[0011] A graph acquisition module is used to read the corresponding knowledge graph according to the virtual training file, wherein different virtual training files correspond to their own knowledge graphs, and the knowledge graph is obtained by inputting all the knowledge systems of the virtual training file into the large language model according to the knowledge information corresponding to the training steps;
[0012] An information collection module is used to monitor and obtain virtual training information of each training step during the virtual training, wherein the virtual training information includes the operation of the virtual model and the status of the virtual model;
[0013] The training module is used to match the virtual training information with the knowledge graph according to the knowledge system corresponding to the training step through the AI teaching assistant virtual role model, and determine whether the virtual training information conforms to the knowledge information of the training step; if not, output the virtual training tutoring information of the training step through the AI teaching assistant virtual role model.
[0014] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the above-mentioned virtual training methods based on a large language model when executing the computer program, so as to solve the technical problem in the prior art of low training effect due to the lack of guidance and supervision of the experimental process.
[0015] An embodiment of the present invention also provides a computer-readable storage medium, which stores a computer program for executing any of the above-mentioned virtual training methods based on a large language model, so as to solve the technical problem in the prior art that the training effect is low due to the lack of guidance and supervision of the experimental process.
[0016] Compared with the prior art, the beneficial effects that can be achieved by at least one of the above-mentioned technical solutions adopted in the embodiments of this specification include at least: it is proposed that the virtual training process of the virtual training file be divided into multiple different training steps according to the knowledge system, and the corresponding knowledge graphs are learned for different virtual training files through a large language model. In the process of each of the training steps of the virtual training, for each of the training steps, the virtual training information of the training step is monitored and obtained, wherein the virtual training information includes the operation of the virtual model and the status of the virtual model; and then, the virtual training information is matched with the knowledge graph according to the knowledge system corresponding to the training step through the AI assistant virtual role model to determine whether the virtual training information conforms to the knowledge information of the training step; if not, the virtual training tutoring information of the training step is output through the AI assistant virtual role model. During each step of virtual training, the training situation of the training step can be monitored in real time to determine whether the relevant training operations of the training step have reached the preset training content. If not, the virtual training coaching information of the training step can be output in time through the AI teaching assistant virtual role model to achieve real-time guidance or guidance of the training step, so as to improve the training effect. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0018] Figure 1 is a flow chart of a virtual training method based on a large language model provided by an embodiment of the present invention;
[0019] Figure 2 It is a schematic diagram of the principle of implementing a virtual training method based on a large language model provided by an embodiment of the present invention;
[0020] Figure 3 is a structural block diagram of a computer device provided by an embodiment of the present invention;
[0021] Figure 4 It is a structural block diagram of a virtual training device based on a large language model provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0022] The embodiments of the present application are described in detail below with reference to the accompanying drawings.
[0023] The following describes the implementation methods of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The present application can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, in the absence of conflict, the following embodiments and the features in the embodiments can be combined with each other. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work belong to the scope of protection of the present application.
[0024] In an embodiment of the present invention, a virtual training method based on a large language model is provided, such as Figure 1 As shown, the method includes:
[0025] Step S101: acquiring a corresponding virtual training file according to a user selection, wherein the virtual training process of the virtual training file is divided into a plurality of different training steps according to a knowledge system;
[0026] Step S102: reading a corresponding knowledge graph according to the virtual training file, wherein different virtual training files correspond to respective knowledge graphs, and the knowledge graph is obtained by inputting all knowledge systems of the virtual training file into a large language model according to knowledge information corresponding to the training steps;
[0027] Step S103: During each training step of the virtual training, for each training step, monitoring and acquiring virtual training information of the training step, wherein the virtual training information includes the operation of the virtual model and the state of the virtual model;
[0028] Step S104: Match the virtual training information with the knowledge graph through the AI teaching assistant virtual role model according to the knowledge system corresponding to the training step, and determine whether the virtual training information conforms to the knowledge information of the training step; if not, output the virtual training tutoring information of the training step through the AI teaching assistant virtual role model.
[0029] Depend on Figure 1 It can be seen from the process shown that in the embodiment of the present invention, during each training step of the virtual training, the training situation of the training step is monitored in real time to determine whether the relevant training operations of the training step meet the preset training content. If not, the virtual training tutoring information of the training step can be output in time through the AI teaching assistant virtual role model to achieve real-time guidance or guidance of the training step, so as to improve the training effect.
[0030] In the specific implementation, in order to achieve real-time supervision of each training step in the virtual training process so as to provide real-time and targeted guidance, it is proposed that the virtual training process of the virtual training file be divided into multiple different training steps according to the knowledge system. The knowledge system can be different training knowledge points in the virtual training file. For example, "path phenomenon of engine oil entering the cylinder" and "diagnostic criteria for engine oil entering the cylinder" are different knowledge systems for different knowledge points. The knowledge system can be a knowledge system corresponding to each knowledge point. The virtual training process is divided into multiple different training steps according to the knowledge system, that is, the training content corresponding to each knowledge system is divided into at least one training step. In the virtual training file, prefabricated information can also be added to each training step. The prefabricated information is used to indicate which knowledge system the training step corresponds to. The division of the training steps can be determined according to the actual training situation. For example, the operation steps corresponding to each model can be divided into one training step, or each operation action or each use process of each tool model can be divided into one training step.
[0031] In the specific implementation, in order to monitor whether the user's operation in each training step is correct or appropriate, it is proposed to match the virtual training information of each training step with the knowledge graph. In order to improve the accuracy of the knowledge graph, it is proposed to use a large language model to learn the knowledge graph in the following way:
[0032] The training process of the virtual training file is divided into different block contents according to the knowledge system, and each block content is divided into different training steps, each of which corresponds to its own knowledge information, wherein the knowledge information includes knowledge points, model information and correct training operation information; the knowledge information of the training steps of all block contents of the virtual training file is input into the large language model for learning, and the knowledge graph is obtained after the large language model is verified by asking questions about the learning content.
[0033] Specifically, the basic capabilities of the large language model can be used to replace the knowledge graph based on "triples". Through the training of relevant knowledge information, the knowledge graph can be used. The relevant knowledge information is based on a set of self-defined knowledge definition specifications, which classifies, defines, describes, and associates knowledge points, allowing the large language model to "learn" and "understand" these relevant knowledge information.
[0034] The basic capabilities of the large language model are as follows:
[0035] 1) Recognition and understanding: The model learns from large amounts of text data to identify semantic relationships, so that it can perform analytical reasoning based on these patterns when encountering new problems.
[0036] 2) Knowledge accumulation and integration: A wide range of knowledge is accumulated during the training process, and when reasoning is performed, this knowledge can be mobilized and integrated to form reasonable inferences.
[0037] 3) Learning logical rules: Through the understanding and analysis of natural language, some logical relationships and reasoning patterns can be independently learned from a large amount of corpus.
[0038] 4) Multi-step thinking: Ability to conduct continuous analysis and deduction, and draw conclusions by processing information step by step.
[0039] 5) Analogy and association: Make analogies and associations based on past experience and similar situations to assist the reasoning process.
[0040] 6) Contextual analysis: Make full use of the contextual information of the input text and combine it with the model’s existing knowledge and understanding to make inferences and judgments.
[0041] 7) Probabilistic inference: Based on the probability distribution of language and the predictive ability of the model, relatively reasonable inference conclusions are made with a certain degree of reliability.
[0042] In specific implementation, after the large language model is trained and learned, the training effect of the trained large language model can also be verified by the proofreading method of question verification. For example, after the large language model is trained using a general method, questions are asked about the training of the relevant knowledge information of the large language model to verify whether its answers and expressions are accurate and whether its reasoning is rigorous, so as to verify whether the large language model has mastered the relevant knowledge information of training and learning.
[0043] In specific implementation, in order to improve the accuracy and comprehensiveness of the knowledge graph, the knowledge information corresponding to each training step includes knowledge points, model information, and correct training operation information.
[0044] Step 1: Disassemble the spark plug;
[0045] Knowledge point: Determine whether the guide holes on the valve stem and cylinder head are loose;
[0046] Model: valve stem, cylinder head, spark plug, screwdriver, wrench;
[0047] Model status: not operated, operating, being operated, effect conversion in progress;
[0048] Correct result: Use a wrench to remove the cylinder head and unscrew the spark plug with a wrench.
[0049] In the specific implementation, in order to more accurately and efficiently judge whether the operation of each training step is accurate or appropriate, it is proposed that the knowledge information corresponding to each training step for training the knowledge graph includes principle knowledge points and practical training knowledge points. The principle knowledge points include the definition, function and composition of the virtual model, and the practical training knowledge points include the description of the actions, sequence, goals, requirements and specifications of the virtual model during the operation process.
[0050] Furthermore, when the knowledge information corresponding to the training step includes at least two of the principle knowledge points, the knowledge information also includes the association relationship between the principle knowledge points.
[0051] Specifically, principle knowledge points mainly describe the definition, function, composition, etc. of a thing. Practical knowledge points mainly describe the actions, sequence, goals, requirements, specifications, etc. of a thing in the actual operation process.
[0052] The definition of principle knowledge points is as follows:
[0053] Knowledge point {serial number}: description of the knowledge point. A training text can contain multiple knowledge points.
[0054] Relationship: Describe the relationship between two or more knowledge points. If the logic of the knowledge point itself is clear, it can be omitted.
[0055] For example:
[0056] Knowledge point 1: A car engine is a device that provides power for a car. It converts the heat energy generated by fuel combustion into mechanical energy to propel the car forward.
[0057] Knowledge point 2: Automobile engines are classified according to fuel type:
[0058] Gasoline engine: uses gasoline as fuel and has the characteristics of high speed, high power and light weight.
[0059] Diesel engine: It uses diesel and has the advantages of high torque, good fuel economy and high reliability.
[0060] According to the cylinder arrangement:
[0061] In-line: The cylinders are arranged in a straight line.
[0062] V-type: The cylinders are arranged in a V shape.
[0063] W type: can be seen as a combination of two V-type engines.
[0064] Association relationship: Knowledge point 2 is an attribute of knowledge point 1.
[0065] For example, the definition of practical training knowledge points is as follows:
[0066] Knowledge point: Description of knowledge points. A training text can contain multiple knowledge points. The description can use Chinese characters or Arabic numerals as labels to indicate items or steps. The large language model can understand it by itself. No special layout and formatting of text is required.
[0067] For example:
[0068] The practical knowledge points for the analysis, diagnosis and repair of engine oil burning are as follows:
[0069] (I) The path of oil entering the cylinder
[0070] 1) The engine oil enters the cylinder along the valve stem. The valve is located above the cylinder. If the valve stem and the guide hole on the cylinder head are loose, or the valve oil seal fails, when the intake valve is opened, the engine oil will be sucked into the cylinder along the valve stem.
[0071] 2) The engine oil enters the cylinder along the gap between the piston and the cylinder wall. If the piston and cylinder are loose in the horizontal direction, or the piston and piston ring are loose in the vertical direction, the engine oil will enter the cylinder from bottom to top.
[0072] 3) The engine oil is sucked into the cylinder through the intake pipe. The crankcase ventilation device can suck the gas leaking into the crankcase into the cylinder, recover it and then burn it again. In this process, the oil particles splashing in the crankcase are inevitably sucked into the cylinder. Under normal circumstances, the oil particles sucked into the cylinder are extremely limited. However, if the air filter is seriously blocked, the amount of oil sucked into the cylinder will increase, causing the engine to burn oil.
[0073] (II) Diagnostic criteria for oil entering the cylinder
[0074] 1) Determine whether there is oil in the valve stem.
[0075] 2) Determine whether the guide holes on the valve stem and cylinder head are loose.
[0076] 3) Determine whether the valve oil seal has failed.
[0077] 4) Determine whether there is oil in the cylinder.
[0078] 5) Determine whether there is oil in the intake pipe.
[0079] (III) Detailed steps for cleaning the engine cylinder:
[0080] 1) Preparation: Make sure the vehicle is parked safely and the engine is cool. Prepare necessary tools such as wrenches, screwdrivers, cylinder cleaners, etc.
[0081] 2) Disassemble related parts: According to the instructions in the vehicle maintenance manual, disassemble the parts related to the cylinder, such as the intake manifold, exhaust manifold, spark plugs, etc.
[0082] 3) Remove old gaskets, etc.: Carefully clean the old gaskets, glue and other residues at the joint between the cylinder head and the engine body.
[0083] 4) Clean the surface of the cylinder head: Use a clean rag or special cleaning tool to carefully wipe the surface of the cylinder head to remove dust, oil and other impurities.
[0084] 5) Check the cylinder wall: Use light to carefully check whether the cylinder wall has any scratches, wear, carbon deposits, etc.
[0085] 6) Remove carbon deposits: If there is carbon deposit, you can use a suitable carbon deposit cleaner to spray into the cylinder. After soaking for a period of time, use a carbon scraper to carefully scrape off the carbon deposits, but be careful not to damage the cylinder wall.
[0086] 7) Clean the oil and water channels: Clean the oil and water channels connected to the cylinder to ensure they are unobstructed.
[0087] 8) Flushing: Use compressed air or clean water to flush the inside of the cylinder to remove residual detergent and impurities.
[0088] 9) Drying: Let the cylinder dry naturally or use compressed air to blow it dry to ensure that there is no moisture inside the cylinder.
[0089] 10) Check and confirm: Carefully check all parts of the cylinder again to ensure that they are thoroughly cleaned and free of damage.
[0090] 11) Installation: Install the previously disassembled parts in reverse order, ensuring that they are installed in place and that bolts and other fasteners are tightened.
[0091] When implementing it, Figure 2 As shown, after the knowledge graph is obtained through training and learning of the large language model, the knowledge graph can be stored in a vector library so that the AI teaching assistant virtual role model can match the virtual training information of each training step with the knowledge graph.
[0092] When implementing it, Figure 2 As shown, during the virtual training process, for each training step, the virtual training information of the user in each training step can be collected in real time through perception, monitoring and other means, such as the operation of the virtual model and the status of the virtual model, and the virtual training information is transmitted to the AI teaching assistant virtual role model in real time. The virtual training information is matched with the knowledge graph through the AI teaching assistant virtual role model to determine whether the virtual training information conforms to the knowledge information of the training step.
[0093] In the specific implementation, in order to improve the supervision and effective guidance of each training step, it is proposed to output targeted guidance and guidance through the AI teaching assistant virtual role model, for example,
[0094] If the operation of the virtual model in the virtual training information does not conform to the knowledge information of the training step (for example, the operation does not conform to the knowledge point or the wrong operation is performed using the wrong virtual tool, etc.), then according to the knowledge information of the training step, the AI teaching assistant virtual role model outputs guidance information and explanation information related to the operation;
[0095] If the state of the virtual model in the virtual training information does not conform to the knowledge information of the training step (such as the operated virtual model is incorrect or the state of the virtual model is incorrect, etc.), the knowledge points of the training step are output through the AI teaching assistant virtual role model based on the knowledge information of the training step.
[0096] In specific implementation, the above-mentioned AI teaching assistant virtual role model can be a virtual role model of any image, and the AI teaching assistant virtual role model can output virtual training tutoring information through voice, text, action, etc.
[0097] In the specific implementation, in order to improve the accuracy and efficiency of the matching process, it is proposed to match the virtual training information with the knowledge graph through the following steps:
[0098] In the large model vector library, the knowledge graph is divided into different block contents according to the knowledge system, each block content is divided into different training steps, each training step corresponds to its own knowledge information, and an index is constructed for each training step and the corresponding knowledge information;
[0099] When obtaining the virtual training file, the index corresponding to each training step and the corresponding knowledge information of the knowledge graph is obtained from the large model vector library, and each training step and the corresponding knowledge information form a data pair with the corresponding index;
[0100] For each of the block contents, sort the data pairs corresponding to the block contents according to the execution order of the practical training steps included in the block contents;
[0101] Hash the sorted data pairs, and store the hash values of the data pairs locally;
[0102] At least one query request for different models and / or different operations is generated based on the virtual training information, and the virtual training information is matched in parallel with the Hash value of the block content to which the training step belongs locally based on the at least one query request.
[0103] During specific implementation, the inventors of the present application found that since the amount of data stored in the vector library is very large, if no index is performed, the query efficiency is very low and the answer feedback takes a long time.
[0104] However, AI teaching assistant applications based on virtual reality training operations are different from conversational query applications based on text inquiries. The former can operate multiple object models in a virtual 3D space at the same time, generate multiple "queries" (multiple knowledge points are judged concurrently), and need to provide real-time feedback and record whether the relevant operations are correct, so that the AI teaching assistant can give users timely reminders and correct guidance. The latter is a text-based "query and feedback", and there is only one query at the same time.
[0105] Therefore, the former cannot adopt the same indexing mechanism as the text vector and needs to be improved, that is, the above-mentioned implementation method of matching the virtual training information with the knowledge graph is proposed.
[0106] Specifically, the process of matching the virtual training information with the knowledge graph may include the following steps:
[0107] Step 1: Global Index
[0108] First, the R-Tree algorithm is used to index the vector, and the knowledge graph is divided into multiple hierarchical structures (i.e., block contents) according to the knowledge system. The advantage is that the query based on a certain block content is a nearest neighbor search, which is very efficient.
[0109] Step 2: Association
[0110] When a training courseware is loaded for the first time, it is necessary to determine whether it has been initialized or whether there is an update mark. If not, the initialization step is performed. The initialization step first submits all the training steps, knowledge points, models, states, correct training operation information and other information of the training courseware to the vector library, and the vector library returns the corresponding index data.
[0111] Step 3: Rearrange
[0112] The training steps, knowledge points and other information of the courseware are combined with the returned index data to form data pairs, and the data pairs are arranged in sequence according to the operation steps of the courseware. The courseware knowledge point information of each step corresponds to the index information.
[0113] Step 4: Hashing
[0114] The corresponding information of the rearranged training steps is hashed and indexed, and the two contents (ie, data pairs) are represented by hash values, and stored locally in the courseware.
[0115] Step 5: Load
[0116] The current training courseware loads the hashed index into the memory. The AI assistant monitoring mechanism is activated to obtain the user's virtual training information at each training step, and traverses and compares the knowledge system prefabricated in the courseware (that is, the knowledge system to which the training step belongs) with the hashed index in the memory, so as to efficiently and in real time determine whether the virtual training information is correct.
[0117] In the specific implementation, in order to further improve the knowledge graph and meet more practical training needs, a method for updating the knowledge graph is proposed, for example,
[0118] When the virtual training information does not exist in the knowledge graph, temporarily storing the virtual training information;
[0119] After a preset time, each piece of temporarily stored virtual training information is checked, calibrated and completed;
[0120] The verified, calibrated and completed virtual training information is input into the large language model as training data to obtain an updated knowledge graph.
[0121] In specific implementation, a sub-mechanism for communicating with the large language model can be established in the feedback mechanism. The mechanism is used to automatically feed back the feedback data set to the large language model through the system, thereby driving the large language model to perform iterative learning to update its knowledge graph. The process may include the following steps:
[0122] Step 1: Data formatting
[0123] Format the user operation feedback data (i.e. virtual training information), including the following:
[0124] {Step name, step ID, knowledge point, current operation model, current operation model status}
[0125] Step 2: Temporary data storage
[0126] Multiple groups of virtual training information of multiple users are temporarily stored in the feedback mechanism.
[0127] Step 3: Manual calibration and completion.
[0128] Through the UI interface of the feedback mechanism, each piece of data is manually checked and calibrated. If it is wrong, it will be deleted. If it is incomplete, it will be manually supplemented and marked as "updated".
[0129] Step 4: Establish a large language model connection
[0130] The feedback mechanism is connected to the large language model and can be used for communication and data transmission to transmit temporarily stored virtual training information to the large language model.
[0131] Step 5: Large language model learning again
[0132] The feedback mechanism uses an interface call to submit multiple copies of data to the large language model in batch processing. This allows the large language model to learn again and output an updated knowledge graph. This mechanism facilitates the formation of a virtuous cycle, and the system is increasingly improved through the continuous use of discovered problem data.
[0133] In this embodiment, a computer device is provided, such as Figure 3 As shown, it includes a memory 301, a processor 302, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, any of the above-mentioned virtual training methods based on a large language model is implemented.
[0134] Specifically, the computer device may be a computer terminal, a server or a similar computing device.
[0135] In this embodiment, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program for executing any of the above-mentioned virtual training methods based on a large language model.
[0136] Specifically, computer-readable storage media include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer-readable 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 technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable storage media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.
[0137] Based on the same inventive concept, a virtual training device based on a large language model is also provided in the embodiments of the present invention, as described in the following embodiments. Since the principle of solving the problem by the virtual training device based on a large language model is similar to that of the virtual training method based on a large language model, the implementation of the virtual training device based on a large language model can refer to the implementation of the virtual training method based on a large language model, and the repeated parts will not be repeated. As used below, the term "unit" or "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived.
[0138] Figure 4 is a structural block diagram of a virtual training device based on a large language model according to an embodiment of the present invention, such as Figure 4 As shown, the device comprises:
[0139] The file acquisition module 401 is used to acquire the corresponding virtual training file according to the user's selection, wherein the virtual training process of the virtual training file is divided into multiple different training steps according to the knowledge system;
[0140] A graph acquisition module 402 is used to read the corresponding knowledge graph according to the virtual training file, wherein different virtual training files correspond to their own knowledge graphs, and the knowledge graph is obtained by inputting all the knowledge systems of the virtual training file into the large language model according to the knowledge information corresponding to the training steps;
[0141] The information collection module 403 is used to monitor and obtain virtual training information of each training step during the virtual training, wherein the virtual training information includes the operation of the virtual model and the status of the virtual model;
[0142] The training module 404 is used to match the virtual training information with the knowledge graph through the AI teaching assistant virtual role model according to the knowledge system corresponding to the training step, and determine whether the virtual training information is consistent with the knowledge information of the training step; if not, output the virtual training tutoring information of the training step through the AI teaching assistant virtual role model.
[0143] In one embodiment, the graph acquisition module is also used to divide the training process of the virtual training file into different block contents according to the knowledge system, and divide each block content into different training steps, each training step corresponds to its own knowledge information, wherein the knowledge information includes knowledge points, model information and correct training operation information; the knowledge information of the training steps of all block contents of the virtual training file is input into the large language model for learning, and the knowledge graph is obtained after the large language model is verified by asking questions about the learning content.
[0144] In one embodiment, the knowledge information includes knowledge points including principle knowledge points and practical training knowledge points. The principle knowledge points include a description of the definition, function and composition of the virtual model, and the practical training knowledge points include a description of the actions, sequence, goals, requirements and specifications of the virtual model during operation.
[0145] In one embodiment, when the knowledge information corresponding to the training step includes at least two of the principle knowledge points, the knowledge information also includes the association relationship between the principle knowledge points.
[0146] In one embodiment, the training module is used to output operation-related guidance information and explanation information through the AI teaching assistant virtual role model according to the knowledge information of the training step if the operation status of the virtual model in the virtual training information does not conform to the knowledge information of the training step; if the state of the virtual model in the virtual training information does not conform to the knowledge information of the training step, then the knowledge points of the training step are output through the AI teaching assistant virtual role model according to the knowledge information of the training step.
[0147] In one embodiment, a training module is used to divide the knowledge graph into different block contents according to the knowledge system in a large model vector library, divide each block content into different training steps, each training step corresponds to its own knowledge information, and construct an index for each training step and the corresponding knowledge information; when obtaining the virtual training file, obtain the index corresponding to each training step and the corresponding knowledge information of the knowledge graph from the large model vector library, and each training step and the corresponding knowledge information form a data pair with the corresponding index; for each block content, sort the data pairs corresponding to the block content according to the execution order of the training steps included in the block content; hash the sorted data pairs, and store the hash values of the data pairs locally; generate at least one query request for different models and / or different operations based on the virtual training information, and match the virtual training information with the hash values of the block content to which the training step belongs in parallel locally based on the at least one query request.
[0148] In one embodiment, the above device further comprises:
[0149] An updating module is used to temporarily store the virtual training information when the virtual training information does not exist in the knowledge graph; after a preset period of time, check, calibrate and complete each temporarily stored virtual training information; input the checked, calibrated and completed virtual training information into the large language model as training data to obtain an updated knowledge graph.
[0150] The embodiments of the present invention achieve the following technical effects: during each training step of virtual training, the training situation of the training step is monitored in real time, and it is determined whether the relevant training operations of the training step reach the preset training content. If not, the virtual training tutoring information of the training step can be output in time through the AI teaching assistant virtual role model, so as to achieve real-time guidance or guidance of the training step, so as to improve the training effect.
[0151] Obviously, those skilled in the art should understand that the modules or steps of the above-mentioned embodiments of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. In this way, the embodiments of the present invention are not limited to any specific combination of hardware and software.
[0152] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A virtual training method based on a large language model, characterized in that: include: Acquire a corresponding virtual training file according to the user's selection, wherein the virtual training process of the virtual training file is divided into multiple different training steps according to the knowledge system; Reading a corresponding knowledge graph according to the virtual training file, wherein different virtual training files correspond to respective knowledge graphs, and the knowledge graph is obtained by inputting all knowledge systems of the virtual training file into a large language model according to knowledge information corresponding to the training steps; In the process of performing each of the training steps of the virtual training, for each of the training steps, monitoring and acquiring the virtual training information of the training step, wherein the virtual training information includes the operation of the virtual model and the state of the virtual model; The AI teaching assistant virtual role model is used to match the virtual training information with the knowledge graph according to the knowledge system corresponding to the training step to determine whether the virtual training information is consistent with the knowledge information of the training step; if not, the virtual training tutoring information of the training step is output through the AI teaching assistant virtual role model.
2. The virtual training method based on a large language model as claimed in claim 1, characterized in that: All knowledge systems of the virtual training file are input into the large language model according to the knowledge information corresponding to the training steps to obtain a knowledge graph, including: Divide the training process of the virtual training file into different block contents according to the knowledge system, divide each block content into different training steps, and each training step corresponds to respective knowledge information, wherein the knowledge information includes knowledge points, model information and correct training operation information; The knowledge information of the training steps of all block contents of the virtual training file is input into the large language model for learning, and the knowledge graph is obtained after the large language model is verified by asking questions about the learning content.
3. The virtual training method based on a large language model as claimed in claim 2, characterized in that: The knowledge points included in the knowledge information include principle knowledge points and practical training knowledge points. The principle knowledge points include a description of the definition, function and composition of the virtual model, and the practical training knowledge points include a description of the actions, sequence, goals, requirements and specifications of the virtual model during operation.
4. The virtual training method based on a large language model as claimed in claim 3, characterized in that: When the knowledge information corresponding to the training step includes at least two of the principle knowledge points, the knowledge information also includes the association relationship between the principle knowledge points.
5. The virtual training method based on a large language model as claimed in claim 1, characterized in that: Outputting virtual training coaching information of the training step through the AI teaching assistant virtual role model includes: If the operation of the virtual model in the virtual training information does not conform to the knowledge information of the training step, then according to the knowledge information of the training step, the AI teaching assistant virtual role model outputs guidance information and explanation information related to the operation; If the state of the virtual model in the virtual training information does not conform to the knowledge information of the training step, the knowledge points of the training step are output through the AI teaching assistant virtual role model according to the knowledge information of the training step.
6. The virtual training method based on a large language model according to any one of claims 1 to 5, characterized in that: Matching the virtual training information with the knowledge graph includes: In the large model vector library, the knowledge graph is divided into different block contents according to the knowledge system, each block content is divided into different training steps, each training step corresponds to its own knowledge information, and an index is constructed for each training step and the corresponding knowledge information; When obtaining the virtual training file, the index corresponding to each training step and the corresponding knowledge information of the knowledge graph is obtained from the large model vector library, and each training step and the corresponding knowledge information form a data pair with the corresponding index; For each of the block contents, sort the data pairs corresponding to the block contents according to the execution order of the practical training steps included in the block contents; Hash the sorted data pairs, and store the hash values of the data pairs locally; At least one query request for different models and / or different operations is generated based on the virtual training information, and the virtual training information is matched in parallel with the Hash value of the block content to which the training step belongs locally based on the at least one query request.
7. The virtual training method based on a large language model according to any one of claims 1 to 5, characterized in that: Also includes: When the virtual training information does not exist in the knowledge graph, temporarily storing the virtual training information; After a preset time, each piece of temporarily stored virtual training information is checked, calibrated and completed; The verified, calibrated and completed virtual training information is input into the large language model as training data to obtain an updated knowledge graph.
8. A virtual training device based on a large language model, characterized in that: include: A file acquisition module, used to acquire a corresponding virtual training file according to a user's selection, wherein the virtual training process of the virtual training file is divided into a plurality of different training steps according to the knowledge system; A graph acquisition module is used to read the corresponding knowledge graph according to the virtual training file, wherein different virtual training files correspond to their own knowledge graphs, and the knowledge graph is obtained by inputting all the knowledge systems of the virtual training file into the large language model according to the knowledge information corresponding to the training steps; An information collection module is used to monitor and obtain virtual training information of each training step during the virtual training, wherein the virtual training information includes the operation of the virtual model and the status of the virtual model; The training module is used to match the virtual training information with the knowledge graph according to the knowledge system corresponding to the training step through the AI teaching assistant virtual role model, and determine whether the virtual training information conforms to the knowledge information of the training step; if not, output the virtual training tutoring information of the training step through the AI teaching assistant virtual role model.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the virtual training method based on a large language model according to any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program for executing the virtual training method based on a large language model according to any one of claims 1 to 7.
Citation Information
Patent Citations
First-aid training device and method based on virtual reality equipment
CN114185435A
Pantovascular interventional operation training auxiliary system and construction and use method thereof
CN116913466A
Knowledge graph construction method and system based on large language model
CN117150050A
Intelligent teaching assisting method and system based on MR equipment
CN117689508A
Intelligent interactive training system based on virtual reality technology
CN117765783A
