Electromagnetic large model training method and device, electronic equipment and storage medium

Abstract

The invention belongs to the technical field of large language models, and particularly discloses an electromagnetic large model training method and device, electronic equipment and a storage medium, and the method comprises the steps: extracting electromagnetic signal features of a preset electromagnetic signal related image and text features of a preset electromagnetic related text through a pre-trained encoder; performing comparative learning on the electromagnetic signal features and the text features, so that the feature vector distances of the images and the texts with the same meaning in the shared feature space are closer, and the feature vector distances of the images and the texts with different meanings in the shared feature space are farther; splicing the feature vectors of the images and the texts with the same meaning in the shared feature space to obtain an input sequence; and inputting the input sequence and preset professional knowledge into a to-be-trained large language model for training to obtain a trained large language model. The method can improve the understanding ability of the trained electromagnetic large model.

Description

Technical Field

[0001] This application belongs to the field of large language model technology, and more specifically, relates to an electromagnetic large model training method, apparatus, electronic device and storage medium. Background Technology

[0002] With the successful application of the Transformer model, large language models have become a research hotspot. In 2018, OpenAI released the GPT series of models, which have gained powerful human language generation capabilities through learning from large-scale corpora.

[0003] Real-world data is multimodal, and human cognitive processing typically targets multimodal information. Compared to large language models like ChatGPT, multimodal large models expand input data from text to multiple modalities such as audio, images, and video, significantly improving the model's understanding and generalization abilities, and more closely resembling the cognitive mechanisms of the human brain. Therefore, a growing number of scholars believe that to build more general artificial intelligence that better understands and handles the complexities of the real world, pre-trained large models must inevitably evolve towards multimodality, and multimodal large models will be a crucial trend in the future development of artificial intelligence technology.

[0004] Electromagnetic large models (EMBs) are a type of multimodal large models capable of supporting signal input from multiple sources. They convert input signals into various input images and train them using electromagnetic coding models to form large language models. However, existing training methods for EMBs suffer from the following drawbacks: aligning the multimodal data (signals, text, images) used during training is challenging; discrepancies easily arise between signal features and text semantic matching, and fine-grained features of electromagnetic signals are easily lost during the conversion of electromagnetic signals into features or images; current general-purpose multimodal large models often lack sufficient "signal-language" cross-modal understanding capabilities, resulting in low accuracy in semantic mapping of specialized electromagnetic terms. These shortcomings collectively lead to low understanding capabilities of the trained EMBs. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application aims to provide a method, apparatus, electronic device, and storage medium for training large electromagnetic models. The goal is to solve the problems of low understanding capabilities of trained large electromagnetic models caused by the high difficulty in aligning multimodal data used during training, the easy loss of fine features of electromagnetic signals during the conversion of electromagnetic signals into features or images, and the low accuracy of semantic mapping of specialized electromagnetic terms.

[0006] To achieve the above objectives, in a first aspect, this application provides a method for training a large electromagnetic model, comprising: The electromagnetic signal features of images related to preset electromagnetic signals and the text features of preset electromagnetic-related texts are extracted using a pre-trained encoder. By comparing and learning the electromagnetic signal features and the text features, the feature vectors of images and texts with the same meaning are closer in the shared feature space, while the feature vectors of images and texts with different meanings are farther apart in the shared feature space. The input sequence is obtained by concatenating the feature vectors of images and texts with the same meaning in a shared feature space; The input sequence and preset professional knowledge are input into the large language model to be trained to obtain the trained large language model.

[0007] This application extracts electromagnetic signal features from images related to preset electromagnetic signals and text features from preset electromagnetic-related texts using a pre-trained encoder. This avoids the loss of fine features of electromagnetic signals during the conversion of electromagnetic signals into features or images. By comparing and learning the electromagnetic signal features and the text features, the alignment of multimodal data is facilitated. By inputting the spliced ​​multimodal data and professional knowledge into the large language model to be trained, the large language model can accurately understand various professional terms and ways of thinking, thereby improving the understanding ability of the trained electromagnetic large model.

[0008] According to the electromagnetic large model training method provided in this application, the encoder is a Transformer-based encoder, and the training process of the pre-trained encoder includes: The Transformer-based encoder is trained using pre-acquired unlabeled electromagnetic spectrum signal images to obtain a trained encoder.

[0009] According to the electromagnetic large-scale model training method provided in this application, the step of inputting the input sequence and preset professional knowledge into the large-scale language model to be trained for training, and obtaining the trained large-scale language model, includes: The input sequence, preset professional knowledge, and multimodal instruction data are input into the large language model to be trained for training, and the trained large language model is obtained.

[0010] Secondly, this application provides an electromagnetic large-scale model training device, comprising: The extraction module is used to extract electromagnetic signal features of preset electromagnetic signal related images and text features of preset electromagnetic related text through a pre-trained encoder. The contrastive learning module is used to compare and learn the electromagnetic signal features and the text features, so that the feature vector distance between images and texts with the same meaning in the shared feature space is closer, while the feature vector distance between images and texts with different meanings is farther. The concatenation module is used to concatenate the feature vectors of images and texts with the same meaning in a shared feature space to obtain the input sequence; The training module is used to input the input sequence and preset professional knowledge into the large language model to be trained for training, so as to obtain the trained large language model.

[0011] Thirdly, this application provides an electronic device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the electromagnetic large model training method described in the first aspect or any possible implementation thereof.

[0012] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the electromagnetic large model training method described in the first aspect or any possible implementation thereof.

[0013] Fifthly, this application provides a computer program product that, when run on a processor, causes the processor to execute the electromagnetic large model training method described in the first aspect or any possible implementation of the first aspect.

[0014] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0015] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art: By extracting electromagnetic signal features from images related to preset electromagnetic signals and text features from preset electromagnetic-related text using a pre-trained encoder, the fine details of electromagnetic signals are avoided during the conversion of electromagnetic signals into features or images. Comparative learning of the electromagnetic signal features and the text features facilitates the alignment of multimodal data. The spliced ​​multimodal data and professional knowledge are then input into the large language model to be trained, enabling the large language model to accurately understand various professional terms and ways of thinking, thus improving the comprehension ability of the trained electromagnetic large model. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the electromagnetic large model training method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the electromagnetic large model training process provided in the embodiments of this application; Figure 3 This is a schematic diagram of the encoder training process provided in an embodiment of this application; Figure 4 This is one of the schematic diagrams of the output of the electromagnetic question-and-answer system provided in the embodiments of this application; Figure 5 This is the second schematic diagram of the output of the electromagnetic question-and-answer system provided in the embodiments of this application; Figure 6 This is a schematic diagram of the structure of an electromagnetic large model training device provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] In this article, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The symbol " / " in this article indicates that the related objects are in an "or" relationship; for example, A / B means A or B.

[0020] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0021] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.

[0022] Next, combined Figures 1-5 The electromagnetic large model training method provided in the embodiments of this application is introduced.

[0023] Figure 1 This is a flowchart illustrating the electromagnetic large model training method provided in the embodiments of this application, as shown below. Figure 1 As shown, the method includes the following steps: Step S1: Extract electromagnetic signal features of preset electromagnetic signal related images and text features of preset electromagnetic related texts through a pre-trained encoder. Electromagnetic large models are a type of multimodal large models that can support signal input from multiple sources. They can convert input signals into various input images and train them using electromagnetic coding models to form large language models. These models can provide a variety of functions, including model classification, feature extraction, situation interpretation, decision support, task execution, report generation, and intelligent assistant.

[0024] Optionally, the preset electromagnetic signal related image can be an image converted from the electromagnetic signal, representing DC signals from different sources, such as IQ signal diagrams, time-frequency diagrams, amplitude-phase diagrams, and constellation diagrams.

[0025] Optionally, the preset electromagnetic-related text can be text describing electromagnetic signals, such as noise ratio description, modulation description, etc.

[0026] Optionally, the encoder can be any encoder capable of extracting electromagnetic signal features from electromagnetically related images and text features from electromagnetically related text.

[0027] The encoder encodes the image related to the preset electromagnetic signal into a signal token, that is, electromagnetic signal features, and converts the signal into a feature representation of the same type as the text token, such as discrete sequence units or continuous embedding vectors.

[0028] Step S2 involves comparative learning of electromagnetic signal features and text features to make the feature vectors of images and texts with the same meaning closer in the shared feature space, while the feature vectors of images and texts with different meanings are farther apart in the shared feature space. The encoder converts the image (signal) and text into high-dimensional feature vectors (embedding vectors) and maps them into the same feature space. Here, "distance" refers to the geometric distance between these two feature vectors in that space, such as cosine distance or Euclidean distance. Through comparative learning of the features, if the image and text have the same meaning (e.g., "picture of a cat" and "a cat"), their feature vectors will be optimized to be closer together (higher similarity); if they have different meanings (e.g., "picture of a cat" and "a car"), their feature vectors will be optimized to be farther apart (lower similarity).

[0029] This training method, which involves "shortening / widening the distance," essentially teaches the model the rule that "semantically similar cross-modal content is closer in the feature space," thereby achieving semantic alignment between images and text.

[0030] Step S3: Concatenate the feature vectors of images and texts with the same meaning in the shared feature space to obtain the input sequence; Optionally, the dimensions and feature space of the signal token can be adjusted through an adaptation layer to make its format match the input requirements of a large language model (e.g., consistent with the dimensions and encoding rules of the text token). Then, the format-aligned signal token and the text token (e.g., the token corresponding to the text describing the signal) are concatenated into a unified input sequence.

[0031] Step S4: Input the input sequence and preset professional knowledge into the large language model to be trained for training, and obtain the trained large language model.

[0032] By inputting common knowledge from various professional fields into the large language model, the large model can accurately understand various professional terms and ways of thinking. By inputting the input sequence into the large language model, the large language model can gain the ability to describe signals and understand the semantics of signals.

[0033] Figure 2 This is a schematic diagram of the electromagnetic large model training process provided in the embodiments of this application, such as... Figure 2 As shown, the training process of the electromagnetic large model can be divided into two stages. The first stage is to input common knowledge from various professional fields into the large language model through a text encoder, so that the large language model can accurately understand various professional terms and ways of thinking. The second stage is to map the input sequence to the input domain of the large language model through an adaptation layer, so that the large language model can obtain the ability to describe signals and understand signal semantics.

[0034] The electromagnetic large-scale model training method provided in this application extracts electromagnetic signal features from images related to preset electromagnetic signals and text features from preset electromagnetic-related text through a pre-trained encoder. This avoids the loss of fine features of electromagnetic signals during the conversion of electromagnetic signals into features or images. By comparing and learning the electromagnetic signal features and the text features, the alignment of multimodal data is facilitated. The method trains the large language model by inputting the spliced ​​multimodal data and professional knowledge into the model, enabling the large language model to accurately understand various professional terms and ways of thinking, thereby improving the comprehension ability of the trained electromagnetic large-scale model.

[0035] In some embodiments, the encoder is a Transformer-based encoder, and the training process of the pre-trained encoder includes: The Transformer-based encoder is trained using pre-acquired unlabeled electromagnetic spectrum signal images to obtain a trained encoder.

[0036] This application uses a Transformer-based encoder to extract features, which can efficiently capture global dependencies within a sequence.

[0037] Optionally, the encoder can be trained on large-scale unlabeled electromagnetic spectrum signal image data to learn visual feature representations.

[0038] Alternatively, the encoder can be used to learn features using already classified graph spectra.

[0039] Figure 3 This is a schematic diagram of the encoder training process provided in the embodiments of this application, as shown below. Figure 3 As shown, in one embodiment of this application, a Transformer-based electromagnetic encoder is trained using an electromagnetic spectrum signal graph and text describing the electromagnetic signal to extract signal features and text features.

[0040] In some embodiments, step S4 specifically includes: Input sequences, pre-defined professional knowledge, and multimodal instruction data are input into the large language model to be trained to obtain the trained large language model.

[0041] By introducing complex multimodal instruction data, the large language model is trained to follow various user instructions, further enhancing the model's multimodal understanding and task execution capabilities, enabling the large model to perform customized tasks such as intelligent evaluation, decision support, external knowledge retrieval, and system calls.

[0042] Once the large electromagnetic model is trained, it can be used for three applications: building a signal library, an electromagnetic question-and-answer system, and an electromagnetic knowledge graph.

[0043] The process of building a signal library using a large electromagnetic model includes: first, preprocessing the data by converting IQ signal data into images, and then extracting feature vectors from DC images; then, storing the signal feature vectors into databases such as MongoDB, Faiss, and Milvus according to different categories; finally, generating a search front-end, inputting signal text information, outputting DC signals, inputting IQ signals, processing them into images, and retrieving and outputting DC signals.

[0044] An electromagnetic question-and-answer system is implemented using a large electromagnetic model: First, the user's question is input and converted into a DC signal. Then, combined with the user's question, instructions are generated and input into the multimodal model along with the signal data. Next, the multimodal model is imported to understand the content of the electromagnetic signal and generate DC signal images, including constellation diagrams, signal diagrams, and time-frequency diagrams. Finally, the DC signal images are output and converted into output signal content.

[0045] Figure 4 This is one of the schematic diagrams of the output of the electromagnetic question-and-answer system provided in the embodiments of this application. Figure 5This is the second schematic diagram of the output of the electromagnetic question-and-answer system provided in the embodiments of this application. In one embodiment of this application, an image and the question "What type of signal modulation is this?" are input into the electromagnetic question-and-answer system, and the system outputs... Figure 4 The constellation diagram shown and Figure 5 The image shows an IQ signal diagram, along with the text "Based on the signal data you provided, we can determine that the DC signal uses 16-QAM modulation, a widely used digital modulation method that combines amplitude and phase modulation to improve data transmission rate," "Signal modulation chart: Based on the signal data you provided, the system has generated the corresponding constellation diagram, IQ signal diagram, and time-frequency diagram for you," and an interpretation of each image.

[0046] The process of constructing an electromagnetic knowledge graph using a large electromagnetic model is as follows: 1a. Construct a knowledge graph by integrating and analyzing the transmission signal characteristics of various electronic devices (communication, navigation, etc.); 2a. Utilize knowledge graphs to achieve real-time monitoring and logical reasoning of target activities in order to identify and predict target actions; 3a. Based on electromagnetic signal analysis and knowledge graph reasoning, provide decision-makers with specific action plans and implementation strategies; 4a. Based on the characteristics of the target electronic device, quickly formulate effective action measures.

[0047] In one embodiment of this application, a system with a modular architecture is used to construct a complete closed loop from data processing to decision support through the collaborative work of two core modules: electromagnetic large model training and optimization and multimodal intelligent agent inference. The system adopts a front-end and back-end separation architecture and combines deep learning frameworks and database technology to realize various functions.

[0048] Specifically, the system includes a model training and tuning module and a multimodal agent reasoning module. The model training and tuning module achieves model training and tuning through data annotation tools, FGADet model training, Qwen3 model fine-tuning, and visual monitoring. The multimodal agent reasoning module achieves multimodal agent reasoning through agent configuration, knowledge base management, and interactive reasoning.

[0049] Optionally, the data annotation tool is developed based on web technologies, using Canvas to implement interactive image annotation, supporting functions such as rectangle drawing and category selection. Annotated data is stored in JSON format, containing target coordinates, category information, etc., and can be exported to standard formats such as COCO for model training.

[0050] Optionally, the FGADet model is trained using the PyTorch framework, modifying the network structure based on YOLO11 to enhance local feature extraction capabilities. The training process employs a distributed training strategy, supporting breakpoint resume and model version management. Training parameters (such as iteration count and learning rate) can be configured via a web interface, and key metrics during training are written to the database in real time.

[0051] Optionally, the Qwen3 model fine-tuning is implemented based on the Hugging Face Transformers library, improving the LoRA algorithm. When constructing low-rank matrices A and B, pre-trained Chinese word vector and character vector models are integrated, and feature fusion is achieved through a custom weight allocation mechanism. The fine-tuning process supports incremental pre-training and can continuously absorb new text data.

[0052] Optionally, ECharts is used to visualize the training process, and WebSocket is used for real-time data updates. The output results of the test samples are displayed in the form of image overlay annotations and text, allowing users to intuitively evaluate the model's performance.

[0053] Optionally, the agent configuration provides model selection and combination functions through the front-end interface, and the back-end stores the model parameters and configuration information selected by the user as agent configuration files, and loads the corresponding model according to the configuration during inference.

[0054] Optionally, a graph database can be used to store various types of knowledge, supporting batch import of formats such as Excel and JSON. This enables knowledge retrieval, updating, and correlation analysis, providing knowledge support for the reasoning process.

[0055] Optionally, the front-end adopts a dialog interface design, supporting text input and image upload. After receiving user input, the back-end calls the FGADet model to process the image and extract target information; simultaneously, it inputs the text information and image analysis results into the Qwen3 model, combines the knowledge base content for reasoning, and generates corresponding analysis and action suggestions. The Prompt project is used to optimize the input format and improve the model output quality.

[0056] Optionally, the front-end and back-end communicate via a RESTful API, using JWT for authentication and access control. The database uses MySQL to store system configuration and annotation data, MongoDB to store unstructured text and image metadata, and Redis to cache frequently accessed model parameters and knowledge base content, improving system response speed.

[0057] The electromagnetic large-scale model training device provided in this application is described below. The electromagnetic large-scale model training device described below can be referred to in correspondence with the electromagnetic large-scale model training method described above.

[0058] Figure 6 This is a schematic diagram of the structure of an electromagnetic large model training device provided in an embodiment of this application, as shown below. Figure 6 As shown, the device 600 includes: Extraction module 610 is used to extract electromagnetic signal features of preset electromagnetic signal related images and text features of preset electromagnetic related text through a pre-trained encoder. The contrastive learning module 620 is used to perform contrastive learning on electromagnetic signal features and text features, so that the feature vectors of images and texts with the same meaning are closer in the shared feature space, while the feature vectors of images and texts with different meanings are farther apart in the shared feature space. The splicing module 630 is used to splice the feature vectors of images and texts with the same meaning in a shared feature space to obtain an input sequence; The training module 640 is used to input the input sequence and preset professional knowledge into the large language model to be trained, and to obtain the trained large language model.

[0059] It should be understood that the above-described device is used to execute the methods in the above embodiments. The implementation principle and technical effect of the corresponding program modules in the device are similar to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.

[0060] Based on the methods in the above embodiments, Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown in the illustration, this application provides an electronic device that may include a processor 710, a communications interface 720, a memory 730, and a communication bus 740. The processor 710, communications interface 720, and memory 730 communicate with each other via the communication bus 740. The processor 710 can call logic instructions stored in the memory 730 to execute the electromagnetic large-scale model training method described in the above embodiment.

[0061] Furthermore, the logic instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a 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, server, or network device, etc.) to execute all or part of the steps of the electromagnetic large model training method described in the various embodiments of this application.

[0062] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the electromagnetic large model training method in the above embodiments.

[0063] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the electromagnetic large model training method in the above embodiments.

[0064] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0065] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0066] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0067] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0068] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An electromagnetic large model training method, characterized in that, The method comprises the steps of: extracting electromagnetic signal features of preset electromagnetic signal related images and text features of preset electromagnetic related texts through a pre-trained encoder; performing contrast learning on the electromagnetic signal features and the text features, so that the feature vectors of images and texts with the same meaning in a shared feature space are closer, and the feature vectors of images and texts with different meanings in the shared feature space are farther apart; splicing the feature vectors of images and texts with the same meaning in the shared feature space to obtain an input sequence; inputting the input sequence and preset professional knowledge into a large language model to be trained to obtain a trained large language model.

2. The electromagnetic macro-model training method of claim 1, wherein, The encoder is a Transformer-based encoder, and the training process of the pre-trained encoder comprises: training the Transformer-based encoder based on pre-acquired unlabeled electromagnetic spectrum signal images to obtain a trained encoder. 3.The electromagnetic macro-model training method of claim 1, wherein, The input sequence and preset professional knowledge are inputted into a large language model to be trained to obtain a trained large language model, comprising: inputting the input sequence, preset professional knowledge and multi-modal instruction data into a large language model to be trained to obtain a trained large language model.

4. An electromagnetic large model training apparatus, characterized by, The method comprises the steps of: extracting electromagnetic signal features of preset electromagnetic signal related images and text features of preset electromagnetic related texts through a pre-trained encoder; performing contrast learning on the electromagnetic signal features and the text features, so that the feature vectors of images and texts with the same meaning in a shared feature space are closer, and the feature vectors of images and texts with different meanings in the shared feature space are farther apart; splicing the feature vectors of images and texts with the same meaning in the shared feature space to obtain an input sequence; inputting the input sequence and preset professional knowledge into a large language model to be trained to obtain a trained large language model.

5. An electronic device, comprising: The method comprises the steps of: at least one memory for storing a computer program; at least one processor for executing the program stored in the memory, when the program stored in the memory is executed, the processor is used to execute the electromagnetic large model training method in any one of claims 1-3.

6. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 5. When the computer program runs on the processor, the processor executes the electromagnetic large model training method in any one of claims 1-3.

7. A computer program product, characterised in that, When the computer program product runs on the processor, the processor executes the electromagnetic large model training method in any one of claims 1-3.

Images