A method, device, storage medium and electronic device for deploying satellite models

Abstract

This specification discloses a satellite model deployment method, device, storage medium and electronic device. In the satellite model deployment method provided in this specification, a pre-trained base large model and a satellite base model are obtained, and the satellite base model is transmitted to the target satellite; the target mission field for the target satellite model to be deployed is determined; the base large model is trained for the target mission field to obtain an expert large model for the target mission field; the satellite base model is trained according to the expert large model to obtain an incremental fine-tuning model for the target mission field; the incremental fine-tuning model is transmitted to the target satellite, so that the target satellite fuses the satellite base model and the incremental fine-tuning model to obtain the target satellite model.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a satellite-borne model deployment method, device, storage medium and electronic device. Background Art

[0002] In recent years, satellite technology has developed rapidly, satellite application areas have continued to increase, functions have continued to enrich, and satellite applications have been integrated into people's daily lives. Due to the high cost of satellite payload launches and the limited computing power of the payload, the intelligence level of satellites needs to be improved. At the same time, with the development of artificial intelligence technology, large models represented by generative artificial intelligence have shown excellent results in various fields and have been applied to various fields. Deploying large models on satellites will help improve the intelligence level of satellites.

[0003] However, on the one hand, large models usually contain tens or hundreds of billions of parameters, and the storage space reaches tens or even hundreds of GB. Running large models requires high computing resources, and it is difficult for satellites to directly deploy and apply large models with limited computing resources. On the other hand, the communication window between satellites and the ground is very short, often only a few minutes or tens of minutes, and it is impossible to connect for a long time. At the same time, the data transmission and communication bandwidth between satellites and the ground is also limited, making it difficult to transfer large models to satellites in a short time.

[0004] Therefore, how to simply and efficiently deploy a spaceborne model with the same performance as a large model is an urgent problem to be solved. Summary of the invention

[0005] The present specification provides a satellite-based model deployment method, device, storage medium and electronic device to at least partially solve the above-mentioned problems existing in the prior art.

[0006] This manual adopts the following technical solutions:

[0007] This specification provides a method for deploying a satellite model, including:

[0008] Acquire a pre-trained large base model and a satellite-borne base model, and transmit the satellite-borne base model to the target satellite;

[0009] Determine the target mission area for the target spaceborne model to be deployed;

[0010] Training the base large model for the target task field to obtain an expert large model for the target task field;

[0011] The onboard base model is trained according to the expert large model to obtain an incremental fine-tuning model for the target mission field;

[0012] The incremental fine-tuning model is transmitted to the target satellite, so that the target satellite fuses the satellite-borne base model and the incremental fine-tuning model to obtain a target satellite-borne model.

[0013] Optionally, the large base model and the satellite-borne base model are models oriented to general fields, and the large base model contains a larger number of parameters than the satellite-borne base model.

[0014] Optionally, training the base large model for the task domain specifically includes:

[0015] Obtaining training samples and training annotations for the task domain;

[0016] Inputting the training sample into the base large model to obtain an output result to be optimized output by the base large model;

[0017] According to the difference between the output result to be optimized and the training annotation, the parameters of the base large model are adjusted.

[0018] Optionally, training the satellite-borne base model according to the expert large model specifically includes:

[0019] The knowledge of the expert large model is transferred to the onboard base model through knowledge distillation.

[0020] Optionally, the method further comprises:

[0021] For each parameter included in the target satellite-borne model, the number of decimal places of the parameter is quantized to a specified number of places to obtain a lightweight target satellite-borne model.

[0022] Optionally, the method further comprises:

[0023] In response to the target satellite needing to update the target onboard model, the base large model is retrained for the target mission field to obtain an updated expert large model for the target mission field;

[0024] The onboard base model is trained according to the updated expert large model to obtain an updated incremental fine-tuning model for the target mission domain;

[0025] The updated incremental fine-tuning model is transmitted to the target satellite, so that the target satellite fuses the onboard base model and the updated incremental fine-tuning model to obtain an updated onboard model.

[0026] Optionally, after obtaining the updated onboard model, the method includes:

[0027] The target satellite tests whether the updated onboard model can operate normally;

[0028] In response to the updated onboard model being able to operate normally, the target satellite adopts the updated onboard model to replace the target onboard model;

[0029] In response to the update onboard model failing to operate normally, the target satellite returns an error prompt.

[0030] This specification provides a satellite-borne model deployment device, the device comprising:

[0031] An acquisition module is used to acquire a pre-trained large base model and a satellite-borne base model, and transmit the satellite-borne base model to a target satellite;

[0032] A determination module, used to determine the target mission area for the target onboard model to be deployed;

[0033] A training module, used for training the base large model for the target task field to obtain an expert large model for the target task field;

[0034] A fine-tuning module, used for training the onboard base model according to the expert large model to obtain an incremental fine-tuning model for the target mission field;

[0035] The transmission module is used to transmit the incremental fine-tuning model to the target satellite, so that the target satellite fuses the satellite-borne base model and the incremental fine-tuning model to obtain a target satellite-borne model.

[0036] This specification provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned satellite model deployment method is implemented.

[0037] This specification provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned satellite-borne model deployment method when executing the program.

[0038] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:

[0039] In the satellite-borne model deployment method provided in the present specification, a pre-trained base large model and a satellite-borne base model are obtained, and the satellite-borne base model is transmitted to the target satellite; the target mission field faced by the target satellite-borne model to be deployed is determined; the base large model is trained for the target mission field to obtain an expert large model facing the target mission field; the satellite-borne base model is trained according to the expert large model to obtain an incremental fine-tuning model facing the target mission field; the incremental fine-tuning model is transmitted to the target satellite, so that the target satellite fuses the satellite-borne base model and the incremental fine-tuning model to obtain the target satellite-borne model.

[0040] When the onboard model deployment method provided in this specification is used to deploy the target onboard model required by the target satellite, the onboard base model can be first transmitted to the target satellite, and the target mission field for the target onboard model can be determined; then, the expert large model in the target mission field is obtained based on the base large model training, and the knowledge of the expert large model is transmitted to the onboard base model to obtain the incremental fine-tuning model; finally, the incremental fine-tuning model is transmitted to the target satellite, so that the target satellite integrates the onboard base model and the incremental fine-tuning model to obtain the target onboard model. This method can deploy an onboard model with performance similar to that of a large model on the satellite while satisfying the storage resource conditions of the satellite, effectively controlling the volume of the onboard model while greatly improving the performance of the onboard model, and making the subsequent update of the onboard model easier. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings described herein are used to provide a further understanding of this specification and constitute a part of this specification. The illustrative embodiments and descriptions of this specification are used to explain this specification and do not constitute an improper limitation on this specification. In the drawings:

[0042] Figure 1 A schematic diagram of a flow chart of a satellite-borne model deployment method in this specification;

[0043] Figure 2 A schematic diagram of a satellite-borne model deployment device provided in this specification;

[0044] Figure 3 The corresponding Figure 1 Schematic diagram of electronic equipment. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of this specification more clear, the technical solutions of this specification will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0046] The technical solutions provided by the embodiments of this specification are described in detail below in conjunction with the accompanying drawings.

[0047] Figure 1 The following is a flow chart of a method for deploying a satellite-borne model in this specification, which specifically includes the following steps:

[0048] S100: Acquire a pre-trained large base model and a satellite-borne base model, and transmit the satellite-borne base model to a target satellite.

[0049] All steps in the onboard model deployment method provided in this specification can be implemented by any electronic device with computing functions, such as a terminal, a server, etc. For ease of understanding, this specification mainly uses the example of a server in a ground station docked with a satellite as the execution subject.

[0050] This method is mainly used to deploy satellite-borne models. Based on this, a pre-trained base large model and a satellite-borne base model can be first obtained in this step, so that a model that can specialize in a specific field can be obtained based on the model base in the subsequent steps.

[0051] Large models are deep learning models with tens of millions or even hundreds of millions of parameters. The model size can reach hundreds of GB or even larger, with strong expression and learning capabilities. Spaceborne models are neural network models carried on satellites. Base models refer to a trained model that can serve as the basis for other models, and can be retrained and fine-tuned on its basis to adapt to new data and tasks in different scenarios.

[0052] Limited by the satellite's ability to carry payloads, the number of parameters and model size contained in the satellite model are usually much smaller than the large model. This is also the case in this application. The number of parameters and model size contained in the large base model are larger than the satellite-borne base model. At the same time, as base models, the large base model and the satellite-borne base model are both models for general fields and can be used as the framework of other models in the future.

[0053] At the same time, in this step, the smaller onboard base model can be transmitted to the satellite first, so that the satellite stores the onboard base model in this step for use in subsequent steps.

[0054] S102: Determine the target mission area for the target satellite-borne model to be deployed.

[0055] After the base large model and the satellite-borne base model are acquired in step S100, it is necessary to first determine the target mission domain for the target satellite-borne model to be deployed before starting to train the base model.

[0056] Based on the established tasks that the satellite needs to perform when it is launched into orbit, the satellite usually carries multiple corresponding onboard models. Common tasks on satellites include onboard task scheduling, onboard intelligent question and answer, target tracking, environmental monitoring, resource exploration, geographic mapping, etc. Each different task can be completed by one or more onboard models that can achieve the corresponding functions. However, in general, a onboard model cannot usually be used to perform multiple different tasks. Therefore, before obtaining the onboard model, it is first necessary to determine the functions required by the target onboard model for this deployment, that is, the target mission area it is facing.

[0057] S104: Train the base large model for the target task domain to obtain an expert large model for the target task domain.

[0058] After determining the target mission domain for the target satellite-borne model, an expert large model for the target mission domain can be trained based on the base large model in this step. In this method, Base-LLM is used to represent the base large model. The base large model can be trained by itself using data for general fields, or it can directly use a relatively mature public large model, such as LLama3-70B, Qwen2-72B and other large models. This specification does not make specific restrictions on this.

[0059] When training the base large model, for example, supervised learning and other methods may be used. Specifically, training samples and training annotations for the task domain may be obtained; the training samples are input into the base large model to obtain output results to be optimized output by the base large model; and the parameters of the base large model are adjusted according to the difference between the output results to be optimized and the training annotations.

[0060] In this method, Base-LLM-Adapter is used to represent the expert large model. When adjusting the parameters of the base large model, LoRA and other methods can be used for fine-tuning, and finally an expert large model for the target task field is obtained.

[0061] S106: Training the satellite-borne base model according to the expert large model to obtain an incremental fine-tuning model for the target mission field.

[0062] After the expert large model in the target task domain is trained in step S104, the expert large model can be further used in this step to train an incremental fine-tuning model with a smaller model size based on the onboard base model. It can be imagined that the incremental fine-tuning model trained using the expert large model is also oriented to the target task domain.

[0063] In this method, the satellite-based model can be expressed as Sat-Base-LLM, and the incremental fine-tuning model can be expressed as Sat-Base-LLM-Adapter. Similar to the large base model, the satellite-based model can be trained by using data for general fields, or it can directly use relatively mature models, such as LLama3-8B, Qwen2-7B, etc. It only needs to ensure that the parameters of the satellite-based model meet the requirements of the graphics processing unit (GPU) and storage resources on the satellite to support its deployment and operation.

[0064] When training the onboard base model, methods such as knowledge distillation can be used, that is, the knowledge of the expert large model is transferred to the onboard base model through knowledge distillation to obtain an incremental fine-tuning model. Knowledge distillation is a model training technology that transfers the knowledge of the large model to the small model. The small model serves as the student model and the large model serves as the teacher model. The student model uses the results of the teacher model's reasoning to learn knowledge from the teacher model. After the training is completed, the student model has knowledge close to that of the teacher model. The prediction accuracy of the student model is close to that of the teacher model, but the number of parameters is much smaller than that of the teacher model, which greatly reduces the size of the model and saves the operating cost of the model.

[0065] Specifically, the base model Base-LLM and its incremental fine-tuning model Base-LLM-Adapter are used as teacher models, and the satellite base model Sat-Base-LLM and its incremental fine-tuning model Sat-Base-LLM-Adapter are used as student models. The teacher model is used to perform reasoning on the data set to generate reasoning results. The data set and the corresponding reasoning results are merged into a new data set to train the student model, and the trained incremental fine-tuning model Sat-Base-LLM-Adapter can be obtained. During the training process, the objective function is to minimize the KL divergence between the teacher model and the student model to ensure that the student model or the teacher model knowledge produces prediction results close to the teacher model.

[0066] At the same time, in this method, when the incremental fine-tuning model is obtained by knowledge distillation, the satellite base model itself still exists for subsequent model updates. Since the incremental fine-tuning model is very small, the transmission cost from the ground to the satellite is greatly reduced, and the incremental update of the model is realized.

[0067] S108: Transmitting the incremental fine-tuning model to the target satellite, so that the target satellite fuses the satellite-borne base model and the incremental fine-tuning model to obtain a target satellite-borne model.

[0068] After obtaining the incremental fine-tuning model based on the satellite-borne base model through the above series of steps of the method, the incremental fine-tuning model can be finally transmitted to the target satellite in this step. Thus, the target satellite can fuse the satellite-borne base model obtained in step S100 and the incremental fine-tuning model obtained in this step, and finally obtain the target satellite-borne model.

[0069] In this method, Sat-LLM is used to represent the fused target satellite model. The satellite base model Sat-Base-LLM with general basic knowledge and the incremental fine-tuning model Sat-Base-LLM-Adapter with professional knowledge in the target mission field are fused. The capability of the satellite base model can complement the incremental fine-tuning model that lacks basic knowledge, so that the target satellite model has complete reasoning capabilities. Since the incremental fine-tuning model itself is obtained based on the satellite base model, the fusion process of the two is not complicated, and it only needs to merge some key codes to complete it.

[0070] Additionally, after the satellite completes the fusion of the onboard base model and the incremental fine-tuning model to obtain the target onboard model, the target onboard model can be further lightweighted to further reduce the storage resources occupied by the target onboard model in the satellite, reduce the model size, and improve the model reasoning speed. Specifically, for each parameter included in the target onboard model, the decimal place of the parameter can be quantized to a specified place to obtain a lightweight target onboard model.

[0071] When performing lightweighting, low-bit quantization techniques such as INT8 and INT4 can be used to quantize the parameters with long decimal places in the target satellite model into parameters with fewer decimal places. Among them, the specified number of bits can be determined according to the specific performance requirements and the resource status of the satellite. For example, it can be set to 2 bits, 4 bits, etc., and this manual does not impose specific restrictions on this. Those skilled in the art should understand that the longer the number of bits of a data, the larger the storage space it occupies in the computer. Low-bit quantization can convert data with longer bits into data with shorter bits, greatly reducing the size of the model with almost no loss in model performance.

[0072] Additionally, when the target satellite-borne model is deployed using this method, a simple and efficient update can be achieved when the target satellite-borne model needs to be updated later. Specifically, in response to the target satellite needing to update the target satellite-borne model, the base large model can be retrained for the target mission field to obtain an updated expert large model for the target mission field; the satellite-borne base model can be trained according to the updated expert large model to obtain an updated incremental fine-tuning model for the target mission field; the updated incremental fine-tuning model can be transmitted to the target satellite, so that the target satellite fuses the satellite-borne base model and the updated incremental fine-tuning model to obtain an updated satellite-borne model.

[0073] The base large model and the satellite-borne base model are always stored in the server and do not need to be re-acquired. Based on this, when the target satellite-borne model needs to be updated, the base large model can be directly trained with the newly acquired training samples in the target mission field to obtain the updated expert large model in the target mission field. Subsequently, the knowledge of the updated expert large model is transmitted to the satellite-borne base model to obtain the updated incremental fine-tuning model in the target mission field. Finally, the updated incremental fine-tuning model is transmitted to the target satellite again, so that the target satellite fuses the satellite-borne base model and the updated incremental fine-tuning model to obtain the updated satellite-borne model.

[0074] Furthermore, after obtaining the updated satellite-borne model, before formally putting it into use, the updated satellite-borne model can be first tested to see whether it can operate normally. Specifically, the target satellite tests whether the updated satellite-borne model can operate normally; in response to the updated satellite-borne model being able to operate normally, the target satellite adopts the updated satellite-borne model to replace the target satellite-borne model; in response to the updated satellite-borne model not being able to operate normally, the target satellite returns an error prompt.

[0075] When the updated onboard model can operate normally, the updated onboard model can be used to replace the current target onboard model. On the contrary, if the updated onboard model is abnormal, the satellite can return an error prompt to the ground station and continue to use the current target onboard model.

[0076] When the onboard model deployment method provided in this specification is used to deploy the target onboard model required by the target satellite, the onboard base model can be first transmitted to the target satellite, and the target mission field for the target onboard model can be determined; then, the expert large model in the target mission field is obtained based on the base large model training, and the knowledge of the expert large model is transmitted to the onboard base model to obtain the incremental fine-tuning model; finally, the incremental fine-tuning model is transmitted to the target satellite, so that the target satellite integrates the onboard base model and the incremental fine-tuning model to obtain the target onboard model. This method can deploy an onboard model with performance similar to that of a large model on the satellite while satisfying the storage resource conditions of the satellite, effectively controlling the volume of the onboard model while greatly improving the performance of the onboard model, and making the subsequent update of the onboard model easier.

[0077] The above is the onboard model deployment method provided in this specification. Based on the same idea, this specification also provides a corresponding onboard model deployment device, such as Figure 2 shown.

[0078] Figure 2 A schematic diagram of a satellite-borne model deployment device provided in this specification specifically includes:

[0079] An acquisition module 200 is used to acquire a pre-trained large base model and a satellite-borne base model, and transmit the satellite-borne base model to a target satellite;

[0080] A determination module 202 is used to determine the target mission area for the target onboard model to be deployed;

[0081] A training module 204 is used to train the base large model for the target task field to obtain an expert large model for the target task field;

[0082] A fine-tuning module 206 is used to train the onboard base model according to the expert large model to obtain an incremental fine-tuning model for the target mission field;

[0083] The transmission module 208 is used to transmit the incremental fine-tuning model to the target satellite, so that the target satellite fuses the satellite-borne base model and the incremental fine-tuning model to obtain a target satellite-borne model.

[0084] Optionally, the large base model and the satellite-borne base model are models oriented to general fields, and the large base model contains a larger number of parameters than the satellite-borne base model.

[0085] Optionally, the training module 204 is specifically used to obtain training samples and training annotations for the task field; input the training samples into the base large model to obtain the output results to be optimized output by the base large model; and adjust the parameters of the base large model according to the difference between the output results to be optimized and the training annotations.

[0086] Optionally, the fine-tuning module 206 is specifically configured to transfer the knowledge of the expert large model to the onboard base model through knowledge distillation.

[0087] Optionally, the device further includes a quantization module 210, which is specifically configured to quantize the number of decimal places of each parameter included in the target satellite-borne model to a specified number of places, so as to obtain a lightweight target satellite-borne model.

[0088] Optionally, the device also includes an update module 212, which is specifically used to respond to the need for the target satellite to update the target onboard model, retrain the base large model for the target mission field, and obtain an updated expert large model for the target mission field; train the onboard base model according to the updated expert large model to obtain an updated incremental fine-tuning model for the target mission field; transmit the updated incremental fine-tuning model to the target satellite, so that the target satellite fuses the onboard base model and the updated incremental fine-tuning model to obtain an updated onboard model.

[0089] Optionally, the update module 212 is specifically used for the target satellite to test whether the updated onboard model can operate normally; in response to the updated onboard model being able to operate normally, the target satellite adopts the updated onboard model to replace the target onboard model; in response to the updated onboard model not being able to operate normally, the target satellite returns an error prompt.

[0090] This specification also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 Provided onboard model deployment method.

[0091] This manual also provides Figure 3 The schematic structure diagram of the electronic device shown in FIG. Figure 3 As mentioned above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1Of course, in addition to the software implementation, this specification does not exclude other implementations, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0092] For the improvement of a technology, it can be clearly distinguished whether it is a hardware improvement (for example, improvement of the circuit structure of diodes, transistors, switches, etc.) or a software improvement (improvement of the method flow). However, with the development of technology, many improvements of the method flow today can be regarded as direct improvements of the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be implemented with a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming themselves, without having to ask chip manufacturers to design and make dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0093] The controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, and the memory controller may also be implemented as part of the control logic of the memory. It is also known to those skilled in the art that, in addition to implementing the controller in a purely computer-readable program code manner, the controller may be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, such a controller may be considered as a hardware component, and the devices for implementing various functions included therein may also be considered as structures within the hardware component. Or even, the devices for implementing various functions may be considered as both software modules for implementing the method and structures within the hardware component.

[0094] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0095] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0096] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0097] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0098] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0099] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0100] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0101] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0102] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic 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 media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0103] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0104] It should be understood by those skilled in the art that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0106] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0107] The above description is only an embodiment of this specification and is not intended to limit this specification. For those skilled in the art, this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification should be included in the scope of the claims of this application.

Claims

1. A method for deploying a satellite model, characterized in that: include: Acquire a pre-trained base large model and a satellite-borne base model, and transmit the satellite-borne base model to the target satellite, wherein the base large model is a large model oriented to a general field, the satellite-borne base model is a model oriented to a general field for deployment on a satellite, and the base large model contains more parameters than the satellite-borne base model; Determine the target mission area for the target spaceborne model to be deployed; Training the base large model for the target task field to obtain an expert large model for the target task field; The onboard base model is trained according to the expert large model to obtain an incremental fine-tuning model for the target mission field; The incremental fine-tuning model is transmitted to the target satellite, so that the target satellite fuses the satellite-borne base model and the incremental fine-tuning model to obtain a target satellite-borne model.

2. The method according to claim 1, characterized in that Training the base large model for the task domain specifically includes: Obtaining training samples and training annotations for the task domain; Inputting the training sample into the base large model to obtain an output result to be optimized output by the base large model; According to the difference between the output result to be optimized and the training annotation, the parameters of the base large model are adjusted.

3. The method according to claim 1, characterized in that Training the satellite-borne base model according to the expert large model specifically includes: The knowledge of the expert large model is transferred to the onboard base model through knowledge distillation.

4. The method according to claim 1, characterized in that The method further comprises: For each parameter included in the target satellite-borne model, the number of decimal places of the parameter is quantized to a specified number of places to obtain a lightweight target satellite-borne model.

5. The method according to claim 1, characterized in that The method further comprises: In response to the target satellite needing to update the target onboard model, the base large model is retrained for the target mission field to obtain an updated expert large model for the target mission field; The onboard base model is trained according to the updated expert large model to obtain an updated incremental fine-tuning model for the target mission domain; The updated incremental fine-tuning model is transmitted to the target satellite, so that the target satellite fuses the onboard base model and the updated incremental fine-tuning model to obtain an updated onboard model.

6. The method according to claim 5, characterized in that After obtaining the updated onboard model, the method includes: The target satellite tests whether the updated onboard model can operate normally; In response to the updated onboard model being able to operate normally, the target satellite adopts the updated onboard model to replace the target onboard model; In response to the update onboard model failing to operate normally, the target satellite returns an error prompt.

7. A satellite-borne model deployment device, characterized in that: include: An acquisition module is used to acquire a pre-trained base large model and a satellite-borne base model, and transmit the satellite-borne base model to a target satellite, wherein the base large model is a large model oriented to a general field, the satellite-borne base model is a model oriented to a general field for deployment on a satellite, and the base large model contains more parameters than the satellite-borne base model; A determination module, used to determine the target mission area for the target onboard model to be deployed; A training module, used for training the base large model for the target task field to obtain an expert large model for the target task field; A fine-tuning module, used for training the onboard base model according to the expert large model to obtain an incremental fine-tuning model for the target mission field; The transmission module is used to transmit the incremental fine-tuning model to the target satellite, so that the target satellite fuses the satellite-borne base model and the incremental fine-tuning model to obtain a target satellite-borne model.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. An electronic 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 program, the method described in any one of claims 1 to 6 is implemented.

Images