Communication method and communication apparatus

Abstract

A communication method and a communication apparatus. In the method, a second entity may provide a first entity with a requirement of the second entity for output information of a large model, and the first entity may select a suitable compression algorithm on the basis of the requirement, such that the first entity no longer executes compression according to a fixed compression algorithm, but on the basis of the actual requirement of the second entity, thereby facilitating the prevention of the context of the large model from exceeding a context window, and facilitating an improvement in the accuracy of the large model to process long text and an enhancement in the capability of the large model to process a large volume of data.

Description

A communication method and a communication device

[0001] This application claims priority to Chinese Patent Application No. 202510125027.1, filed on January 26, 2025, entitled "A Communication Method and Communication Device", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of communication technology, and more specifically, to a communication method and a communication device. Background Technology

[0003] In multi-agent architectures based on artificial intelligence (AI), there is interaction and collaboration between agents. In multi-turn dialogue scenarios, the complexity of the task continues to increase, typically involving chain-of-thought (COT), context learning, and retrieving relevant documents or historical dialogues. For example, wireless base station outages involve multiple domains, and the localization process relies on multi-turn interactions between agents in each domain.

[0004] Interactions between agents can be achieved with the assistance of a large model. The large model can generate responses within a context window, which refers to the number of tokens considered when generating a response. A larger context window helps the large model better understand user input, ensuring the consistency and coherence of its responses. However, considering factors such as computing resources, memory, response quality, and technical implementation, the context window is not infinitely large and is usually limited.

[0005] However, because the large model's interface is stateless, each round of interaction needs to carry all historical dialogues (or context) from the start of the interaction to the current interaction. This will significantly increase the context length of the large model, which may exceed the large model's context window. Once the large model's context length exceeds the context window, it will affect the logic and coherence of the session, reducing the satisfaction of network management services. Summary of the Invention

[0006] Embodiments of this application provide a communication method and a communication device to ensure the logic and coherence of large-scale model sessions, thereby improving the satisfaction of network management services.

[0007] In a first aspect, embodiments of this application provide a communication method that can be applied to a first entity, modules within the first entity (e.g., circuits, chips, or chip systems such as modem chips (also known as baseband chips), or system-on-chip (SoC) chips or system-in-package (SIP) chips containing modem cores), or logical nodes, logical modules, or software that can implement all or part of the first entity. Exemplarily, the first entity can be a manufacturing entity.

[0008] The method includes: receiving first information and second information from a second entity, wherein the first information is input information of a large model and the second information is used to indicate the second entity's demand for output information of the large model; and compressing the first information according to the second information to obtain third information.

[0009] Based on the above method, the second entity can provide the first entity with its requirements for the output information of the large model. The first entity can then select a suitable compression algorithm based on these requirements. This allows the first entity to flexibly perform compression according to the actual needs of the second entity, rather than following a fixed compression algorithm. This helps to prevent the context of the large model from exceeding the context window and improves the accuracy of the large model in processing long texts and enhances its ability to handle large amounts of data.

[0010] In conjunction with the first aspect, in some possible implementations, the second information includes the compression type preferred by the second entity; wherein the compression type is a first compression type or a second compression type, the first compression type supporting the guarantee of the quality of the output information, and the second compression type supporting the guarantee of the length of the output information.

[0011] In the above implementation, a compression type attribute is added to the information exchanged between the first entity and the second entity to indicate the compression type preferred by the second entity. This allows the first entity to no longer compress according to a fixed compression method, but to execute different compression algorithms according to user preferences in order to achieve the operation desired by the first entity.

[0012] In conjunction with the first aspect or any of its implementations, in some other possible implementations, the second information further includes length information, which is used to indicate the length range of the output information.

[0013] Based on the above implementation, the second entity can provide more accurate compression constraint information, allowing the first entity to select a more suitable compression algorithm to meet the expectations of the second entity.

[0014] In conjunction with the first aspect or any of its implementations, in some other possible implementations, the step of compressing the first information based on the second information to obtain the third information includes: determining a first compression algorithm based on the second information; and using the first compression algorithm to compress the first information to obtain the third information.

[0015] In conjunction with the first aspect or any of its implementations, in some other possible implementations, the method further includes: sending the third information to a third entity, the third entity including the large model.

[0016] In conjunction with the first aspect or any of its implementations, in some other possible implementations, the first entity and the third entity are located in the same physical entity; or, the first entity and the third entity are located in different physical entities.

[0017] In conjunction with the first aspect or any of its implementations, in some other possible implementations, the method further includes: sending fourth information to the third entity, the fourth information being used to indicate a first compression algorithm, the third information being obtained based on the first compression algorithm.

[0018] Based on the above implementation, the first entity provides the third entity with the first compression algorithm corresponding to the third information, which enables the third entity to better understand the third information, thus helping to improve consumer satisfaction with the dialogue.

[0019] In conjunction with the first aspect or any implementation thereof, in some other possible implementations, before compressing the first information according to the second information, the method further includes: determining that the length of the first information exceeds the context window of the large model.

[0020] Based on the above implementation, the first entity can perform compression of the first information when the context of the large model exceeds the context window. This can avoid unnecessary compression operations and thus help improve the dialogue response speed.

[0021] In conjunction with the first aspect or any of its implementations, in some other possible implementations, the first information is a first prompt and the third information is a second prompt; or, the first information is text to be compressed and the third information is a third prompt.

[0022] Secondly, embodiments of this application provide a communication method that can be applied to a second entity, modules within the second entity (e.g., circuits, chips, or chip systems (such as modem chips, or SoC chips or SIP chips containing modem cores)), or logical nodes, logical modules, or software that can implement all or part of the second entity. For example, the second entity can be a consumer entity.

[0023] The method includes: determining first information and second information, wherein the first information is input information of a large model, and the second information is used to indicate the second entity's demand for output information of the large model; and sending the first information and the second information to a first entity.

[0024] Based on the above method, the second entity can provide the first entity with its requirements for the output information of the large model, so that the first entity can select a suitable compression algorithm based on the requirements. In this way, the first entity no longer performs compression according to a fixed compression algorithm, but performs compression flexibly according to the actual needs of the second entity. This helps to avoid the context of the large model from exceeding the context window, and also helps to improve the accuracy of the large model in processing long texts and enhance the large model's ability to process large amounts of data.

[0025] In conjunction with the second aspect, in some possible implementations, the second information includes the compression type preferred by the second entity; wherein the compression type is a first compression type or a second compression type, the first compression type supporting the guarantee of the quality of the output information, and the second compression type supporting the guarantee of the length of the output information.

[0026] In the above implementation, a compression type attribute is added to the information exchanged between the first entity and the second entity to indicate the compression type preferred by the second entity. This allows the first entity to no longer compress according to a fixed compression method, but to execute different compression algorithms according to user preferences in order to achieve the operation desired by the first entity.

[0027] In conjunction with the second aspect or any of its implementations, in some other possible implementations, the second information further includes length information, which is used to indicate the length range of the output information.

[0028] Based on the above implementation, the second entity can provide more accurate compression constraint information, allowing the first entity to select a more suitable compression algorithm to meet the expectations of the second entity.

[0029] In conjunction with the second aspect or any of its implementations, in some other possible implementations, the first information is a first prompt; or, the first information is text to be compressed.

[0030] Thirdly, embodiments of this application provide a communication method that can be applied to a third entity, a module within the third entity (e.g., a circuit, chip, or chip system (such as a modem chip, or a SoC chip or SIP containing a modem core), or a logical node, logical module, or software that can implement all or part of the third entity. For example, the third entity can be the entity containing a large model.

[0031] The method includes: receiving third information and fourth information from a first entity, wherein the third information is input information of the large model and is compressed information, and the fourth information is used to indicate a first compression algorithm and is obtained based on the first compression algorithm; and interacting with a second entity through the first entity based on the third information and the fourth information.

[0032] Based on the above method, the first entity provides the third entity with the first compression algorithm corresponding to the third information, which enables the third entity to better understand the third information, thus helping to improve consumer satisfaction with the dialogue.

[0033] Fourthly, embodiments of this application provide a communication method that can be applied to a first entity and a second entity. Unless otherwise specified, the first entity and the second entity may refer to the first entity and the second entity themselves, or to modules (e.g., circuits, chips, or chip systems (such as modem chips, or SoC chips or SIP chips containing modem cores)) in the first entity and the second entity, or logical nodes, logical modules, or software that can implement all or part of the first entity and the second entity.

[0034] The method includes: a second entity sending first information and second information, wherein the first information is input information of a large model, and the second information is used to indicate the second entity's demand for output information of the large model; a first entity receiving the first information and the second information; and the first entity compressing the first information according to the second information to obtain third information.

[0035] In conjunction with the fourth aspect, in some possible implementations, the method further includes: the first entity sending the third information, the third entity including the large model; and the third entity receiving the third information.

[0036] It should be noted that the steps performed by the first entity in the fourth aspect or its implementation can refer to the first aspect or its implementation; the steps performed by the second entity in the fourth aspect or its implementation can refer to the second aspect or its implementation; the steps performed by the third entity in the fourth aspect or its implementation can refer to the third aspect or its implementation; the terms or features that are the same as those in the first, second, and third aspects or their implementations in the fourth aspect or its implementation can refer to the first, second, and third aspects or their implementations; and the technical effects of the fourth aspect or its implementation can refer to the technical effects in the first, second, and third aspects or their implementations, and will not be repeated in the fourth aspect.

[0037] Fifthly, a communication apparatus is provided for performing the method provided in any of the above aspects or implementations thereof. Specifically, the apparatus may include units and / or modules for performing the method provided in any of the above aspects or implementations thereof, such as a processing unit and / or a transceiver unit. The processing unit is used to perform the processing steps in the method provided in any of the above aspects or implementations thereof. The transceiver unit is used to perform the transceiver steps in the method provided in any of the above aspects or implementations thereof.

[0038] In one implementation, the device is a first entity, a second entity, or a third entity. When the device is a first entity, a second entity, or a third entity, the transceiver unit can be a transceiver, an input / output interface, or a communication interface; the processing unit can be at least one processor. Optionally, the transceiver is a transceiver circuit. Optionally, the input / output interface is an input / output circuit.

[0039] In another implementation, the device is a chip, chip system, or circuit used in a first entity, a second entity, or a third entity. When the device is a chip, chip system, or circuit used in a first entity, a second entity, or a third entity, the transceiver unit may be an input / output interface, interface circuit, output circuit, input circuit, pin, or related circuit on the chip, chip system, or circuit; the processing unit may be at least one processor, processing circuit, or logic circuit.

[0040] In a sixth aspect, a communication device is provided, comprising: a memory for storing a program; and at least one processor for executing the computer program or instructions stored in the memory to perform the method provided in any of the foregoing aspects or their implementations.

[0041] In one implementation, the device is a first entity, a second entity, or a third entity.

[0042] In another implementation, the device is a chip, chip system, or circuit for use in a first entity, a second entity, or a third entity.

[0043] In a seventh aspect, a communication device is provided, comprising: at least one processor and a communication interface, the at least one processor being configured to obtain a computer program or instructions stored in a memory via the communication interface to execute the method provided in any of the foregoing aspects or their implementations. The communication interface may be implemented in hardware or software.

[0044] In one implementation, the device further includes the memory.

[0045] Eighthly, a processor is provided for executing the methods provided in the above aspects.

[0046] Unless otherwise specified, or if it does not contradict its actual function or internal logic in the relevant description, the transmission and acquisition / reception operations involved in the processor can be understood as processor output and reception, input and other operations, or as transmission and reception operations performed by radio frequency circuits and antennas. This application does not limit them in this regard.

[0047] Ninthly, a computer-readable storage medium is provided that stores program code for execution by a device, the program code including methods for performing any of the foregoing aspects or their implementations.

[0048] In a tenth aspect, a computer program product containing instructions is provided, which, when run on a computer, causes the computer to perform the method provided in any of the foregoing aspects or their implementations.

[0049] Eleventhly, a chip is provided, comprising a processor and a communication interface. The processor reads instructions stored in memory through the communication interface and executes the methods provided in any of the above aspects or their implementations. The communication interface can be implemented in hardware or software.

[0050] Optionally, as one implementation, the chip also includes a memory that stores computer programs or instructions. The processor is used to execute the computer programs or instructions stored in the memory. When the computer programs or instructions are executed, the processor is used to perform the methods provided by any of the above aspects or their implementations.

[0051] When the method provided in this application is executed by a chip, this application does not limit the specific number of chips implementing the method. For example, it can be executed by one chip, or by two or more chips. Furthermore, when the number of chips implementing the method is two or more, the chip manufacturers are not limited; they can be from the same manufacturer or different manufacturers.

[0052] In a twelfth aspect, a communication system is provided, comprising at least one of the first entity, second entity, or third entity described above.

[0053] In a thirteenth aspect, a computer program is provided that, when run on a computer, causes the methods provided by any of the foregoing aspects or their implementations to be executed. Attached Figure Description

[0054] Figure 1 is a schematic structural diagram of a communication system applicable to an embodiment of this application.

[0055] Figure 2 is a schematic structural diagram of another communication system applicable to the embodiments of this application.

[0056] Figure 3 is a schematic flowchart of the communication method 300 provided in this application.

[0057] Figure 4 is a schematic flowchart of the communication method 400 provided in this application.

[0058] Figure 5 is a schematic flowchart of the communication method 500 provided in this application.

[0059] Figure 6 is a schematic diagram of a device provided in an embodiment of this application.

[0060] Figure 7 is another structural schematic diagram of the device provided in an embodiment of this application.

[0061] Figure 8 is a schematic diagram of a chip system provided in an embodiment of this application. Detailed Implementation

[0062] Before introducing the embodiments of this application, the following description is provided.

[0063] The terms "for indicating" or "instruction" can include both direct and indirect indication, or they can be explicit and / or implicit. The various numerical designations such as "first," "second," etc., are merely for descriptive convenience and are not intended to limit the scope of the embodiments of this application, such as distinguishing different messages or different information. "Predefined" can be implemented by pre-storing corresponding codes, tables, or other methods that can be used to indicate relevant information in the device; this application does not limit the specific implementation method. The "protocol" involved can refer to standard protocols in the field of communication, such as the Long Term Evolution (LTE) protocol, the New Radio (NR) protocol, and related protocols applied to future communication systems; this application does not limit this. The words "exemplary," "for example," "exemplarily," "as another example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as an "example" in this application should not be construed as being better or more advantageous than other embodiments or designs. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized. "A plurality" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and c can mean: a, or, b, or, c, or, a and b, or, a and c, or, b and c, or, a, b, and c. Here, a, b, and c can be single or multiple. Descriptions involving network element A sending messages, information, or data to network element B, and network element B receiving messages, information, or data from network element A, aim to specify which network element the message, information, or data is to be sent to, without specifying whether they are sent directly or indirectly through other network elements. Descriptions such as “when…”, “under…”, “if”, and “if” all indicate that the device will take corresponding actions under certain objective circumstances. They are not time-limited, nor do they require the device to make a judgment action when implementing the action, nor do they imply any other limitations.

[0064] The embodiments of this application can be applied to various communication systems, including but not limited to: 5th generation (5G) systems or NR systems, LTE systems, Long Term Evolution-Advanced (LTE-A) systems, LTE Frequency Division Duplex (FDD) systems, LTE Time Division Duplex (TDD) systems, etc. They can also be applied to future communication systems. Furthermore, they can be applied to device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, long term evolution-vehicle (LTE-V) communication, long term evolution-machine (LTE-M) communication, machine-to-machine (M2M) communication, machine-type communication (MTC), Internet of Things (IoT) communication systems, narrowband Internet of Things (NB-IoT) systems, or other communication systems. The embodiments of this application can also be applied to satellite communication systems, high altitude platform station (HAPS) communications, unmanned aerial vehicles (UAVs), and other non-terrestrial network (NTN) systems, such as integrated communication and navigation (ICAN) systems, global navigation satellite systems (GNSS), and ultra-dense low-Earth orbit (LEO) satellite communication systems. Furthermore, it can be extended to similar wireless communication systems, such as wireless local area networks (WLANs), wireless-fidelity (WiFi), worldwide interoperability for microwave access (WIMAX), and other communication systems related to the 3rd generation partnership project (3GPP).

[0065] The embodiments of this application can also be applied to various network function virtualization (NFV) systems. These systems can use standard formal language to describe current operator service technology solutions, network construction solutions, and network operation and maintenance methods as patterns and strategies, and implement the technology solutions and construction solutions based on these patterns and strategies. For example, the technical solutions of this application can be applied to one or more of the following systems: wireless intent-driven network (wIDN) systems, experiential networked intelligence (ENI) systems, intent-driven management service (IDMS) systems, or open network automation platform (ONAP) systems, etc.

[0066] To facilitate understanding of the embodiments of this application, some terms involved in the embodiments of this application will be explained first.

[0067] 1. Intelligent agent

[0068] Intelligent agents generally refer to entities that are capable of taking initiative, possess consciousness or will, and have the ability to make decisions and choices. Common intelligent agents include at least one of the following:

[0069] Artificial intelligence (AI) agent / LLM-based agent: Based on the intelligent agent, AI or LLM is used to realize the perception, decision-making and other capabilities of the intelligent agent.

[0070] Autonomous agent: A system that can autonomously perceive its environment, make decisions and execute actions, and can complete tasks independently in complex environments.

[0071] Autonomous networks agent: A self-closing system that, in response to a specific network scenario, can proactively perceive the network state and its environment, analyze the perceived information, make adaptive decisions, and perform corresponding controls to achieve user goals.

[0072] 2. AI agent

[0073] An AI agent can be an instance of an intelligent agent within a communication device. In embodiments of this application, the AI ​​agent can be an application instance generated by a large model application framework based on a large model. AI agents can interact to achieve business goals.

[0074] An AI agent framework can include the following key components:

[0075] 1) Brain: Based on a large model, it provides memory and decision-making abilities;

[0076] 2) Perception: Responsible for receiving external input, perceiving information from the environment, and providing it to the brain;

[0077] 3) Action: Responsible for executing the action instructions given by the brain, including generating text, calling tools, and controlling embodied devices. Embodied devices typically refer to devices with physical form and sensory capabilities, enabling them to communicate and interact naturally with humans.

[0078] Application paradigms for AI agents can include:

[0079] 1) Machine-attended, i.e., single agent, can handle simple tasks;

[0080] 2) Human-machine co-driving, also known as agent-human interaction, can handle complex tasks;

[0081] 3) Swarm intelligence, namely agent-agent interaction, can handle cross-team collaborative tasks.

[0082] In the embodiments of this application, the AI ​​agent can also be simply referred to as agent.

[0083] 3. Interaction and collaboration between AI agents

[0084] In a multi-agent architecture based on AI agents, there is interaction and collaboration between AI agents. The following example, using a wireless base station outage troubleshooting process, provides a simple description of the interaction and collaboration between AI agents.

[0085] Wireless base station outages involve multiple domains, such as the wireless domain, core network domain (e.g., cloud core network), data communication domain, optical domain, and environmental domain. The network management system for the wireless domain and / or core network domain can be a mobile broadband automation engine (MAE), while the network management system for the data communication and / or optical domain can be a network cloud engine (NCE). Through the autonomous collaboration between the MAE agent (i.e., the wireless agent) and the NCE agent (i.e., the transmission agent), the location and repair of base station outages can be completed. For example, if a batch of base station outage alarms occur in a certain district of a city, maintenance personnel can issue a task, requesting the MAE agent and NCE agent to assist in troubleshooting and resolving the issue. Upon receiving the task, the MAE agent checks the environmental monitoring system and, through routine troubleshooting, identifies a list of base stations with transmission failures. The MAE agent can then provide this list to the NCE agent for collaborative troubleshooting. After receiving the list, the NCE agent can check each base station on the list, generating work orders for each faulty base station to repair. If base stations without transmission failures are identified, the NCE agent can provide a list of these base stations to the MAE agent for detailed investigation. The MAE agent can then perform detailed investigations on these base stations, generating work orders for faulty base stations and conducting on-site investigations of the non-transmission failure base stations. After repairing the base stations with transmission failures, the NCE agent can notify the MAE agent that all base stations with transmission failures have been repaired, and the MAE agent confirms the repair result. Once the agent confirms that the base station with the transmission failure has been repaired, and all base stations without the transmission failure have also been repaired, the process of locating and repairing the base station outage failure ends.

[0086] It should be noted that the aforementioned "MAE agent" or "NCE agent" can refer to an agent within the MAE or NCE that possesses the function or capability of troubleshooting wireless base station faults. The MAE or NCE may also have agents with other functions or capabilities; that is, different agents can be set up within the MAE or NCE for different functions or capabilities. Alternatively, the aforementioned "MAE agent" or "NCE agent" can refer to an agent representing the MAE or NCE. This agent, in addition to possessing the function or capability of troubleshooting wireless base station faults, may also possess other functions or capabilities. In other words, there is only one agent representing the MAE or NCE, and this agent may possess one or more functions or capabilities.

[0087] 4. Large Model

[0088] Large models are used to provide large-scale model services. A large model refers to a neural network model containing an extremely large number of parameters. Large models can also be called foundation models. Large models play a crucial role in many fields and applications, with common applications including natural language processing, computer vision, speech recognition and synthesis, recommendation systems, financial risk control, intelligent dialogue systems, game AI, and healthcare. In the future, large models will evolve towards multimodal models, which can handle data from multiple modalities. For example, multimodal large models can process natural language as well as text, images, or videos.

[0089] Large language models (LLMs) are a type of large model that is a deep learning model trained on large amounts of text data. LLMs can generate natural language text or understand the meaning of language text. LLMs can handle various natural language tasks, such as text classification, question answering, and dialogue.

[0090] 5. Large-scale model application framework

[0091] The Large Model Application Framework is a framework for developing large model applications. Based on components such as models, prompts, and memory, it provides capabilities like prompt templates, model orchestration, large model services, and security isolation, helping developers achieve a simple, secure, and reliable large model application building experience. As a user module of the large model, the Large Model Application Framework can provide functions such as agent instance generation and local agent configuration information storage. The Large Model Application Framework can be represented as a copilot or an agent. It can also be replaced with other names, such as an artificial intelligence (AI) module, a prompt unit, or a prompt engine. The Large Model Application Framework can be deployed in consumer or production entities, or in other locations, such as other devices or cloud platforms, without restriction.

[0092] 6. Prompt

[0093] Prompt is an interaction method based on natural language processing. It enables communication between users and machines by parsing natural language. The main implementation of prompt is to convert natural language into machine-recognizable instructions by building a corresponding corpus and semantic parsing model. Prompt can generate finer-grained tasks by piecing together prompts from the original text based on context, thought processes, or cue words, guiding larger models to better complete tasks.

[0094] "Prompt" can also be described as a hint, a message, or a guide.

[0095] 7. token

[0096] In the field of artificial intelligence, especially in natural language processing (NLP), a token refers to the smallest unit or basic element of text processing. A token can be a word, a phrase, a punctuation mark, a subword, or a character, etc. Currently, many large models, whether in terms of their demonstration capabilities or pricing, use tokens as the unit. A token can be described as a word unit.

[0097] 8. Context Window

[0098] The context window refers to the number of tokens a model considers when generating a response. It determines the range of information the model can capture, similar to the contextual information humans rely on when thinking about a problem. A larger context helps the model better understand user input, maintaining coherence and consistency, thereby improving the quality and accuracy of task completion.

[0099] Typically, context windows have limitations:

[0100] 1) Computational resource limitations: Processing and generating a large number of tokens requires significant computational resources and time. To maintain responsiveness in practical applications, models typically set token limits to balance computational resource usage and performance.

[0101] 2) Memory Limitations: When generating text, the model needs to maintain context information in memory. If the generated text is too long, the required memory will increase rapidly, which places very high demands on the hardware, especially when handling multiple concurrent requests.

[0102] 3) Quality control: The longer the generated text, the more difficult it is for the model to maintain contextual consistency, which may result in repetitive, irrelevant or illogical output text.

[0103] 4) Technical Implementation: Existing transformer architectures and implementations have certain limitations on sequence length. During model training and inference, excessively long sequences can lead to an exponential increase in computational complexity.

[0104] Therefore, the limitation of the context window is intended to provide better performance and user experience in practical applications, and it is also an optimization choice under the current technology and hardware conditions.

[0105] 9. Retrieval-augmented generation (RAG)

[0106] RAG (Related Aspects of Generative Grammar) is a natural language processing approach that combines retrieval and generation techniques, aiming to enhance the capabilities of generative models by leveraging external knowledge bases or documents. The core idea of ​​RAG is to combine the generative capabilities of a large model with the accuracy of a retrieval system, enabling the large model to generate more accurate and information-rich content. RAG can expand the parameterized knowledge of the large model by incorporating external data sources, thereby improving the relevance and accuracy of the generated content. This method does not require retraining the model; instead, it supplements the original knowledge of the large model by introducing retrieved background information, improving the search experience and generated results.

[0107] The relevant terms used in the embodiments of this application have been explained above, and will not be explained again below. The system architecture applicable to the embodiments of this application will be described below with reference to Figures 1 and 2.

[0108] Figure 1 is a schematic structural diagram of a communication system applicable to an embodiment of this application.

[0109] Management Service Consumer Entity: The entity that invokes the management service is called the management service consumer entity. The management service consumer entity in Figure 1 can be used to invoke the large model management service or the large model service, or in other words, to invoke the large model management service to manage the network system. The management service consumer entity can act as the initiator of interactions between agents and supports communication with agents from other domains.

[0110] The management service consumption entity can also be referred to as a large-scale management service consumption entity, management service consumption network element, management service consumer (MnS consumer), intent management service consumer, service consumer, intent consumer, intent owner, network management service consumption entity, etc. In future communication systems, the management service consumption entity may have other names, and this application does not impose any particular limitation on this. The capabilities or functions of the management service consumption entity can be deployed on a certain network element, which is called a management service consumption network element; the capabilities or functions of the management service consumption entity can also be deployed on other devices, and this application does not limit this. For ease of description, this application uses the management service consumption entity as an example for illustration, but all can be replaced by other devices that have deployed the capabilities or functions of the management service consumption entity.

[0111] Management Service Producing Entity: The entity that provides management services is called the management service producing entity. The management service producing entity in Figure 1 can be used to provide large-scale model management services or large-scale model services. For example, the management service producing entity can receive prompts from management service consuming entities and perform network management based on the received prompts. The management service consuming entity can act as a receiver for interactions between agents and supports communication with agents in other domains.

[0112] The management service production entity can also be referred to as a large-scale management service production entity, management service production network element, intent management service provider, intent service provider, management service producer (MnS producer), intent provider, intent handler, network management service production entity, etc. In future communication systems, the management service production entity may have other names, which this application does not specifically limit. The capabilities or functions of the management service production entity can be deployed on a network element, which is called a management service production network element; the capabilities or functions of the management service production entity can also be deployed on other devices, which this application embodiment does not limit. For ease of description, this application embodiment uses a management service production entity as an example, but all can be replaced by other devices that have deployed the capabilities or functions of a management service production entity.

[0113] The management service consumer entity and management service producer entity in Figure 1 perform network management with the assistance of the large model, based on the large model management service (MnS) interface.

[0114] Figure 2 is a schematic structural diagram of another communication system applicable to the embodiments of this application.

[0115] Unlike the communication system shown in Figure 1, in addition to the management service production entity, the large model sets up a management service consumption entity and a management service production entity, which perform network management with the assistance of the large model based on the management service (MnS) interface (such as the intent-driven management service (MnS) interface).

[0116] The entities in Figures 1 and 2 can be network elements within hardware devices, software functions running on dedicated hardware, or virtualization functions instantiated on a platform (e.g., a cloud platform). It is understood that these entities can be implemented by a single device or by multiple devices working together. Furthermore, these entities can also be functional modules within a system, such as a network management system (NMS), an equipment management system (EMS), or functional modules within a device, such as one or more functional modules within network equipment (NE). Network equipment can be a base station or a core network element.

[0117] As an example, the management service consumer entity can be deployed within the NMS, while the management service production entity can be deployed within different EMSs. The interaction between the management service consumer entity and the management service production entity can be achieved through the interface between the NMS and the EMS.

[0118] As another example, the management service consumer entity can be deployed within the EMS, while the management service production entity can be deployed within different NEs. The interaction between the management service consumer entity and the management service production entity can be achieved through the interface between the NMS and the EMS.

[0119] Optionally, the aforementioned communication system may also include other devices or network elements, such as a knowledge base or a network twin service production entity. A network twin service production entity may be, for example, a network digital twin (NDT) system or an entity deployed with an NDT system, which can be used to provide network management services.

[0120] The solution proposed in this application can also be applied to other systems that contain the corresponding entities. This application does not limit the scope of the solution, and the embodiments of this application do not impose any specific limitations on this.

[0121] It should be noted that the management service consumer entity and / or management service producer entity in Figures 1 and 2 can be an agent, or can refer to an entity in general.

[0122] It should be understood that the network architecture and business scenarios described in the embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided in the embodiments of this application. As those skilled in the art will know, with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0123] For the sake of simplicity, the following text will refer to the management service production entity as the first entity, the management service consumption entity as the second entity, and the entity with a large model deployed as the third entity.

[0124] In multi-turn dialogue scenarios involving agents, task complexity continuously increases, typically involving thought chains, context learning, and retrieving relevant documents or historical dialogues. For example, the wireless base station outage fault mentioned above involves multiple domains, and the localization process relies on multi-turn interactions between agents in each domain. However, because the interface of the large model is stateless, the prompt for each round of interaction needs to carry all historical dialogues (or context) from the start of the interaction to the current interaction, gradually increasing the prompt length. For long text tasks, it is necessary to analyze massive amounts of data based on the large model, identify key information, and answer questions. Due to the sheer volume of data, the prompt length increases accordingly, also leading to an increase in the context length of the large model. Furthermore, the demand for information retrieval based on RAG in the telecommunications field is constantly increasing. When using RAG technology, it is usually necessary to combine multiple relevant documents to provide background knowledge to the large model. This approach can bring richer external knowledge, but it also significantly increases the prompt length. As the prompt length continues to increase, the context length of the large model also increases significantly. In this case, the context length of the large model is likely to exceed the context window. If the context length of a large model exceeds the context window, it will affect the logic and coherence of the session, reducing the satisfaction of network management services.

[0125] Currently, when the number of tokens exceeds the context window of the large model, the agent can compress the received prompt according to its internally configured compression algorithm before sending the compressed prompt to the large model. However, compression based on the internally configured algorithm is prone to over-compression or under-compression.

[0126] 1) Excessive compression can reduce accuracy: In lossy compression, the compression of the original data reduces the amount of information and decreases the accuracy of LLM inference.

[0127] 2) Insufficient compression can affect the integrity of the output: The input and output of a large model share the same context window. If the input data takes up too many tokens, the output of the large model may not have enough tokens available, resulting in output truncation.

[0128] To address the aforementioned issues, this application provides a communication method and a communication device to ensure the logic and coherence of large-scale model sessions, thereby improving the satisfaction of network management services.

[0129] The method embodiments of this application are described below.

[0130] Figure 3 is a schematic flowchart of the communication method 300 provided in this application.

[0131] The method shown in Figure 3 can be performed by a first entity, a second entity, and an optional third entity. Unless otherwise specified, "first entity," "second entity," or "third entity" can refer to the first entity, the second entity, or the third entity itself, or to components within the first entity, the second entity, or the third entity (e.g., circuits, chips, or chip systems (such as modem chips, or SoC chips or SIP chips containing modem cores, etc.), or to logic modules or software that can implement all or part of the functions of the first entity, all or part of the functions of the second entity, or all or part of the functions of the third entity).

[0132] For example, the first entity can be the management service production entity in Figure 1 or Figure 2, the second entity can be the management service consumption entity in Figure 1 or Figure 2, and the third entity can be the entity containing the large model in Figure 2.

[0133] For example, the first entity can be a transport domain agent, such as an NCE agent, and the second entity can be a radio domain agent, such as a MAE agent.

[0134] For example, the first entity can be a transport domain agent, such as an NCE agent, and the second entity can be an upper-layer network management agent, such as an operation support system (OSS) agent.

[0135] Method 300 may include some or all of the following.

[0136] Step 301: The second entity sends first information and second information to the first entity, and correspondingly, the second entity receives the first information and second information from the first entity.

[0137] The first piece of information refers to the input information of the large model. The second piece of information indicates the second entity's requirements for the output information of the large model, such as requirements for the accuracy or length of the output information. Output information can also be replaced with output, output text, output data, or output results, etc., without restriction. The second piece of information indicating the second entity's requirements for the output information of the large model can also be replaced with: the second piece of information indicating the second entity's requirements for compression results, compression methods, or compression performance, etc.

[0138] In embodiments of this application, a prompt can be generated by a second entity, and then the first entity can compress the received prompt. In this case, the first information can be the prompt itself. Alternatively, the second entity can provide information for generating the prompt, and then the first entity can generate and compress the prompt based on the information provided by the second entity. In this case, the first information can be the original text or the text to be compressed.

[0139] The embodiments of this application do not limit the implementation of the second information.

[0140] In some implementations, the second information includes the second entity's preference for compression types. That is, the second information indirectly indicates the second entity's demand for the output information of the large model through the second entity's preference for compression types.

[0141] For example, the second information may include the compression type preferred by the second entity, which may be either a first compression type or a second compression type.

[0142] The first compression type supports ensuring the quality of the output information of the large model, or in other words, the first compression type prioritizes ensuring the quality of the output information of the large model, or in other words, the first compression type guarantees the quality of the output information of the large model. For example, the scenarios to which the first compression type is applicable have at least one of the following characteristics: high requirements for the large model's understanding of the input information, high requirements for the accuracy of the large model's output information, or low requirements for the length of the large model's output information, for example, scenarios where the large model provides yes / no judgments.

[0143] The second compression type supports ensuring the length of the output information of large models, or in other words, the second compression type prioritizes ensuring the length of the output information of large models. For example, the second compression type is applicable to scenarios with at least one of the following characteristics: low requirements for the large model's understanding of the input information, low requirements for the accuracy of the large model's output information (e.g., the large model's output has no fixed answer), or high requirements for the length of the large model's output information, such as scenarios where the large model generates high-quality articles or performs text optimization.

[0144] For example, the first compression type is lossless compression, and the second compression type is lossy compression.

[0145] For example, the first compression type is length-first, and the second compression type is quality-first.

[0146] The compression type of the second entity preference can also be replaced with: the compression method of the second entity preference, or the compression algorithm of the second entity preference, or the type of compression algorithm of the second entity preference, without restriction.

[0147] In some implementations, the second information includes the second entity's restrictions on the compression type. That is, the second information indirectly indicates the second entity's demand for the output information of the large model through the second entity's restrictions on the compression type.

[0148] For example, the second information includes a compression type limited by a second entity, which can be either a first compression type or a second compression type.

[0149] The first compression type prioritizes ensuring the quality of the output information of the large model, or in other words, it guarantees the quality of the output information of the large model. For example, the second compression type is applicable to scenarios with at least one of the following characteristics: low requirements for the large model's understanding of the input information, low requirements for the precision of the large model's output information (e.g., the large model's output has no fixed answer), or high requirements for the length of the large model's output information, such as scenarios where the large model generates high-quality articles or performs text optimization.

[0150] The second compression type supports ensuring the length of the output information of the large model, or in other words, the second compression type prioritizes ensuring the length of the output information of the large model, or in other words, the second compression type guarantees the length of the output information of the large model. For example, the scenarios to which the first compression type is applicable have at least one of the following characteristics: high requirements for the large model's understanding of the input information, high requirements for the accuracy of the large model's output information, or low requirements for the length of the large model's output information, for example, scenarios where the large model provides yes / no judgments.

[0151] For example, the first compression type is lossless compression, and the second compression type is lossy compression.

[0152] For example, the first compression type is length-first, and the second compression type is quality-first.

[0153] The compression type of the second entity restriction can also be replaced with: the compression method of the second entity restriction, or the compression algorithm of the second entity restriction, or the type of compression algorithm of the second entity restriction, which is not restricted.

[0154] Optionally, before step 301, method 300 may further include: the second entity determining the first information and the second information.

[0155] Step 302: The first entity compresses the first information based on the second information to obtain the third information.

[0156] In some implementations, the first entity can determine a first compression algorithm based on the second information, and then use the first compression algorithm to compress the first information to obtain the third information. For example, if the second information includes the second entity's preference for a compression type, the first entity determines a first compression algorithm that satisfies that preference based on the second information, and then uses the first compression algorithm to compress the first information to obtain the third information. For example, if the second information includes the second entity's restrictions on the compression type, the first entity determines a first compression algorithm that follows those restrictions based on the second information, and then uses the first compression algorithm to compress the first information to obtain the third information.

[0157] In other implementations, the first entity can determine a first compression algorithm based on the second information, and then use the first compression algorithm to compress the first and fourth information to obtain third information, where the fourth information is used to indicate the first compression algorithm. For example, if the second information includes the second entity's preference for a compression type, the first entity determines a first compression algorithm that satisfies that preference based on the second information, and then uses the first compression algorithm to compress the first and fourth information to obtain the third information. For example, if the second information includes the second entity's restriction on the compression type, the first entity determines a first compression algorithm that follows that restriction based on the second information, and then uses the first compression algorithm to compress the first and fourth information to obtain the third information.

[0158] For example, the fourth information may include the number, name, etc. of the first compression algorithm.

[0159] Based on method 300, the second entity can provide the first entity with its requirements for the output information of the large model. The first entity can select a suitable compression algorithm based on these requirements, so that the first entity no longer compresses the prompt according to a fixed compression algorithm. This helps to prevent the context of the large model from exceeding the context window, and also helps to improve the accuracy of the large model in processing long text prompts and enhance the large model's ability to process large amounts of data.

[0160] In other embodiments of this application, the second information may further include length information, which indicates the length range (or token quantity range) of the output information of the large model. That is, the second entity can also provide the length range of the output information of the large model. For example, the length range may be less than a first threshold, such as a token quantity less than 500. Alternatively, the length range may be greater than a second threshold, such as a token quantity greater than 140. Yet another example is a length range greater than a third threshold and less than a fourth threshold, such as a token quantity range of 2048 to 4000. In this way, the second entity can provide more precise compression constraint information, allowing the first entity to select a more suitable compression algorithm to meet the expectations of the second entity.

[0161] In some other embodiments of this application, before the first entity compresses the first information based on the second information, method 300 may further include: the first entity determining that the length of the first information exceeds the context window of the large model. In this implementation, the first entity can perform compression of the first information even when the context of the large model exceeds the context window, thus avoiding unnecessary compression operations and helping to improve dialogue response speed.

[0162] In some other embodiments of this application, method 300 may further include step 303, namely: the first entity sends third information to the third entity, and correspondingly, the third entity receives the third information from the first entity, wherein the third information is information obtained by compressing the first information based on the second information or information obtained by compressing the first information and the fourth information based on the second information; the third entity interacts with the second entity through the first entity based on the third information.

[0163] The first entity and the third entity can be deployed on different physical devices. For example, the first entity can be a production entity, and the third entity can be deployed on a third-party cloud. The first entity and the third entity can also be deployed on the same physical device, where they can be different modules or units of that physical entity. For example, both the first entity and the third entity can be deployed on a production entity, where they are different modules of that production entity. When the first entity and the third entity can be deployed on the same physical device, they can also be considered as the same entity, such as the first entity.

[0164] In some other embodiments of this application, when the third information does not include the fourth information, method 300 may further include step 304, namely: the first entity sends the fourth information to the third entity, and correspondingly, the third entity receives the fourth information from the first entity, wherein the fourth information is used to indicate the first compression algorithm. For example, the fourth information includes the number, name, etc. of the first compression algorithm. In this case, the third entity interacts with the second entity through the first entity based on the third information, which may include: the third entity interacts with the second entity through the first entity based on both the third and fourth information. In this implementation, the first entity provides the third entity with the first compression algorithm corresponding to the third information, which enables the third entity to better understand the third information, thus helping to improve consumer satisfaction with the dialogue.

[0165] The following describes method 300 in detail, taking the example of the second entity providing the first entity with a preferred compression type, the large model being an LLM, and the large model being deployed outside the first entity.

[0166] Figure 4 is a schematic flowchart of the communication method 400 provided in this application.

[0167] In method 400, agent1 can correspond to the second entity mentioned above, agent2 can correspond to the first entity mentioned above, and LLM can correspond to the third entity mentioned above.

[0168] Method 400 may include at least a portion of the following.

[0169] Step 401: agent1 sends information #1 to agent2, and agent2 receives information #1 from agent1 accordingly.

[0170] Information #1 includes the raw information of the natural language and the compression type (CompressType), which is the compression type preferred by agent1. The raw information of the natural language can correspond to the first information above, and the compression type can correspond to the second information above.

[0171] For example, information #1 can be as follows:

[0172] {

[0173] "header": {"CompressType": "quality_first"}

[0174] "content":

[0175] }

[0176] Where CompressType is the compression type, and content is the raw information in natural language.

[0177] Step 402: Agent2 selects a suitable compression algorithm (CompressMethod) based on information #1 to compress the original information and obtain the compressed text.

[0178] Step 403: agent2 generates the corresponding prompt based on the compressed text and the compression algorithm.

[0179] The prompt can correspond to the third piece of information above.

[0180] In step 404, agent2 sends a prompt to the LLM, and the LLM receives the prompt from agent2 accordingly.

[0181] The prompt includes the compressed text and the compression algorithm.

[0182] As an example, a prompt can look like this:

[0183] {

[0184] Template Content: The following data has been compressed using the lossless BPE algorithm. The compressed data is provided below. Please answer the following questions based on this data…

[0185] }

[0186] BPE stands for byte pair encoding.

[0187] As another example, a prompt can look like this:

[0188] {

[0189] "Template Content": The following data has undergone lossy compression, with the compression logic being ["table characters in the Markdown text may be preserved"]. The compressed data is provided below; please answer the following questions based on this data…

[0190] }

[0191] In method 400, a compression type attribute is added to the information exchanged between agent1 and agent2. This attribute indicates the compression method preferred by agent1, allowing agent2 to execute different compression algorithms based on user preferences, rather than a fixed compression method, to achieve the desired operation from agent1. This enables agent2 to flexibly perform compression according to agent1's actual needs, thus enhancing the accuracy of agent1's information transmission.

[0192] The following describes method 300 in detail, taking the example of the second entity providing the first entity with a preferred compression type, the large model being an LLM, and the large model being deployed outside the first entity.

[0193] Figure 5 is a schematic flowchart of the communication method 500 provided in this application.

[0194] In method 500, agent1 can correspond to the second entity mentioned above, agent2 can correspond to the first entity mentioned above, and LLM can correspond to the third entity mentioned above.

[0195] Method 500 may include at least a portion of the following.

[0196] Step 501: agent1 sends information #2 to agent2, and agent2 receives information #2 from agent1.

[0197] Information #2 includes the raw information of the natural language, the compression type, and the compression limit (CompressLimit). The compression type is the one preferred by agent1. The raw information of the natural language can correspond to the first information above, the compression type can correspond to the second information above, and the compression limit can correspond to the length information above.

[0198] For example, information #2 can be as follows:

[0199] Here, CompressType is the compression type, CompressLimit is the compression limit, and content is the original information in natural language.

[0200] Step 502: agent2 selects a suitable compression algorithm (CompressMethod) based on information #2 to compress the original information and obtain the compressed text.

[0201] Step 503: agent2 generates the corresponding prompt based on the compressed text and compression algorithm.

[0202] The prompt can correspond to the third piece of information above.

[0203] In step 504, agent2 sends a prompt to the LLM, and the LLM receives the prompt from agent2 accordingly.

[0204] The prompt includes the compressed text and the compression algorithm.

[0205] As an example, a prompt can look like this:

[0206] {

[0207] Template Content: The following data has been compressed using the lossless BPE algorithm. The compressed data is provided below. Please answer the following questions based on this data…

[0208] }

[0209] BPE stands for byte pair encoding.

[0210] As another example, a prompt can look like this:

[0211] {

[0212] "Template Content": The following data has undergone lossy compression, with the compression logic being ["table characters in the Markdown text may be preserved"]. The compressed data is provided below; please answer the following questions based on this data…

[0213] }

[0214] In method 500, compression type and compression limit attributes are added to the information exchanged between agent1 and agent2. These attributes indicate the compression method preferred by agent1, allowing agent2 to execute different compression algorithms based on user preferences instead of a fixed compression method, thus achieving the desired operation from agent1. This enables agent2 to flexibly perform compression according to agent1's actual needs, helping to improve the accuracy of agent1's information transmission.

[0215] The method embodiments provided by this application have been described in detail above with reference to Figures 1 to 5. The device embodiments of this application will be described below with reference to Figures 6 to 8.

[0216] It is understood that, in order to achieve the functions in the above embodiments, the devices in Figures 6 to 8 include hardware structures and / or software modules corresponding to each function. These devices can be used to implement the functions of the first entity, the second entity, or the third entity in the above method embodiments, and thus can also achieve the beneficial effects of the above method embodiments. Those skilled in the art should readily recognize that, based on the units and method steps of the various examples described in conjunction with the embodiments disclosed in this application, this application can be implemented in hardware or a combination of hardware and computer software.

[0217] Figure 6 is a schematic diagram of a device provided in an embodiment of this application.

[0218] This application embodiment can divide the first entity, second entity, or third entity into functional units according to the above method examples. For example, each function can be divided into different functional units, or two or more functions can be integrated into one unit. Each function can be implemented in hardware or as a software functional module. It should be noted that the division shown in Figure 6 is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0219] As shown in Figure 6, the device 10 includes a transceiver unit 11 and a processing unit 12. Optionally, the device 10 also includes a storage unit 13 for storing instructions and / or data.

[0220] When device 10 is used to implement the function of the first entity in the above method embodiments, transceiver unit 11 is used to execute the transceiver steps of the first entity, such as steps 301, 303, 304, 401, 404, 501, or 504, and processing unit 12 is used to execute the processing steps of the first entity, such as steps 302, 402, 403, 502, or 503. When device 10 is used to implement the function of the second entity in the above method embodiments, transceiver unit 11 is used to execute the transceiver steps of the second entity, such as steps 301, 401, or 501, and processing unit 12 is used to execute the processing steps of the second entity. When device 10 is used to implement the function of the third entity in the above method embodiments, transceiver unit 11 is used to execute the transceiver steps of the third entity, such as steps 303, 304, 404, or 504, and processing unit 12 is used to execute the processing steps of the third entity.

[0221] For a more detailed description of the transceiver unit 11 and the processing unit 12, please refer to the relevant descriptions in the above method embodiments, which will not be repeated here.

[0222] Figure 7 is another structural schematic diagram of the device provided in an embodiment of this application.

[0223] The device 20 includes a processing circuit 21. The processing circuit 21 is coupled to a memory 23, which stores instructions. When the device 20 is used to implement the method described above, the processing circuit 21 executes the instructions in the memory 23 to implement the function of the processing unit 12 described above.

[0224] Optionally, the device 20 further includes a memory 23 for implementing the functions of the aforementioned storage unit 13.

[0225] Optionally, the device 20 further includes a transceiver circuit 22. The transceiver circuit can be referred to as a communication interface. The processing circuit 21 and the transceiver circuit 22 are coupled to each other. It is understood that the transceiver circuit 22 can be a transceiver or an input / output interface. When the device 20 is used to implement the method described above, the processing circuit 21 executes instructions to implement the function of the processing unit 12, and the transceiver circuit 22 implements the function of the transceiver unit 11.

[0226] Optionally, device 20 can be a first entity, a second entity, or a third entity, and correspondingly, the transceiver circuit can be a transceiver.

[0227] Optionally, the device 20 can be a chip applied to a first entity, a second entity, or a third entity, and correspondingly, the transceiver circuit can be an input / output interface.

[0228] For example, when device 20 is a chip applied to a first entity, a second entity, or a third entity, the chip implements the functions of the first entity, the second entity, or the third entity in the above method embodiments. The chip receives information from other modules (such as radio frequency modules or antennas) in the first entity, the second entity, or the third entity, which is sent to the first entity, the second entity, or the third entity by other devices; or, the chip sends information to other modules (such as radio frequency modules or antennas) in the first entity, the second entity, or the third entity, which is sent to other devices by the first entity, the second entity, or the third entity.

[0229] Figure 8 is a schematic diagram of a chip system provided in an embodiment of this application. The chip system 30 (or may also be called a processing system) includes logic circuitry 31 and an input / output interface 32.

[0230] The logic circuit 31 can be a processing circuit in the chip system 30. The logic circuit 31 can be coupled to a memory unit, calling instructions from the memory unit, enabling the chip system 30 to implement the methods and functions of the embodiments of this application. The input / output interface 32 can be an input / output circuit in the chip system 30, outputting processed information from the chip system 30, or inputting data or signaling information to be processed into the chip system 30 for processing.

[0231] As an alternative, the chip system 30 may also include a memory unit.

[0232] As one approach, the chip system 30 is used to implement the operations performed by the first entity, the second entity, or the third entity in the various method embodiments described above.

[0233] For example, logic circuit 31 is used to implement processing-related operations performed by the first entity, second entity, or third entity in the above method embodiments; input / output interface 32 is used to implement sending and / or receiving-related operations performed by the first entity, second entity, or third entity in the above method embodiments.

[0234] This application also provides a communication device including a processing circuit coupled to a memory for storing computer programs or instructions and / or data. The processing circuit is used to execute the computer programs or instructions stored in the memory, or to read the data stored in the memory, to perform the methods in the above-described method embodiments. Optionally, the processing circuit may be one or more. Optionally, the communication device includes a memory. Optionally, the memory may be one or more. Optionally, the memory may be integrated with the processing circuit, or may be separately disposed.

[0235] This application also provides a chip including a processing circuit coupled to a memory for storing computer programs or instructions. The processing circuit executes the computer programs or instructions stored in the memory to implement the methods performed by the first entity, second entity, or third entity in the above-described method embodiments. The memory may be located within the chip or independently of the chip, located outside the chip; this is not limited thereto.

[0236] This application also provides a computer-readable storage medium having stored thereon computer instructions for implementing the methods executed by the first entity, the second entity, or the third entity in the above-described method embodiments.

[0237] This application also provides a computer program product comprising instructions which, when executed by a computer, implement the methods performed by the first entity, the second entity, or the third entity in the above-described method embodiments.

[0238] This application also provides a computer program that, when executed by a computer, implements the methods performed by the first entity, the second entity, or the third entity in the above-described method embodiments.

[0239] This application also provides a communication system that includes at least one of the first entity, second entity, or third entity in the embodiments described above.

[0240] The explanations and beneficial effects of the relevant contents in any of the devices provided above can be found in the corresponding method embodiments provided above, and will not be repeated here.

[0241] It is understood that the processing circuit in the embodiments of this application may be a processor or a circuit in a processor for performing processing operations. The processor may include one or more of the following: a central processing unit (CPU), a digital signal processor (DSP), a microprocessor unit (MPU), a microcontroller unit (MCU), a graphics processing unit (GPU), a field programmable gate array (FPGA), an artificial intelligence processor (AI processor), or a neural processing unit (NPU).

[0242] The aforementioned memory may include one or more of the following storage media: random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), phase-change memory (PCM), resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), cache, register, read-only memory (ROM), flash memory, erasable programmable read-only memory (EPROM), hard disk, etc. In one example, computer program instructions for executing the above embodiments may be stored in non-volatile memory, such as the aforementioned memory 23 or at least a portion of the storage cells (e.g., one or more of ROM, flash memory, EPROM, or hard disk).

[0243] The method steps in the embodiments of this application 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, flash memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, registers, hard disks, portable hard disks, compact disc read-only memory (CD-ROM), or any other form of storage medium well known in the art. An exemplary storage medium is coupled to a 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 storage medium can reside in an application-specific integrated circuit (ASIC). Furthermore, the ASIC can reside in a first entity, a second entity, or a third entity. Alternatively, the processor and storage medium can exist as discrete components in the first entity, the second entity, or the third entity.

[0244] 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 in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed on a computer, the processes or functions described in the embodiments of this application are performed entirely or partially. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user equipment, or other programmable device. The computer program or instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless 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, such as a floppy disk, hard disk, or magnetic tape; it can also be an optical medium, such as a digital video optical disc; or it can be a semiconductor medium, such as a solid-state drive.

[0245] In the various embodiments of this application, unless otherwise specified or in case of logical conflict, the terminology and / or descriptions of different embodiments are consistent and can be referenced by each other. The technical features of different embodiments can be combined to form new embodiments according to their inherent logical relationship.

[0246] Unless otherwise stated, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this application is for the purpose of describing specific embodiments only and is not intended to limit the scope of this application. It should be understood that the above are illustrative examples, and the examples above are merely to help those skilled in the art understand the embodiments of this application, and are not intended to limit the embodiments of the application to the specific numerical values ​​or specific scenarios exemplified. Those skilled in the art can obviously make various equivalent modifications or variations based on the examples given above, and such modifications and variations also fall within the scope of the embodiments of this application.

Claims

1. A communication method characterized by comprising: The method is applied to a first entity or a unit within the first entity, and the method includes: Receive first information and second information from the second entity, wherein the first information is the input information of the large model and the second information is used to indicate the second entity's requirement for the output information of the large model; Based on the second information, the first information is compressed to obtain the third information.

2. The method according to claim 1, characterized in that, The second information includes the compression type preferred by the second entity; The compression type is either a first compression type or a second compression type. The first compression type supports ensuring the quality of the output information, while the second compression type supports ensuring the length of the output information.

3. The method according to claim 2, characterized in that, The second information also includes length information, which is used to indicate the length range of the output information.

4. The method according to any one of claims 1 to 3, characterized in that, The step of compressing the first information based on the second information to obtain the third information includes: Based on the second information, the first compression algorithm is determined; The first information is compressed using the first compression algorithm to obtain the third information.

5. The method according to any one of claims 1 to 4, characterized in that, The method further includes: The third information is sent to a third entity, which includes the large model.

6. The method according to claim 5, characterized in that, The first entity and the third entity are located in the same physical entity; or, The first entity and the third entity are located in different physical entities.

7. The method according to claim 5 or 6, characterized in that, The method further includes: A fourth message is sent to the third entity, the fourth message being used to instruct the first compression algorithm, the third message being obtained based on the first compression algorithm.

8. The method according to any one of claims 1 to 7, characterized in that, Before compressing the first information based on the second information, the method further includes: It is determined that the length of the first information exceeds the context window of the large model.

9. The method according to any one of claims 1 to 8, characterized in that, The first information is a first prompt, and the third information is a second prompt; or, The first information is the text to be compressed, and the third information is the third prompt.

10. A communication method characterized by comprising: The method is applied to a second entity or a unit within the second entity, and the method includes: Determine the first information and the second information, where the first information is the input information of the large model, and the second information is used to indicate the second entity's demand for the output information of the large model; Send the first information and the second information to the first entity.

11. The method according to claim 10, characterized in that, The second information includes the compression type preferred by the second entity; The compression type is either a first compression type or a second compression type. The first compression type supports ensuring the quality of the output information, while the second compression type supports ensuring the length of the output information.

12. The method according to claim 11, characterized in that, The second information also includes length information, which is used to indicate the length range of the output information.

13. The method according to any one of claims 10 to 12, characterized in that, The first information is a first prompt; or, the first information is text to be compressed.

14. A communication method, comprising: The method is applied to a third entity or a unit within the third entity, the third entity comprising a large model, and the method includes: Receive third and fourth information from the first entity, wherein the third information is the input information of the large model and is compressed information, and the fourth information is used to indicate the first compression algorithm, and the third information is obtained based on the first compression algorithm; Based on the third and fourth information, the first entity interacts with the second entity.

15. A method of communication, comprising: The method includes: The second entity sends a first message and a second message, wherein the first message is the input information of the large model, and the second message is used to indicate the second entity's requirement for the output information of the large model; The first entity receives the first information and the second information; The first entity compresses the first information based on the second information to obtain the third information.

16. The method of claim 15, wherein, The method further includes: The first entity sends the third information to the third entity, the third entity including the large model; The third entity receives the third information.

17. A communications device, characterized by The apparatus includes modules or units for implementing the method as described in any one of claims 1 to 16.

18. A communications device, characterized by The device includes a processor and an interface circuit, the interface circuit being used to receive signals from other devices besides the communication device and transmit them to the processor, or to send signals from the processor to other devices besides the communication device, the processor being used to implement the method as described in any one of claims 1 to 16 via logic circuits or executing code instructions.

19. The communication apparatus according to claim 18, wherein The communication device is a chip or chip system.

20. A computer-readable storage medium, characterized in that, The storage medium stores a computer program or instructions, which, when executed by a communication device, implement the method as described in any one of claims 1 to 16.

21. A communication system, characterized by It includes at least one of the following devices: a communication device for performing the method as claimed in any one of claims 1 to 9, a communication device for performing the method as claimed in any one of claims 10 to 13, or a communication device for performing the method as claimed in claim 14.

22. A computer program product, characterised in that, It includes a computer program or instructions that, when executed in a computer, implement the steps of the method as described in any one of claims 1 to 16.

Images