WO2025108271A1 - 通信系统中的模型推理方法、装置、设备以及介质 - Google Patents
通信系统中的模型推理方法、装置、设备以及介质 Download PDFInfo
- Publication number
- WO2025108271A1 WO2025108271A1 PCT/CN2024/133006 CN2024133006W WO2025108271A1 WO 2025108271 A1 WO2025108271 A1 WO 2025108271A1 CN 2024133006 W CN2024133006 W CN 2024133006W WO 2025108271 A1 WO2025108271 A1 WO 2025108271A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- information
- model
- knowledge
- calling interface
- indication
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/06—Testing, supervising or monitoring using simulated traffic
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/10—Scheduling measurement reports ; Arrangements for measurement reports
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/20—Services signaling; Auxiliary data signalling, i.e. transmitting data via a non-traffic channel
Definitions
- the present application belongs to the field of communication technology, and specifically relates to a model reasoning method, device, equipment and medium in a communication system.
- AI artificial intelligence
- CSI channel state information
- AI-based beam management AI-based positioning
- AI-based positioning etc.
- large models usually perform better in solving general problems, but there are many different problems in communication systems, such as beamforming, resource allocation, channel prediction, etc. Therefore, how to ensure the performance of large models in solving specific downstream problems is an urgent problem to be solved.
- the embodiments of the present application provide a model reasoning method, apparatus, device and medium in a communication system, which can improve the performance of the model.
- a model reasoning method in a communication system comprising:
- a first device receives first information sent by a second device, wherein at least one of knowledge information and calling interface information of the knowledge information is stored on the first device, wherein the knowledge information includes at least one of the following: a knowledge graph and a knowledge vector library;
- the first device sends second information to a third device, the first model is deployed on the third device, and the second information is determined according to the first information and knowledge information stored or called on the first device;
- the first information includes at least one of the following:
- a model reasoning method in a communication system comprising:
- the second device sends the first information to the first device, wherein the first device stores at least one of knowledge information and calling interface information of the knowledge information, wherein the knowledge information includes at least one of the following: a knowledge graph and a knowledge vector library;
- the first information includes at least one of the following:
- a model reasoning method in a communication system including:
- the third device obtains the second information from the first device or the target device, wherein the first model is deployed on the third device, and at least one of knowledge information and calling interface information of knowledge information is stored on the first device or the target device, wherein the target device is a device accessed using the calling interface information, and the knowledge information includes at least one of the following: a knowledge graph and a knowledge vector library;
- the second information is determined according to the first information and the knowledge information stored or called on the first device;
- the first information includes at least one of the following:
- a communication device including:
- a communication unit configured to receive first information sent by a second device, wherein the communication device stores at least one of knowledge information and calling interface information of the knowledge information, wherein the knowledge information includes at least one of the following: a knowledge graph, a knowledge vector library; and
- the first information includes at least one of the following:
- a communication device including:
- a communication unit configured to send first information to a first device, wherein the first device stores at least one of knowledge information and calling interface information of the knowledge information, wherein the knowledge information includes at least one of the following: a knowledge graph and a knowledge vector library;
- the first information includes at least one of the following:
- a communication device including:
- a communication unit configured to obtain second information from a first device or a target device, wherein the first model is deployed on the third device, and at least one of knowledge information and calling interface information of knowledge information is stored on the first device or the target device, wherein the target device is a device accessed using the calling interface information, and the knowledge information includes at least one of the following: a knowledge graph and a knowledge vector library;
- the second information is determined according to the first information and the knowledge information stored or called on the first device;
- the first information includes at least one of the following:
- a communication device which network side device includes a processor and a memory, the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the steps of the method described in any one of the first aspect to the third aspect are implemented.
- a readable storage medium on which a program or instruction is stored.
- the program or instruction is executed by a processor, the steps of the method described in any one of the first to third aspects are implemented.
- a wireless communication system comprising: a first device, a second device and a third device, wherein the first device can be used to execute the steps of the method described in the first aspect, the second device can be used to execute the steps of the method described in the second aspect, and the third device can be used to execute the steps of the method described in the third aspect.
- a chip comprising a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run a program or instruction to implement the method as described in any one of the first to third aspects.
- a computer program/program product is provided, wherein the computer program/program product is stored in a storage medium, and the program/program product is executed by at least one processor to implement the method as described in any one of the first to third aspects.
- the second device can assist the third device in obtaining the second information by sending the first information to the first device. Furthermore, the third device can use the second information to assist in model reasoning, which is beneficial to improving the reasoning performance of the model.
- FIG1 is a schematic diagram of a communication system provided in an embodiment of the present application.
- FIG2 is a schematic diagram of a model reasoning method in a communication system provided in an embodiment of the present application.
- FIG. 3 is a schematic diagram of a hidden variable provided in an embodiment of the present application.
- Figures 4 to 14 are schematic interaction diagrams of the model reasoning method provided in the embodiments of the present application.
- FIG15 is a schematic diagram of a communication device provided in an embodiment of the present application.
- FIG16 is a schematic diagram of another communication device provided in an embodiment of the present application.
- FIG17 is a schematic diagram of another communication device provided in an embodiment of the present application.
- FIG18 is a schematic diagram of a communication device provided in an embodiment of the present application.
- FIG. 19 is a hardware structure diagram of a terminal provided in an embodiment of the present application.
- Figure 20 is a hardware structure diagram of a network side device provided in an embodiment of the present application.
- Figure 21 is a hardware structure diagram of another network side device provided in an embodiment of the present application.
- first, second, etc. of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way are interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by “first” and “second” are generally of one type, and the number of objects is not limited, for example, the first object can be one or more.
- “or” in the present application represents at least one of the connected objects.
- “A or B” covers three schemes, namely, Scheme 1: including A but not including B; Scheme 2: including B but not including A; Scheme 3: including both A and B.
- the character "/" generally indicates that the objects associated with each other are in an "or” relationship.
- indication in this application can be a direct indication (or explicit indication) or an indirect indication (or implicit indication).
- a direct indication can be understood as the sender explicitly informing the receiver of specific information, operations to be performed, or request results in the sent indication;
- an indirect indication can be understood as the receiver determining the corresponding information according to the indication sent by the sender, or making a judgment and determining the operation to be performed or the request result according to the judgment result.
- LTE Long Term Evolution
- LTE-A Long Term Evolution
- CDMA Code Division Multiple Access
- TDMA Time Division Multiple Access
- FDMA Frequency Division Multiple Access
- OFDMA Orthogonal Frequency Division Multiple Access
- SC-FDMA Single-carrier Frequency Division Multiple Access
- NR New Radio
- 6G 6th Generation
- FIG1 shows a block diagram of a wireless communication system applicable to the embodiment of the present application.
- the wireless communication system includes a terminal 11 and a network side device 12 .
- the terminal 11 can be a mobile phone, a tablet computer (Tablet Personal Computer), a laptop computer (Laptop Computer), a notebook computer, a personal digital assistant (PDA), a handheld computer, a netbook, an ultra-mobile personal computer (Ultra-mobile Personal Computer, UMPC), a mobile Internet device (Mobile Internet Device, MID), an augmented reality (Augmented Reality, AR), a virtual reality (Virtual Reality, VR) device, a robot, a wearable device (Wearable Device), a flight vehicle (flight vehicle), a vehicle user equipment (VUE), a shipborne equipment, a pedestrian terminal (Pedestrian User Equipment, PUE), a smart home (home appliances with wireless communication functions, such as refrigerators, televisions, washing machines or furniture, etc.), a game console, a personal computer (Personal Computer, PC
- Wearable devices include: smart watches, smart bracelets, smart headphones, smart glasses, smart jewelry (smart bracelets, smart bracelets, smart rings, smart necklaces, smart anklets, smart anklets, etc.), smart wristbands, smart clothing, etc.
- the vehicle-mounted device can also be called a vehicle-mounted terminal, a vehicle-mounted controller, a vehicle-mounted module, a vehicle-mounted component, a vehicle-mounted chip or a vehicle-mounted unit, etc. It should be noted that the specific type of the terminal 11 is not limited in the embodiment of the present application.
- the terminal can also be called user equipment (UE), terminal equipment, access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication equipment, user agent or user device, etc.
- UE user equipment
- terminal equipment access terminal
- user unit user station
- mobile station mobile station
- remote station remote terminal
- mobile device user terminal
- terminal wireless communication equipment
- user agent or user device etc.
- the network side equipment 12 may include access network equipment or core network equipment, wherein the access network equipment may also be referred to as radio access network (RAN) equipment, radio access network function or radio access network unit.
- the access network equipment may include a base station, a wireless local area network (WLAN) access point (AS) or a wireless fidelity (WiFi) node, etc.
- WLAN wireless local area network
- AS wireless local area network
- WiFi wireless fidelity
- the base station may be referred to as a node B (NB), an evolved node B (eNB), the next generation node B (gNB), a new radio node B (NR Node B), an access point, a relay station (RBS), a serving base station (SBS), a base transceiver station (BTS), a radio base station, a radio transceiver, a base Basic Service Set (BSS), Extended Service Set (ESS), home Node B (HNB), home evolved Node B (home evolved Node B), Transmission Reception Point (TRP) or other appropriate term in the field, as long as the same technical effect is achieved, the base station is not limited to specific technical vocabulary. It should be noted that, in the embodiments of the present application, only the base station in the NR system is taken as an example for introduction, and the specific type of the base station is not limited.
- the core network equipment may include but is not limited to at least one of the following: core network nodes, core network functions, mobility management entity (Mobility Management Entity, MME), access mobility management function (Access and Mobility Management Function, AMF), session management function (Session Management Function, SMF), user plane function (User Plane Function, UPF), policy control function (Policy Control Function, PCF), policy and charging rules function unit (Policy and Charging Rules Function, PCRF), edge application service discovery function (Edge Application Server Discovery, Function, EASDF), Unified Data Management (UDM), Unified Data Repository (UDR), Home Subscriber Server (HSS), Centralized network configuration (CNC), Network Repository Function (NRF), Network Exposure Function (NEF), Local NEF (L-NEF), Binding Support Function (BSF), Application Function (AF), etc.
- MME mobility management entity
- AMF Access Mobility Management Function
- SMF Session Management Function
- SMF Session Management Function
- UPF User Plane Function
- Policy Control Function Policy
- the specific type of the core network device is not limited. But not limited to at least one of the following: core network node, core network function, Mobility Management Entity (MME), Access and Mobility Management Function (AMF), Session Management Function (SMF), User Plane Function (UPF), Policy Control Function (PCF), Policy and Charging Rules Function (PCRF), Edge Application Server Discovery Function (Edge Application Server Discovery Function,
- EASDF Unified Data Management
- UDM Unified Data Repository
- USR Unified Data Repository
- HSS Home Subscriber Server
- CNC Centralized network configuration
- NRF Network Repository Function
- NEF Network Exposure Function
- L-NEF Binding Support Function
- BF Binding Support Function
- AF Application Function
- Knowledge graph is a knowledge base that uses graph structure or topology to represent and integrate data. It can store descriptions of associations between entities (objects, events, situations or abstract concepts) and can also encode semantic relationships between entities.
- the application prospects of knowledge graph are very broad. It can not only improve the effectiveness of data services such as information retrieval, search engines, and recommendation systems, but also support intelligent interactions such as natural language question and answer, dialogue, and reasoning. It has the characteristics of strong representation ability, high flexibility, strong computability, and strong cross-domain.
- Knowledge graphs store a large amount of knowledge in an explicit and structured way, which can be used to enhance the knowledge awareness of large models. Incorporating knowledge graphs into large models during the reasoning phase can significantly improve the performance of large models in accessing domain-specific knowledge by retrieving knowledge from knowledge graphs.
- network data knowledge graphs are mainly used for knowledge representation, relationship analysis and deep mining, providing effective knowledge rules and knowledge computing support for the intelligentization of communication systems.
- the network structure, terminal type, terminal behavior, data service requirements, and system resources of the communication system are all highly dynamic, time-sensitive, and mutually coupled.
- Mobile communication data faces many challenges, such as the difficulty in obtaining scattered data, the wide variety of data, the complex structure, and the difficulty in mining complex associations.
- knowledge graphs can effectively clarify the various relationships between data fields and communication network indicators, and further in-depth mining can be carried out based on the established relationships, such as quantifying the degree of correlation between relationships, characterizing the characteristic attributes of data fields and indicators, etc.
- the knowledge graph can be represented by the correlation between entities.
- One representation method can be:
- Entity 1 Correlation coefficient between entity 1 and entity 2, Entity 2).
- entity 1 is the cell throughput
- entity 2 is the L1-RSRP of the strongest beam
- the correlation coefficient is 0.5.
- the knowledge graph can be represented by a subject attribute object (Subject, Predicate, Object, SPO) triple, for example, an SPO triple of (cell id, shaped codebook, codebook indication).
- Subject Predicate, Object, SPO
- SPO SPO triple
- the knowledge vector library is a database used to store, retrieve, and analyze vectors. It is called a database because it has the following characteristics:
- the knowledge vector library can be simply understood as a database used to store model input feature vectors.
- the knowledge vector library can be a model, and the input is a picture or text.
- the input of the model can be the measurement result of a wireless signal, etc.
- the output is a feature vector used to characterize the measurement result.
- the processed data is input into the model together with the original model input, or the processed data is directly input into the model.
- Prompt engineering is a technique in Natural Language Processing (NLP) that can create text snippets, prompts, or templates to guide pre-trained large language models to produce high-quality output for specific tasks or applications. It is widely used in fields such as question answering, summarization, translation, sentiment analysis, and text generation. Prompt engineering can leverage the powerful capabilities of pre-trained large language models to implement a variety of complex natural language processing tasks, reduce dependence on labeled data and model fine-tuning, reduce development costs and time, and improve the interpretability and controllability of pre-trained large language models, increasing user trust and satisfaction. Common prompts can include the following: Zero-Shot Prompting, Few-Shot Prompting, Chain of Thought Prompting, etc.
- AI artificial intelligence
- CSI channel state information
- AI-based beam management AI-based beam management
- AI-based positioning etc.
- AI-based energy saving and AI-based load balancing is also considered.
- future use cases there will be more use cases combined with AI in mobile communication systems.
- Question 1 The size of the large model is very large. How to deploy the large model to devices with limited storage resources such as base stations or terminals for inference? A typical method is to process the large model through compression methods such as quantization and pruning, and then deploy it to the communication equipment.
- Problem 2 Large models are often used to solve general problems, but there are many different problems in communications, such as beamforming, resource allocation, and channel prediction. There is currently no mature solution in the communication system to match large models to specific downstream problems.
- FIG2 shows a schematic diagram of a model reasoning method in a communication system according to an embodiment of the present application.
- the method 200 includes:
- a first device receives first information sent by a second device, wherein at least one of knowledge information and calling interface information of the knowledge information is stored on the first device;
- the first device sends second information to a third device, a first model is deployed on the third device, and the second information is determined based on the first information and knowledge information.
- the first model is a compressed AI model.
- the first model is an AI model that is compressed from a large model.
- the compression here may include but is not limited to quantization, pruning (for example, deleting some layers in the model, or deleting some nodes in a layer), etc.
- the AI model may also be referred to as an AI unit, an ML (machine learning) model, an ML unit, an AI structure, an AI function, an AI characteristic, a machine learning model, a neural network, a neural network function, a neural network function, etc., or, the AI model may also refer to a processing unit that can implement specific algorithms, formulas, processing flows, capabilities, etc.
- the AI model may be a processing method, algorithm, function, module or unit for a specific data set, or the AI model may be a processing method, algorithm, function, module or unit running on AI/ML related hardware such as a graphics processing unit (GPU), a neural network processing unit (NPU), a tensor processing unit (TPU), an application specific integrated circuit (ASIC), etc., and the present application does not make any specific limitations on this.
- GPU graphics processing unit
- NPU neural network processing unit
- TPU tensor processing unit
- ASIC application specific integrated circuit
- the first device can be considered as a device for storing knowledge-related information, or a knowledge storage device.
- the first device stores knowledge information, calling interface information of the knowledge information, etc.
- the second device can be considered as a demand initiating device, such as a device for initiating reasoning requirements.
- the third device can be considered as an inference device, or an AI model deployment device. Among them, the second device and the third device can be the same device, or they can be different devices.
- the knowledge information includes but is not limited to at least one of a knowledge graph and a knowledge vector library.
- the third device may utilize the second information to expand the input of the first model to assist the first model in performing accurate reasoning.
- the second information and the input information of the first model can be used, so that the model can obtain more input information and improve the reasoning performance of the model.
- the first device is an access network device (e.g., a base station) or a core network function, such as a core network function with a data storage function, such as a database, a data function, or a network storage function (Network Repository Function, NRF) or UDM, or an AI model library, or an AI model management function or a third-party server of a terminal, etc.
- the second device is a terminal
- the third device is a terminal.
- the first device is a core network function, such as a core network function with a data storage function, such as a database, a data function or a network repository function (NRF) or UDM, or an AI model library, or an AI model management function or a third-party server of a terminal, etc.
- the second device is a terminal or an access network device (such as a base station)
- the third device is an access network device (such as a base station).
- the first device is a core network function, such as a core network function with a data storage function, such as a database, a data function or a network repository function (NRF) or UDM, or an AI model library, or an AI model management function or a third-party server of a terminal, etc.
- the second device is a terminal or an access network device (such as a base station), a third-party server, a core network function (such as an AI control function, a task control function, a collaborative control function or other core network functions, etc.)
- the third device is the reasoning function of the core network.
- the AI control node can be used to control AI-related functions
- the task control node can be used to control service-related functions, or task-related control functions.
- the task may include but is not limited to new network capabilities involving the coordination and deployment of connections, computing, data and algorithm resources in multi-node scenarios to jointly achieve a specific goal.
- the collaborative control node can be used for collaborative management between multiple functions (such as communication functions, data functions, computing power functions, algorithm functions, and model functions).
- the information exchange between the terminal and the access network device may be through at least one of the following signaling: layer 1 signaling, layer 2 signaling, layer 3 signaling, data plane signaling, AI layer signaling.
- the information exchange between the terminal and the core network device may be through at least one of the following signaling: NAS signaling, data plane signaling, AI layer signaling.
- the terminal and the core network device may directly exchange information, or the information may be forwarded through other devices, such as through communication control functions (such as access mobility management function AMF), AI control functions, task control functions, collaborative control functions, network open functions, etc.
- access network devices and core network devices can directly interact with each other, or information can be forwarded through other devices, such as through communication control functions (such as access mobility management function AMF), AI control functions, task control functions, collaborative control functions, network open functions, etc.
- communication control functions such as access mobility management function AMF
- AI control functions such as access mobility management function AMF
- task control functions such as access mobility management function AMF
- collaborative control functions such as network open functions, etc.
- the first information is carried in at least one of the following signaling: layer 1 signaling, layer 2 signaling, layer 3 signaling, data plane signaling, and AI layer signaling.
- the layer 1 signaling may include but is not limited to PDCCH.
- the layer 2 signaling may include but is not limited to downlink MAC CE, and the layer 3 signaling may include but is not limited to RRC signaling.
- the first device is a core network function
- the second device is a terminal
- the first information is carried in at least one of the following signaling: NAS signaling, data plane signaling, and AI layer signaling.
- the second information is carried in at least one of the following signaling: layer 1 signaling, layer 2 signaling, layer 3 signaling, data plane signaling, and AI layer signaling.
- the layer 1 signaling may include but is not limited to PDCCH.
- the layer 2 signaling may include but is not limited to downlink MAC CE, and the layer 3 signaling may include but is not limited to RRC signaling.
- the third device is a terminal
- the second information is carried in at least one of the following signaling: NAS signaling, data plane signaling, and AI layer signaling.
- the first device may also send the first information to the third device to assist the third device in model reasoning.
- the first information may be carried in the second information, such as the second information includes the content in the first information.
- the first information includes at least one of the following:
- the demand initiating device indicates the relevant information (such as the first information) used for model reasoning to the knowledge storage device, and the knowledge storage device can determine the second information based on the first information and the knowledge information.
- the knowledge information of the professional field is obtained according to the first information, thereby obtaining the second information, and further sending the second information to the reasoning device.
- the reasoning device uses the second information to assist in model reasoning, which can achieve accurate reasoning of professional field tasks, reduce the illusion problem in model reasoning, and improve the reasoning performance of the model.
- the receiving device indication may be identification information of the receiving device of the second device, which is used by the first device to determine to which device the second information is to be sent.
- the sample indication may be which sample the input information of the first model in the first information corresponds to, so as to avoid the inference device using the first information and the second information corresponding to different samples to perform model inference, thereby affecting the inference result.
- the sample indication may be a sample identifier.
- the auxiliary information used for reasoning with the first model includes at least one of the following:
- AI model identification information AI model function indication, AI model characteristic indication, target knowledge graph indication, target knowledge vector library indication, cell identification, and bandwidth part (Band Width Part, BWP) identification.
- the identification information of the AI model may be, for example, a model identifier (model ID) of the AI model, which may be used for AI structure identification, AI algorithm identification, or identification of a specific data set associated with the AI model, or identification of specific scenarios, environments, channel characteristics, and devices related to AI/ML, or identification of functions, features, capabilities, or modules related to AI/ML.
- model ID model identifier
- the characteristic indication of the AI model may explicitly or implicitly indicate the characteristics supported by the AI model, such as supporting CSI prediction and compression feedback, beam prediction, positioning, load balancing, resource allocation, etc.
- the first device can be assisted in obtaining knowledge information related to the characteristics, and then corresponding second information can be obtained based on the knowledge information to assist the third device in reasoning the model.
- the characteristics of the AI model under a specific configuration may be considered to be the functions of the AI model.
- the characteristic of the AI model is beam prediction
- the function of the AI model is time domain beam prediction when the base station is configured with 32 transmit beams.
- the function indication of the AI model may explicitly or implicitly indicate the functions supported by the AI model, such as supporting CSI prediction and compression feedback, beam prediction, positioning, load balancing, resource allocation, etc. under a specific configuration.
- the first device can be assisted in obtaining knowledge information related to the function, and then corresponding second information can be obtained based on the knowledge information to assist the third device in reasoning the model.
- the target knowledge graph indication can be used to indicate the knowledge graph that the second device expects to use.
- the first device can be assisted in using the target knowledge graph to obtain corresponding second information for assisting the third device in reasoning about the model.
- the target knowledge graph indication may explicitly indicate the target knowledge graph.
- the target knowledge graph indicates at least one of the following:
- the subject attribute object (Subject, Predicate, Object, SPO) triple of the sentence in the target knowledge graph;
- the relationship between entities in the target knowledge graph can be expressed in the following format:
- the relationship between entities may include but is not limited to hierarchical relationship, inheritance relationship, etc.
- the attribute information of an entity in the target knowledge graph may be represented in the following format:
- the relationship between nodes can be represented by a triple, for example, a relationship can be (node 1, edge, node 2).
- the correlation between entities in the target knowledge graph can be expressed in the following format:
- Entity 1 Correlation coefficient between entity 1 and entity 2, Entity 2).
- entity 1 is the cell throughput
- entity 2 is the L1-RSRP of the strongest beam
- the correlation coefficient is 0.8.
- the SPO triples of a sentence in the target knowledge graph are used to describe the subject, predicate, and object of the sentence.
- the target knowledge graph indication may implicitly indicate the target knowledge graph.
- the target knowledge graph indication includes but is not limited to at least one of the following:
- multiple knowledge graph identifiers can be predefined or preconfigured (for example, pre-indicated by a network-side device), and each knowledge graph identifier corresponds to a knowledge graph.
- the second device can select one or more knowledge graph identifiers from the multiple knowledge graph identifiers and indicate them to the first device.
- the first device can determine the corresponding knowledge graph based on the indicated knowledge graph identifier, and further determine the second information based on the knowledge graph.
- multiple knowledge graph file addresses can be predefined or preconfigured (for example, pre-indicated by a network side device), and each knowledge graph file address corresponds to a knowledge graph file.
- the second device can select one or more knowledge graph file addresses from the multiple knowledge graph file addresses and indicate them to the first device.
- the first device can obtain the corresponding knowledge graph file according to the indicated knowledge graph file address, and further determine the second information based on the knowledge graph file.
- the knowledge graph may be used in the following ways:
- the knowledge graph can be used to retrieve the cell transmit beam configuration, such as the direction or shaping codebook of the transmit beam.
- the target knowledge vector library indication may implicitly indicate the target knowledge vector library that the second device expects to use. By indicating the expected target knowledge vector library to the first device, the first device may be assisted in using the target knowledge vector library to obtain corresponding second information for assisting the third device in reasoning about the model.
- the target knowledge vector library indicates at least one of the following:
- multiple knowledge vector library identifiers can be predefined or preconfigured (for example, pre-indicated by a network side device), and each knowledge vector library identifier corresponds to a knowledge vector library.
- the second device can select one or more knowledge vector library identifiers from the multiple knowledge vector library identifiers to indicate to the first device.
- the first device can determine the corresponding knowledge vector library based on the indicated knowledge vector library identifier, and further determine the second information based on the knowledge vector library.
- multiple knowledge vector library file addresses can be predefined or preconfigured (for example, pre-indicated by a network side device), and each knowledge vector library file address corresponds to a knowledge vector library file.
- the second device can select one or more knowledge vector library file addresses from the multiple knowledge vector library file addresses and indicate them to the first device.
- the first device can obtain the corresponding knowledge vector library file according to the indicated knowledge vector library file address, and further determine the second information based on the knowledge vector library file.
- the method for using the knowledge vector library may be:
- the knowledge vector library is used to obtain the cell transmission beam configuration, such as the direction of the transmission beam or the feature vector related to the shaped codebook.
- the cell identifier may be used by the first device to obtain knowledge information of the cell corresponding to the cell identifier, such as cell configuration, etc. Further, the second information may be determined based on the knowledge information.
- the BWP identifier may be used by the first device to obtain knowledge information of the BWP corresponding to the BWP identifier, such as BWP configuration, etc. Further, the second information may be determined based on the knowledge information.
- the requirement information for the first model to perform reasoning includes at least one of the following:
- QoE Service quality of experience
- the first device may select knowledge information related to the target task according to the target task indication, and then determine the second information according to the knowledge information.
- the second device may select knowledge information that meets QoE requirements, or processing delay, processing accuracy, or processing scale based on the QoE requirements, or processing delay, processing accuracy, or processing scale, and then determine the second information based on the knowledge information.
- the input information of the first model is related to the target task that the first model needs to adapt.
- the input information may include a cell identifier and a measurement quality of a beam transmitted by the cell.
- the calling interface information of the knowledge information includes but is not limited to at least one of the following:
- the name of the calling interface, the input data format of the calling interface, the output data format of the calling interface, and the input data of the calling interface are related to the domain, task, scenario, etc.
- the calling interface may include but is not limited to an application programming interface (API) and other interfaces for accessing external devices.
- API application programming interface
- the communication system will be continuously updated and the knowledge of the communication system will also be updated.
- the updated system knowledge such as the latest knowledge from organizations such as 3GPP, can be stored in an external database.
- the API interface by calling the API interface, services are provided to the communication system, which improves the applicability of the technical solution of this application.
- both the second device and the third device are terminals, and the first information may be sent under specific circumstances, for example:
- the second device sends the first information to the first device:
- Event 1 receiving a first instruction, where the first instruction is used to instruct the terminal to activate or switch a model;
- Event 2 the terminal determines to activate or switch the model
- Event 3 the terminal determines that the currently used model does not meet the requirements
- Event 4 A second indication is received, where the second indication is used to indicate that the model currently used by the terminal does not meet the requirements.
- the event that triggers the second device to send the first information to the first device can be predefined, or configured by the network side device.
- the second device can send identification information of the event to the network side device, for example, carried in the first information and sent to the network side device, so that the network side device can be aware of the event occurring on the second device.
- the first indication may be sent by a network side device.
- the network side device may control the activation or switching of the AI model.
- the network side device may send a first indication to the terminal, so that the second device may send the first information according to the first indication.
- the terminal may autonomously determine to activate or switch a model when the performance of the currently used model does not meet the requirements.
- the second indication may be sent by a network side device.
- the network side device may send a second indication to the terminal to indicate that the model currently used by the terminal does not meet the requirements.
- the terminal may independently determine whether to activate or switch the model based on the second indication.
- the model not meeting the requirements may refer to the model not meeting the requirement information, for example, it may include but is not limited to at least one of the following: the model's reasoning does not meet the service quality of experience (Quality of Experience, QoE) does not meet the QoE requirements, the model's service processing delay does not meet the delay requirements, and the model's service processing accuracy does not meet the accuracy requirements.
- QoE Quality of Experience
- the second information includes at least one of the following:
- Hidden variables where the hidden variables are feature vectors after mapping the input information of the first model
- the sample indication corresponding to the input information of the first model, or the sample indication in the first information used to determine the second information is not limited.
- the sample indication can be used to indicate which sample the second information is determined based on, and the sample indication can be a sample identifier.
- the sample indication can be a sample identifier.
- the hidden variable may refer to a low-dimensional feature obtained by mapping input information with physical meaning through an encoder, or a feature vector of the original input after redundancy is removed.
- the autoencoder includes an encoder and a decoder
- the encoder maps the input image to low-dimensional features (hidden variables in FIG3 )
- the decoder restores the input image based on the hidden variables, where the hidden variables contain the feature information of the input image.
- the prompt information can be used to prompt or guide the first model to infer an output that meets expectations, thereby improving the reasoning performance of the model.
- the prompt information includes at least one of the following:
- a prompt text mark related to the target task used to mark a prompt text
- the address of the prompt file related to the target task is the address of the prompt file related to the target task.
- the configuration information of the network side device includes but is not limited to at least one of the following:
- the number of transmitting antennas of the network-side equipment is the number of transmitting antennas of the network-side equipment.
- the number of transmit beams of the network-side device is the number of transmit beams of the network-side device
- the transmission power of the network side equipment is the transmission power of the network side equipment
- the scene type information includes at least one of the following:
- Line of sight LOS
- NLOS non line of sight
- outdoor indoor, urban microcell (UMi), urban macrocell (UMa), rural macrocell (RMa), rural microcell (RMi), high speed, low speed.
- UMi urban microcell
- UMa urban macrocell
- RMa rural macrocell
- RMi high speed, low speed.
- the second information may be determined by the first device, for example, based on the first information and knowledge information stored on the first device, or based on the first information and called knowledge information, for example, the first device may use calling interface information to obtain knowledge information from the target device, and further determine the second information based on the knowledge information, or the first information and the knowledge information.
- the second information may also be determined by the target device that needs to be accessed using the calling interface information.
- the target device may be sent directly to the third device, or it may be sent to the third device through other devices, such as forwarding through a network development function, or it may be sent to the third device through the first device.
- Embodiment 1 The first device stores knowledge information for assisting the first model in reasoning.
- the first device can determine the second information based on the first information combined with the knowledge information stored on the first device. Furthermore, the first device can send the second information to the third device, so that the third device can use the second information as auxiliary information when using the first model for reasoning, which can improve the reasoning performance of the model.
- Embodiment 2 The first device does not store knowledge information for assisting the first model in reasoning, and the first information includes calling interface information of the knowledge information.
- the following steps can be used to implement:
- the second device sends first information to the first device, the first information includes calling interface information of knowledge information, and the calling interface information may be calling interface information of knowledge information that the second device expects to use.
- the calling interface information of knowledge information includes a calling interface name and input data of the calling interface.
- the first device sends first information to a target device according to the calling interface information of the knowledge information, where the target device is a device that needs to be accessed using the calling interface information;
- the target device determines second information according to the first information
- the target device sends second information to the first device
- S215 The first device sends second information to the third device.
- first information and the second information between the first device and the target device can be forwarded through other nodes (such as network open functions), or the information can be directly interacted.
- Embodiment 3 The first device does not store knowledge information for assisting the first model in reasoning, and the first information does not include calling interface information of the knowledge information.
- the second device sends first information to the first device, where the first information does not include calling interface information of the knowledge information.
- S222 The first device determines the calling interface information of the knowledge information according to the first information.
- the first device can determine the target calling interface among multiple calling interfaces based on the requirement information for first model reasoning in the first information, thereby obtaining information such as the calling interface name of the target calling interface, the input data format of the calling interface, the output data format of the calling interface, and the input data of the calling interface.
- the first device sends a third message to the target device, the third message including the calling interface information of the knowledge information.
- the target device is a device that needs to be accessed using the calling interface information.
- the calling interface information of the knowledge information includes a calling interface name and input data of the calling interface. For example, the calling interface name is determined based on the requirement information or auxiliary information or calling interface information in the first message, and the input data of the calling interface is determined based on the input information of the first model in the first message.
- the target device determines the second information according to the third information
- the target device sends second information to the first device
- the first device sends second information to the third device.
- the first device can autonomously determine the calling interface information of the knowledge information, and further send the calling interface information, or the calling interface information and the first information to the target device, so that the target device can obtain the corresponding knowledge information based on the calling interface information, thereby obtaining the second information. Then the target device can send the determined second information to the first device, and the first device sends the second information to the third device.
- interaction between the third information and the second information between the first device and the target device can be forwarded through other nodes (such as network open functions), or the information can be directly interacted.
- the second device sends first information to the first device, where the first information does not include calling interface information of the knowledge information.
- the first device determines the calling interface information of the knowledge information according to the first information.
- the first device can determine the target calling interface among multiple calling interfaces based on the requirement information for first model reasoning in the first information, thereby obtaining information such as the calling interface name of the target calling interface, the input data format of the calling interface, the output data format of the calling interface, and the input data of the calling interface.
- the first device sends fourth information to the third device, the fourth information includes the calling interface information of the knowledge information, or the fourth information includes the first information and the calling interface information of the knowledge information.
- the calling interface information of the knowledge information includes the calling interface name of the target calling interface, or the calling interface name of the target calling interface and the input data format of the calling interface, or the calling interface name of the target calling interface and the input data format of the calling interface and the output data format of the calling interface.
- the third device sends the seventh information to the target device according to the fourth information, wherein the seventh information includes the calling interface information of the knowledge information; wherein the calling interface information of the knowledge information includes the calling interface name and the input data of the calling interface.
- the calling interface name is included in the calling interface information of the knowledge information in the fourth information
- the input data of the calling interface is determined based on the input data format of the calling interface of the knowledge information in the fourth information and the input information of the first model that can be provided by the third device.
- the calling interface name is determined based on the requirement information or auxiliary information or calling interface information in the first information (contained in the fourth information), and the input data of the calling interface is determined based on the input information of the first model in the first information (contained in the fourth information).
- the target device determines second information according to the seventh information, and sends the second information to the third device.
- the first device can autonomously determine the calling interface information of the knowledge information, and further send the calling interface information, or the calling interface information and the first information to a third device, and the third device autonomously requests the second information from the target device, and then the target device can send the determined second information to the third device.
- interaction between the seventh information and the second information between the third device and the target device can be forwarded through other nodes (such as network open functions), or the information can be directly interacted.
- the method 200 further includes at least one of the following steps:
- the first device receives fifth information sent by the target device
- the first device registers at least one calling interface information according to the fifth information
- the first device sends sixth information to the target device.
- the fifth information includes but is not limited to at least one of the following:
- the calling interface description is used to explain the function of the calling interface.
- the sixth information includes but is not limited to at least one of the following:
- the first device may determine the target calling interface information in the registered calling interface information according to the first information, and acquire the corresponding knowledge information using the target calling interface information.
- the method 200 further includes:
- the third device sends at least one of the eighth information and the first capability information to the network side device;
- the first model is deployed on the third device
- the eighth information is used to indicate the auxiliary information required for the third device to perform reasoning based on the first model
- the first capability information is used to indicate the capability information of the third device to perform reasoning based on the first model.
- the eighth information or the first capability information can be used by the network side device to determine the second information to indicate to the terminal.
- the eighth information includes at least one of the following:
- the third device indicates a network element where knowledge information required for reasoning based on the first model is located
- the third device indicates the knowledge information required for reasoning based on the first model
- the format of the second information is indicated.
- the network element indication where the knowledge information required for the third device to perform reasoning based on the first model is located may be a network element identifier.
- the third device may indicate the network element identifier of the first network element to the network side device.
- the indication of the knowledge information required for the third device to perform reasoning based on the first model may include the indication information of the knowledge graph and/or the indication information of the knowledge vector library required for the third device to perform reasoning based on the first model.
- the indication information of the knowledge graph is one or more knowledge graph identifiers, or a knowledge graph group identifier, a knowledge graph file address, etc., or explicitly indicates one or more knowledge graphs.
- the indication information of the knowledge vector library is used to indicate one or more knowledge vector library identifiers, or explicitly indicates one or more knowledge vector libraries.
- the indication information of the knowledge graph includes at least one of the following:
- the calling interface indication of the knowledge information required for reasoning by the third device based on the first model may explicitly indicate one or more calling interface information, or may implicitly indicate one or more calling interface identifiers corresponding to a set of calling interface information.
- the format indication of the second information may be used to indicate the model input format of the second information.
- the model input format may include the content included in the model input (e.g., including prompt information, including prompt information and calling interface information, including prompt information and configuration information of network side devices, etc.), as well as the dimension, quantization accuracy, size limit, etc. of the second information.
- the preprocessing method indication of the second information can be used to indicate the method for preprocessing the second information, such as biasing the numerical value, scaling the numerical value, transforming the numerical value into a polynomial function, performing multidimensional mapping on a single numerical value, etc.
- the first capability information includes at least one of the following:
- the third device indicates the knowledge information supported by reasoning based on the first model
- the indication of knowledge information supported by the third device for reasoning based on the first model may include indication information of the knowledge graph supported by the third device for reasoning based on the first model and/or indication information of the knowledge vector library.
- the indication information of the knowledge graph is one or more knowledge graph identifiers, or a knowledge graph group identifier, a knowledge graph file address, etc., or explicitly indicates one or more knowledge graphs.
- the indication information of the knowledge vector library is used to indicate one or more knowledge vector library identifiers, or explicitly indicates one or more knowledge vector libraries.
- the calling interface indication of the knowledge information supported by the third device for reasoning based on the first model may explicitly indicate one or more calling interface information, or may implicitly indicate one or more calling interface identifiers corresponding to a set of calling interface information.
- the format indication of the second information supported by the third device can be used to indicate the model input format of the second information supported by the third device.
- the model input format may include what the model input includes (for example, including prompt information, including prompt information and calling interface information, including prompt information and configuration information of the network side device, etc.), as well as the dimensions, quantization accuracy, size limit, etc. of this information.
- the preprocessing method of the second information supported by the third device can be used to indicate the processing method supported by the third device.
- These processing methods are ways in which the network side device preprocesses the second information, such as biasing the value, scaling the value, performing polynomial function changes on the value, performing multidimensional mapping on a single value, etc.
- the third device sends the eighth information and/or the first capability information to the network side device during a model registration (or model identification) process or a capability reporting process.
- the eighth information and the first capability information may be reported to the network side device via one signaling, or may be reported to the network side device via different signaling, and this application does not limit this.
- Example 1 Inference device is a terminal
- the first device is an access network device or a core network function
- the second device and the third device are the same device and are terminals
- the first model is deployed on the terminal.
- the reasoning method may include the following steps:
- Step 1 The terminal sends first information to an access network device or a core network device to request second information.
- Step 2 The access network device or the core network device sends the second information to the terminal according to the first information.
- the inference device can obtain the second information. Further, the inference device can use the second information to assist in the reasoning of the model, thereby improving the reasoning performance of the model.
- Embodiment 2 The inference device is an access network device, such as a base station.
- the first device is a core network function
- the second device and the third device are the same device and are access network devices
- the first model is deployed on the access network device.
- the reasoning method may include the following steps:
- Step 1 The access network device sends first information to the core network device to request second information.
- Step 2 The core network device sends second information to the access network device according to the first information.
- the inference device can obtain the second information. Further, the inference device can use the second information to assist in the reasoning of the model, thereby improving the reasoning performance of the model.
- the first device is a core network function
- the second device is a terminal
- the third device is an access network device.
- the first model is deployed on the access network device, such as the inference device is an access network device, such as a base station.
- the reasoning method may include the following steps:
- Step 1 The terminal sends first information to the core network function to request second information.
- the terminal may also send the first information to the access network device.
- the first information sent by the terminal to the access network device may be different from the first information sent to the core network function.
- the first information sent by the terminal to the core network may include auxiliary information used for reasoning with the first model and input information of the first model; while the first information sent by the terminal to the access network device only includes the input information of the first model.
- the first information may also include a receiving device indication, which is used to indicate the receiving device of the second information, such as a third device.
- Step 2 The core network device sends second information to the access network device according to the first information.
- the input of the access network device for the first model includes two parts, one part comes from the first information sent by the terminal, and the other part comes from the second information sent by the core network. Therefore, the access network device needs to synchronize the two parts of information to generate complete input information of the first model. This prevents the first information of sample 1 and the second information of sample 2 from being combined to generate an erroneous sample. Therefore, at this time, the first information and the second information need to include a sample indication, such as a sample identifier.
- the core network device when the core network device sends the second information to the access network device, it may also send the first information to the access network device.
- the inference device can obtain the first information and the second information. Further, the inference device can use the first information and the second information to assist in the inference of the model, which can improve the inference performance of the model.
- the first device is a core network function
- the second device is a terminal
- the third device is an access network device.
- the first model is deployed on the access network device, such as the inference device is an access network device, such as a base station.
- the reasoning method may include the following steps:
- Step 1 The terminal sends first information to the core network device to request second information.
- Step 2 The core network device sends second information to the terminal according to the first information.
- Step 3 The terminal sends the second information to the access network device.
- the terminal may also send the first information to the access network device.
- the second information is not sent directly from the knowledge storage device to the reasoning device, but is forwarded to the reasoning device by the demand initiating device.
- the inference device can obtain the first information and the second information. Further, the inference device can use the first information and the second information to assist in the reasoning of the model, thereby improving the reasoning performance of the model.
- Step 1 The terminal sends first information to the core network device to request second information.
- the first information includes at least one of the following:
- the receiving device indication is used to indicate the receiving device of the second information.
- Step 2 The core network device sends the first information and the second information to the access network device according to the first information.
- the inference device can obtain the first information and the second information. Further, the inference device can use the first information and the second information to assist in the reasoning of the model, thereby improving the reasoning performance of the model.
- Embodiment 3 The inference device is a core network function.
- the first device is a core network function
- the second device is a terminal or an access network device or a core network device
- the third device is a core network function.
- the reasoning method may include the following steps:
- Step 1 The second device sends first information to the first device to request second information.
- the first information includes at least one of the following:
- a receiving device indication used to indicate a receiving device of the second information
- the input information of the first model is the input information of the first model.
- the second device may also send the first information to the third device to assist the third device in model reasoning.
- the input of the third device for the first model includes two parts, one part comes from the first information sent by the second device, and the other part comes from the second information sent by the first device. Therefore, the third device needs to synchronize the two parts of information to generate complete input information for the first model. This prevents the first information of sample 1 and the second information of sample 2 from being combined to generate an erroneous sample. Therefore, at this time, the first information and the second information need to include a sample indication, such as a sample identifier.
- Step 2 The first device sends second information to the third device according to the first information.
- the inference device can obtain the first information and the second information. Further, the inference device can use the first information and the second information to assist in the reasoning of the model, thereby improving the reasoning performance of the model.
- the first device is a core network function
- the second device is a terminal or an access network device or a core network device
- the third device is a core network function.
- the reasoning method may include the following steps:
- Step 1 The second device sends first information to the first device to request second information.
- Step 2 The first device sends second information to the second device according to the first information.
- Step 3 The second device sends the second information to the third device.
- the first information may also be sent to the third device.
- the inference device can obtain the second information, or the first information and the second information. Furthermore, the inference device can use the second information, or the first information and the second information to assist in the reasoning of the model, thereby improving the reasoning performance of the model.
- the first device is a core network function
- the second device is a terminal or an access network device or a core network device
- the third device is a core network function.
- the reasoning method may include the following steps:
- Step 1 The second device sends first information to the first device to request second information.
- the first information includes at least one of the following:
- a receiving device indication used to indicate a receiving device of the second information
- the input information of the first model is the input information of the first model.
- Step 2 The first device sends the first information and the second information to the third device according to the first information.
- the inference device can obtain the second information, or the first information and the second information. Furthermore, the inference device can use the second information, or the first information and the second information to assist in the reasoning of the model, thereby improving the reasoning performance of the model.
- the first information and the second information include a sample indication, which is used to avoid the inference device using the first information and the second information of different samples for inference, affecting the inference result.
- the model reasoning method in the communication system provided in the embodiment of the present application can be executed by a communication device.
- the communication device provided in the embodiment of the present application is described by taking the communication device executing the model reasoning method in the communication system as an example.
- FIG15 shows a schematic block diagram of a communication device 500 according to an embodiment of the present application.
- the device 500 includes:
- the communication unit 510 is used to receive the first information sent by the second device, wherein the communication device 500 stores at least one of knowledge information and calling interface information of the knowledge information; wherein the knowledge information includes at least one of the following: a knowledge graph, a knowledge vector library; and
- the first information includes at least one of the following:
- the auxiliary information includes at least one of the following:
- Artificial intelligence AI model identification information AI model function indication, AI model feature indication, target knowledge graph indication, cell identification, bandwidth part BWP identification.
- the requirement information includes at least one of the following:
- QoE Service quality of experience
- the second information includes at least one of the following:
- Hidden variables where the hidden variables are feature vectors after mapping the input information of the first model
- the prompt information includes at least one of the following:
- the reminder file identifier related to the target task
- the address of the prompt file related to the target task is the address of the prompt file related to the target task.
- the configuration information of the network side device includes at least one of the following:
- the number of transmitting antennas of the network-side equipment is the number of transmitting antennas of the network-side equipment.
- the number of transmit beams of the network-side device is the number of transmit beams of the network-side device
- the transmission power of the network side equipment is the transmission power of the network side equipment
- the scene type information includes at least one of the following:
- Line-of-sight LOS Non-line-of-sight NLOS, indoor, outdoor, urban microcell, urban macrocell, suburban microcell, high speed, low speed.
- the communication device 500 further includes:
- a processing unit is used to determine the second information according to the first information and the knowledge information stored in the communication device 500.
- the communication unit 510 is further configured to:
- the second information is received from a target device, where the target device is a device accessed using the calling interface information.
- the first information includes calling interface information of knowledge information
- the communication device 500 further includes: a processing unit, configured to determine the second information according to the first information and the knowledge information stored on the communication device 500 .
- the communication unit 510 is also used to: send the first information to a target device according to the calling interface information of the knowledge information, wherein the target device is a device accessed using the calling interface information; and receive the second information from the target device.
- the first information does not include calling interface information of the knowledge information
- the communication device 500 further includes: a processing unit, configured to determine the calling interface information of the knowledge information according to the first information;
- the communication device 500 is also used to: send third information to a target device, wherein the target device is a device accessed using the calling interface information, and the third information includes the calling interface information of the knowledge information; and receive the second information from the target device.
- the first information does not include calling interface information of the knowledge information
- the communication device 500 further includes: a processing unit, configured to determine the calling interface information of the knowledge information according to the first information;
- the communication unit 510 is also used to: send fourth information to the third device, the fourth information includes calling interface information of the knowledge information, or the fourth information includes the first information and the calling interface information of the knowledge information, the fourth information is used by the third device to request the target device to obtain the second information, and the target device is the device accessed using the calling interface information.
- the communication unit 510 is further configured to:
- the communication device 500 further includes a processing unit, configured to register at least one calling interface information according to the fifth information;
- the communication unit 510 is further used to: send sixth information to the target device;
- the fifth information includes at least one of the following: a calling interface name, a calling interface parameter list, a calling interface description, a calling interface usage example, and a calling interface instruction set;
- the sixth information includes at least one of the following:
- the communication unit 510 is further configured to:
- the first information is sent to the third device.
- the communication device 500 is an access network device or a core network function, and the second device and the third device are terminals; or
- the communication device 500 is a core network function, the second device is a terminal or an access network device, and the third device is an access network device; or
- the communication device 500 is a core network function
- the second device is a terminal or an access network device or a core network function
- the third device is a core network function.
- the communication device 500 is an access network device, the second device is a terminal, and the first information is carried in at least one of the following signaling: layer 1 signaling, layer 2 signaling, layer 3 signaling, AI layer signaling, data plane signaling; or
- the communication device 500 is a core network function
- the third device is a terminal
- the first information is carried in at least one of the following signaling: non-access layer NAS signaling, AI layer signaling, and data plane signaling.
- the calling interface information includes at least one of the following:
- Calling interface name calling interface input data format, calling interface output data format, calling interface input data.
- the communication unit may be a communication interface or a transceiver, or an input/output interface of a communication chip or a system on chip.
- the communication device 500 may correspond to the first device in the method embodiment of the present application, and the above-mentioned and other operations and/or functions of each unit in the communication device 500 are respectively for realizing the corresponding processes of the first device in the method embodiment shown in Figures 2 to 14, and achieving the same technical effect. To avoid repetition, they will not be repeated here.
- FIG16 shows a schematic block diagram of a communication device 600 according to an embodiment of the present application.
- the device 600 includes:
- the communication unit 610 is configured to send first information to a first device, wherein the first device stores at least one of knowledge information and calling interface information of the knowledge information, wherein the knowledge information includes at least one of the following: a knowledge graph and a knowledge vector library;
- the first information includes at least one of the following:
- the auxiliary information includes at least one of the following:
- Artificial intelligence AI model identification information AI model function indication, AI model feature indication, target knowledge graph indication, cell identification, bandwidth part BWP identification.
- the requirement information includes at least one of the following:
- QoE Service quality of experience
- the second information includes at least one of the following:
- Hidden variables where the hidden variables are feature vectors after mapping the input information of the first model
- the reminder file identifier related to the target task
- the address of the prompt file related to the target task is the address of the prompt file related to the target task.
- the configuration information of the network side device includes at least one of the following:
- the number of transmitting antennas of the network-side equipment is the number of transmitting antennas of the network-side equipment.
- the number of transmit beams of the network-side device is the number of transmit beams of the network-side device
- the transmission power of the network side equipment is the transmission power of the network side equipment
- the scene type information includes at least one of the following:
- Line-of-sight LOS Non-line-of-sight NLOS, indoor, outdoor, urban microcell, urban macrocell, suburban microcell, high speed, low speed.
- the calling interface information includes at least one of the following:
- Calling interface name calling interface input data format, calling interface output data format, calling interface input data.
- the communication device 600 is a terminal, and the communication unit 610 is specifically used for:
- the communication device 600 determines to activate or switch to the first model
- the communication device 600 determines that the first model does not meet the requirements
- a second indication is received, where the second indication is used to indicate that the first model does not meet the requirements.
- the communication unit 610 is further configured to:
- the first information is sent to a third device, and the first model is deployed on the third device.
- the communication unit may be a communication interface or a transceiver, or an input/output interface of a communication chip or a system on chip.
- the communication device 600 may correspond to the second device in the method embodiment of the present application, and the above-mentioned and other operations and/or functions of each unit in the communication device 600 are respectively for realizing the corresponding processes of the second device in the method embodiment shown in Figures 2 to 14, and achieving the same technical effect. To avoid repetition, they will not be repeated here.
- FIG17 shows a schematic block diagram of a communication device 700 according to an embodiment of the present application.
- the device 700 includes:
- the communication unit 710 is used to obtain second information from a first device or a target device, wherein a first model is deployed on the communication device 700, and at least one of knowledge information and calling interface information of knowledge information is stored on the first device or the target device, wherein the target device is a device accessed using the calling interface information, and the knowledge information includes at least one of the following: a knowledge graph and a knowledge vector library;
- the second information is determined according to the first information and the knowledge information stored or called on the first device;
- the first information includes at least one of the following:
- the auxiliary information includes at least one of the following:
- Artificial intelligence AI model identification information AI model function indication, AI model feature indication, target knowledge graph indication, cell identification, bandwidth part BWP identification.
- the requirement information includes at least one of the following:
- QoE Service quality of experience
- the second information includes at least one of the following:
- Hidden variables where the hidden variables are feature vectors after mapping the input information of the first model
- the prompt information includes at least one of the following:
- the reminder file identifier related to the target task
- the address of the prompt file related to the target task is the address of the prompt file related to the target task.
- the configuration information of the network side device includes at least one of the following:
- the number of transmitting antennas of the network-side equipment is the number of transmitting antennas of the network-side equipment.
- the number of transmit beams of the network-side device is the number of transmit beams of the network-side device
- the transmission power of the network side equipment is the transmission power of the network side equipment
- the scene type information includes at least one of the following:
- Line-of-sight LOS Non-line-of-sight NLOS, indoor, outdoor, urban microcell, urban macrocell, suburban microcell, high speed, low speed.
- the communication unit 710 is further configured to:
- the fourth information including the calling interface information of the knowledge information, or the fourth information including the first information and the calling interface information of the knowledge information;
- the seventh information includes calling interface information of the knowledge information
- the second information is received from the target device.
- the communication device 700 is a terminal, and the communication unit 710 is further used for:
- the eighth information is used to indicate the auxiliary information required for the communication device 700 to perform reasoning based on the first model
- the first capability information is used to indicate the capability information of the communication device 700 to perform reasoning based on the first model
- the eighth information includes at least one of the following:
- the communication device 700 indicates the knowledge information required for reasoning based on the first model
- the communication device 700 indicates a calling interface for the knowledge information required for reasoning based on the first model
- the format of the second information is indicated.
- the first capability information includes at least one of the following:
- the communication device 700 indicates the knowledge information supported by the reasoning based on the first model
- the communication device 700 indicates a calling interface for the knowledge information supported by the reasoning based on the first model
- the communication unit 710 is further configured to:
- At least one of the eighth information and the first capability information is sent to the network side device.
- the communication unit 710 is further configured to:
- the first information is received from the second device, and the first model is deployed on the communication apparatus 700.
- the communication device 700 further includes:
- a processing unit is used to use the first model to perform reasoning based on the second information and input information of the first model.
- the calling interface information includes at least one of the following:
- Calling interface name calling interface input data format, calling interface output data format, calling interface input data.
- the communication unit may be a communication interface or a transceiver, or an input/output interface of a communication chip or a system on chip.
- the communication device 700 may correspond to the third device in the method embodiment of the present application, and the above-mentioned and other operations and/or functions of each unit in the communication device 700 are respectively for realizing the corresponding processes of the third device in the method embodiment shown in Figures 2 to 14, and achieving the same technical effect. To avoid repetition, they will not be repeated here.
- the apparatus 500, apparatus 600, and apparatus 700 in the embodiments of the present application may be an electronic device, such as an electronic device having an operating system, or a component in an electronic device, such as an integrated circuit or a chip.
- the electronic device may be a terminal, or may be another device other than a terminal.
- the terminal may include but is not limited to the types of terminal 11 listed above, and other devices may be servers, network attached storage (NAS), etc., which are not specifically limited in the embodiments of the present application.
- NAS network attached storage
- an embodiment of the present application further provides a communication device 1000, including a processor 1001 and a memory 1002, wherein the memory 1002 stores a program or instruction that can be run on the processor 1001.
- the communication device 1000 is a first device
- the program or instruction is executed by the processor 1001 to implement the steps performed by the first device in the above-mentioned reasoning method embodiment, and the same technical effect can be achieved.
- the communication device 1000 is a second device
- the program or instruction is executed by the processor 1001 to implement the steps performed by the second device in the above-mentioned reasoning method embodiment, and the same technical effect can be achieved.
- the communication device 1000 is a third device
- the program or instruction is executed by the processor 1001 to implement the steps performed by the third device in the above-mentioned reasoning method embodiment, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.
- the embodiment of the present application also provides a terminal, including a processor and a communication interface, the communication interface is coupled to the processor, and the processor is used to run a program or instruction to implement the steps in the method embodiment shown in Figures 2 to 14.
- This terminal embodiment corresponds to the above-mentioned terminal side method embodiment, and each implementation process and implementation method of the above-mentioned method embodiment can be applied to the terminal embodiment and can achieve the same technical effect.
- Figure 19 is a schematic diagram of the hardware structure of a terminal implementing an embodiment of the present application.
- the terminal 1100 includes but is not limited to: a radio frequency unit 1101, a network module 1102, an audio output unit 1103, an input unit 1104, a sensor 1105, a display unit 1106, a user input unit 1107, an interface unit 1108, a memory 1109 and at least some of the components of a processor 1110.
- the terminal 1100 may also include a power source (such as a battery) for supplying power to each component, and the power source may be logically connected to the processor 1110 through a power management system, so as to implement functions such as managing charging, discharging, and power consumption management through the power management system.
- a power source such as a battery
- the terminal structure shown in FIG19 does not constitute a limitation on the terminal, and the terminal may include more or fewer components than shown in the figure, or combine certain components, or arrange components differently, which will not be described in detail here.
- the input unit 1104 may include a graphics processing unit (GPU) 11041 and a microphone 11042, and the graphics processor 11041 processes the image data of the static picture or video obtained by the image capture device (such as a camera) in the video capture mode or the image capture mode.
- the display unit 1106 may include a display panel 11061, and the display panel 11061 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc.
- the user input unit 1107 includes a touch panel 11071 and at least one of other input devices 11072.
- the touch panel 11071 is also called a touch screen.
- the touch panel 11071 may include two parts: a touch detection device and a touch controller.
- Other input devices 11072 may include, but are not limited to, a physical keyboard, function keys (such as a volume control key, a switch key, etc.), a trackball, a mouse, and a joystick, which will not be repeated here.
- the RF unit 1101 can transmit the data to the processor 1110 for processing; in addition, the RF unit 1101 can send uplink data to the network side device.
- the RF unit 1101 includes but is not limited to an antenna, an amplifier, a transceiver, a coupler, a low noise amplifier, a duplexer, etc.
- the memory 1109 can be used to store software programs or instructions and various data.
- the memory 1109 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, an application program or instruction required for at least one function (such as a sound playback function, an image playback function, etc.), etc.
- the memory 1109 may include a volatile memory or a non-volatile memory.
- the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory.
- the volatile memory may be a random access memory (RAM), a static random access memory (SRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDRSDRAM), an enhanced synchronous dynamic random access memory (ESDRAM), a synchronous link dynamic random access memory (SLDRAM) and a direct memory bus random access memory (DRRAM).
- RAM random access memory
- SRAM static random access memory
- DRAM dynamic random access memory
- SDRAM synchronous dynamic random access memory
- DDRSDRAM double data rate synchronous dynamic random access memory
- ESDRAM enhanced synchronous dynamic random access memory
- SLDRAM synchronous link dynamic random access memory
- DRRAM direct memory bus random access memory
- the processor 1110 may include one or more processing units; optionally, the processor 1110 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communication signals, such as a baseband processor. It is understandable that the modem processor may not be integrated into the processor 1110.
- the embodiment of the present application also provides a network side device, including a processor and a communication interface, the communication interface is coupled to the processor, and the processor is used to run a program or instruction to implement the steps of the method embodiment shown in Figures 2 to 14.
- the network side device embodiment corresponds to the above-mentioned access network device side or core network function side method embodiment, and each implementation process and implementation method of the above-mentioned method embodiment can be applied to the network side device embodiment, and can achieve the same technical effect.
- the embodiment of the present application also provides a network side device.
- the network side device 1200 includes: an antenna 1201, a radio frequency device 1202, a baseband device 1203, a processor 1204 and a memory 1205.
- the antenna 1201 is connected to the radio frequency device 1202.
- the radio frequency device 1202 receives information through the antenna 1201 and sends the received information to the baseband device 1203 for processing.
- the baseband device 1203 processes the information to be sent and sends it to the radio frequency device 1202.
- the radio frequency device 1202 processes the received information and sends it out through the antenna 1201.
- the method executed by the network-side device in the above embodiment may be implemented in the baseband device 1203, which includes a baseband processor.
- the baseband device 1203 may include, for example, at least one baseband board, on which multiple chips are arranged, as shown in Figure 20, one of which is, for example, a baseband processor, which is connected to the memory 1205 through a bus interface to call the program in the memory 1205 and execute the network device operations shown in the above method embodiment.
- the network side device may also include a network interface 1206, which is, for example, a Common Public Radio Interface (CPRI).
- CPRI Common Public Radio Interface
- the network side device 1200 of the embodiment of the present application also includes: instructions or programs stored in the memory 1205 and executable on the processor 1204.
- the processor 1204 calls the instructions or programs in the memory 1205 to execute the methods executed by the modules shown in Figures 15 to 17 and achieve the same technical effect. To avoid repetition, it will not be repeated here.
- the embodiment of the present application further provides a network side device.
- the network side device 1300 includes: a processor 1301, a network interface 1302, and a memory 1303.
- the network interface 1302 is, for example, a common public radio interface (CPRI).
- CPRI common public radio interface
- the network side device 1300 of the embodiment of the present application also includes: instructions or programs stored in the memory 1303 and executable on the processor 1301.
- the processor 1301 calls the instructions or programs in the memory 1303 to execute the methods executed by the modules shown in Figures 15 to 17 and achieve the same technical effect. To avoid repetition, it will not be repeated here.
- An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored.
- a program or instruction is stored.
- the various processes of the model reasoning method embodiment in the above-mentioned communication system are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.
- the processor is a processor in the communication device, communication equipment, terminal or network side equipment described in the above embodiments.
- the readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.
- the readable storage medium may be a non-transient readable storage medium.
- An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the model reasoning method embodiment in the above-mentioned communication system, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
- the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.
- An embodiment of the present application further provides a computer program/program product, which is stored in a storage medium, and is executed by at least one processor to implement the various processes of the model reasoning method embodiment in the above-mentioned communication system, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
- An embodiment of the present application also provides a communication system, including: a first device, a second device and a third device, wherein the first device can be used to execute the steps performed by the first device in the model inference method in the communication system as described above, the second device can be used to execute the steps performed by the second device in the model inference method in the communication system as described above, and the third device can be used to execute the steps performed by the third device in the model inference method in the communication system as described above.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Computing Systems (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
本申请公开了一种通信系统中的模型推理方法、装置、设备及介质,属于通信领域,本申请实施例的方法包括:第一设备接收第二设备发送的第一信息,第一设备上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,知识信息包括以下至少之一:知识图谱、知识向量库;第一设备向第三设备发送第二信息,第三设备上部署第一模型,第二信息是根据第一信息和第一设备上存储的或调用的知识信息确定的;其中,第一信息包括以下至少之一:用于第一模型进行推理的辅助信息;用于第一模型进行推理的需求信息;第一模型的输入信息;知识信息的调用接口信息;第二信息的接收设备指示;第一模型的输入信息对应的样本指示。
Description
相关申请的交叉引用
本申请要求于2023年11月20日提交中国专利局、申请号为202311551035.X、发明名称为“通信系统中的模型推理方法、装置、设备以及介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请属于通信技术领域,具体涉及一种通信系统中的模型推理方法、装置、设备以及介质。
在移动通信系统中,开始有越来越多的用例结合人工智能(AI)。例如在物理层有基于AI的CSI(channel state information)预测和反馈压缩,基于AI的波束管理,基于AI的定位等。为了使不同通信场景具有普适性,提出了将大模型应用到通信网络中来解决通信系统中的问题,但是,大模型通常在解决通用问题上的性能较优,但是通信系统中存在许多不同的问题,例如,波束赋形,资源分配,信道预测等,因此,如何保证大模型在解决具体的下游问题时的性能是一项亟需解决的问题。
本申请实施例提供一种通信系统中的模型推理方法、装置、设备以及介质,能够提升模型的性能。
第一方面,提供了一种通信系统中的模型推理方法,该方法包括:
第一设备接收第二设备发送的第一信息,其中,所述第一设备上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,所述知识信息包括以下至少之一:知识图谱、知识向量库;
所述第一设备向第三设备发送第二信息,所述第三设备上部署第一模型,所述第二信息是根据所述第一信息和所述第一设备上存储的或调用的知识信息确定的;
其中,所述第一信息包括以下至少之一:
用于所述第一模型进行推理的辅助信息;
用于所述第一模型进行推理的需求信息;
所述第一模型的输入信息;
知识信息的调用接口信息;
所述第二信息的接收设备指示;
所述第一模型的输入信息对应的样本指示。
第二方面,提供了一种通信系统中的模型推理方法,该方法包括:
第二设备向第一设备发送第一信息,其中,所述第一设备上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,所述知识信息包括以下至少之一:知识图谱、知识向量库;
其中,所述第一信息包括以下至少之一:
用于第一模型进行推理的辅助信息;
用于第一模型进行推理的需求信息;
第一模型的输入信息;
知识信息的调用接口信息;
所述第二信息的接收设备指示;
所述第一模型的输入信息对应的样本指示。
第三方面,提供了一种通信系统中的模型推理方法,包括:
第三设备从第一设备或目标设备获取第二信息,其中,所述第三设备上部署第一模型,所述第一设备或所述目标设备上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,所述目标设备是使用所述调用接口信息所访问的设备,所述知识信息包括以下至少之一:知识图谱、知识向量库;
其中,所述第二信息根据第一信息和所述第一设备上存储或调用的知识信息确定;
其中,所述第一信息包括以下至少之一:
用于所述第一模型进行推理的辅助信息;
用于所述第一模型进行推理的需求信息;
所述第一模型的输入信息;
知识信息的调用接口信息;
所述第二信息的接收设备指示;
所述第一模型的输入信息对应的样本指示。
第四方面,提供了一种通信装置,包括:
通信单元,用于接收第二设备发送的第一信息,其中,所述通信装置上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,所述知识信息包括以下至少之一:知识图谱、知识向量库;以及
向第三设备发送第二信息,所述第三设备上部署第一模型,所述第二信息是根据所述第一信息和所述第一设备上存储的或调用的知识信息确定的;
其中,所述第一信息包括以下至少之一:
用于所述第一模型进行推理的辅助信息;
用于所述第一模型进行推理的需求信息;
所述第一模型的输入信息;
知识信息的调用接口信息;
所述第二信息的接收设备指示;
所述第一模型的输入信息对应的样本指示。
第五方面,提供了一种通信装置,包括:
通信单元,用于向第一设备发送第一信息,其中,所述第一设备上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,所述知识信息包括以下至少之一:知识图谱、知识向量库;
其中,所述第一信息包括以下至少之一:
用于第一模型进行推理的辅助信息;
用于第一模型进行推理的需求信息;
第一模型的输入信息;
知识信息的调用接口信息;
所述第二信息的接收设备指示;
所述第一模型的输入信息对应的样本指示。
第六方面,提供了一种通信装置,包括:
通信单元,用于从第一设备或目标设备获取第二信息,其中,所述第三设备上部署第一模型,所述第一设备或所述目标设备上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,所述目标设备是使用所述调用接口信息所访问的设备,所述知识信息包括以下至少之一:知识图谱、知识向量库;
其中,所述第二信息根据第一信息和所述第一设备上存储或调用的知识信息确定;
其中,所述第一信息包括以下至少之一:
用于所述第一模型进行推理的辅助信息;
用于所述第一模型进行推理的需求信息;
所述第一模型的输入信息;
知识信息的调用接口信息;
所述第二信息的接收设备指示;
所述第一模型的输入信息对应的样本指示。
第七方面,提供了一种通信设备,该网络侧设备包括处理器和存储器,所述存储器存储可在所述处理器上运行的程序或指令,所述程序或指令被所述处理器执行时实现如第一方面至第三方面中的任一方面所述的方法的步骤。
第八方面,提供了一种可读存储介质,所述可读存储介质上存储程序或指令,所述程序或指令被处理器执行时实现如第一方面至第三方面中的任一方面所述的方法的步骤。
第九方面,提供了一种无线通信系统,包括:第一设备、第二设备及第三设备,所述第一设备可用于执行如第一方面所述的方法的步骤,所述第二设备可用于执行如第二方面所述的方法的步骤,所述第三设备可用于执行如第三方面所述的方法的步骤。
第十方面,提供了一种芯片,所述芯片包括处理器和通信接口,所述通信接口和所述处理器耦合,所述处理器用于运行程序或指令,实现如第一方面至第三方面中的任一方面中所述的方法。
第十一方面,提供了一种计算机程序/程序产品,所述计算机程序/程序产品被存储在存储介质中,所述程序/程序产品被至少一个处理器执行以实现如第一方面至第三方面中的任一方面所述的方法。
在本申请实施例中,第二设备可以通过向第一设备发送第一信息,辅助第三设备获得第二信息,进一步地,第三设备可以使用第二信息辅助进行模型的推理,有利于提升模型的推理性能。
图1是本申请实施例提供的一种通信系统的示意性图。
图2是本申请实施例提供的一种通信系统中的模型推理方法的示意性图。
图3是本申请实施例提供的一种隐藏变量的示意图。
图4至图14是本申请实施例提供的模型推理方法的示意性交互图。
图15是本申请实施例提供的一种通信装置的示意性图。
图16是本申请实施例提供的另一种通信装置的示意性图。
图17是本申请实施例提供的又一种通信装置的示意性图。
图18是本申请实施例提供的一种通信设备的示意性图。
图19是本申请实施例提供的一种终端的硬件结构图。
图20是本申请实施例提供的一种网络侧设备的硬件结构图。
图21是本申请实施例提供的另一种网络侧设备的硬件结构图。
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚描述,显然,所描述的实施例是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员所获得的所有其他实施例,都属于本申请保护的范围。
本申请的术语“第一”、“第二”等是用于区别类似的对象,而不用于描述特定的顺序或先后次序。应该理解这样使用的术语在适当情况下可以互换,以便本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施,且“第一”、“第二”所区别的对象通常为一类,并不限定对象的个数,例如第一对象可以是一个,也可以是多个。此外,本申请中的“或”表示所连接对象的至少其中之一。例如“A或B”涵盖三种方案,即,方案一:包括A且不包括B;方案二:包括B且不包括A;方案三:既包括A又包括B。字符“/”一般表示前后关联对象是一种“或”的关系。
本申请的术语“指示”既可以是一个直接的指示(或者说显式的指示),也可以是一个间接的指示(或者说隐含的指示)。其中,直接的指示可以理解为,发送方在发送的指示中明确告知了接收方具体的信息、需要执行的操作或请求结果等内容;间接的指示可以理解为,接收方根据发送方发送的指示确定对应的信息,或者进行判断并根据判断结果确定需要执行的操作或请求结果等。
值得指出的是,本申请实施例所描述的技术不限于长期演进型(Long Term Evolution,LTE)/LTE的演进(LTE-Advanced,LTE-A)系统,还可用于其他无线通信系统,诸如码分多址(Code Division Multiple Access,CDMA)、时分多址(Time Division Multiple Access,TDMA)、频分多址(Frequency Division Multiple Access,FDMA)、正交频分多址(Orthogonal Frequency Division Multiple Access,OFDMA)、单载波频分多址(Single-carrier Frequency-Division Multiple Access,SC-FDMA)或其他系统。本申请实施例中的术语“系统”和“网络”常被可互换地使用,所描述的技术既可用于以上提及的系统和无线电技术,也可用于其他系统和无线电技术。以下描述出于示例目的描述了新空口(New Radio,NR)系统,并且在以下大部分描述中使用NR术语,但是这些技术也可应用于NR系统以外的系统,如第6代(6th Generation,6G)通信系统。
图1示出本申请实施例可应用的一种无线通信系统的框图。无线通信系统包括终端11和网络侧设备12。其中,终端11可以是手机、平板电脑(Tablet Personal Computer)、膝上型电脑(Laptop Computer)、笔记本电脑、个人数字助理(Personal Digital Assistant,PDA)、掌上电脑、上网本、超级移动个人计算机(Ultra-mobile Personal Computer,UMPC)、移动上网装置(Mobile Internet Device,MID)、增强现实(Augmented Reality,AR)、虚拟现实(Virtual Reality,VR)设备、机器人、可穿戴式设备(Wearable Device)、飞行器(flight vehicle)、车载设备(Vehicle User Equipment,VUE)、船载设备、行人终端(Pedestrian User Equipment,PUE)、智能家居(具有无线通信功能的家居设备,如冰箱、电视、洗衣机或者家具等)、游戏机、个人计算机(Personal Computer,PC)、柜员机或者自助机等终端侧设备。可穿戴式设备包括:智能手表、智能手环、智能耳机、智能眼镜、智能首饰(智能手镯、智能手链、智能戒指、智能项链、智能脚镯、智能脚链等)、智能腕带、智能服装等。其中,车载设备也可以称为车载终端、车载控制器、车载模块、车载部件、车载芯片或车载单元等。需要说明的是,在本申请实施例并不限定终端11的具体类型。
终端也可以称为用户设备(User Equipment,UE)、终端设备、接入终端、用户单元、用户站、移动站、移动台、远方站、远程终端、移动设备、用户终端、终端、无线通信设备、用户代理或用户装置等。
网络侧设备12可以包括接入网设备或核心网设备,其中,接入网设备也可以称为无线接入网(Radio Access Network,RAN)设备、无线接入网功能或无线接入网单元。接入网设备可以包括基站、无线局域网(Wireless Local Area Network,WLAN)接入点(Access Point,AS)或无线保真(Wireless Fidelity,WiFi)节点等。其中,基站可被称为节点B(Node B,NB)、演进节点B(Evolved Node B,eNB)、下一代节点B(the next generation Node B,gNB)、新空口节点B(New Radio Node B,NR Node B)、接入点、中继站(Relay Base Station,RBS)、服务基站(Serving Base Station,SBS)、基收发机站(Base Transceiver Station,BTS)、无线电基站、无线电收发机、基本服务集(Basic Service Set,BSS)、扩展服务集(Extended Service Set,ESS)、家用B节点(home Node B,HNB)、家用演进型B节点(home evolved Node B)、发送接收点(Transmission Reception Point,TRP)或所属领域中其他某个合适的术语,只要达到相同的技术效果,所述基站不限于特定技术词汇,需要说明的是,在本申请实施例中仅以NR系统中的基站为例进行介绍,并不限定基站的具体类型。
核心网设备可以包含核心网设备可以包含但不限于如下至少一项:核心网节点、核心网功能、移动管理实体(Mobility Management Entity,MME)、接入移动管理功能(Access and Mobility Management Function,AMF)、会话管理功能(Session Management Function,SMF)、用户平面功能(User Plane Function,UPF)、策略控制功能(Policy Control Function,PCF)、策略与计费规则功能单元(Policy and Charging Rules Function,PCRF)、边缘应用服务发现功能(Edge Application Server Discovery Function,EASDF)、统一数据管理(Unified Data Management,UDM)、统一数据仓储(Unified Data Repository,UDR)、归属用户服务器(Home Subscriber Server,HSS)、集中式网络配置(Centralized network configuration,CNC)、网络存储功能(Network Repository Function,NRF)、网络开放功能(Network Exposure Function,NEF)、本地NEF(Local NEF,或L-NEF)、绑定支持功能(Binding Support Function,BSF)、应用功能(Application Function,AF)等。需要说明的是,在本申请实施例中仅以NR系统中的核心网设备为例进行介绍,并不限定核心网设备的具体类型。但不限于如下至少一项:核心网节点、核心网功能、移动管理实体(Mobility Management Entity,MME)、接入移动管理功能(Access and Mobility Management Function,AMF)、会话管理功能(Session Management Function,SMF)、用户平面功能(User Plane Function,UPF)、策略控制功能(Policy Control Function,PCF)、策略与计费规则功能单元(Policy and Charging Rules Function,PCRF)、边缘应用服务发现功能(Edge Application Server Discovery Function,EASDF)、统一数据管理(Unified Data Management,UDM)、统一数据仓储(Unified Data Repository,UDR)、归属用户服务器(Home Subscriber Server,HSS)、集中式网络配置(Centralized network configuration,CNC)、网络存储功能(Network Repository Function,NRF)、网络开放功能(Network Exposure Function,NEF)、本地NEF(Local NEF,或L-NEF)、绑定支持功能(Binding Support Function,BSF)、应用功能(Application Function,AF)等。需要说明的是,在本申请实施例中仅以NR系统中的核心网设备为例进行介绍,并不限定核心网设备的具体类型。
为便于理解理解本申请实施例,对本申请相关的技术进行说明。
一、知识图谱
知识图谱是一种利用图结构或拓扑来表示和整合数据的知识库,可以存储实体(对象、事件、情况或抽象概念)之间的关联描述,同时也可以编码实体之间的语义关系。知识图谱的应用前景非常广阔,既能提高信息检索、搜索引擎、推荐系统等数据服务的效果,还支持自然语言问答、对话、推理等智能交互,具备表示能力强、灵活性高、可计算性强和跨领域性强等特点。
知识图谱以明确和结构化的方式存储了大量的知识,可以用于增强大模型的知识意识。在推理阶段将知识图谱纳入大模型中,通过从知识图谱检索知识,可以显著提高大模型在访问领域特定知识方面的性能。
在通信领域,主要利用网络数据知识图谱进行知识表示、关联关系分析和深度挖掘,为通信系统的智能化提供有效的知识规则与知识计算支撑。
通信系统的网络结构、终端类型、终端行为、数据业务需求、系统资源都具有高动态性、时效性强及相互耦合性等特点,移动通信数据面临诸多挑战,数据分散获取困难,种类繁多结构复杂,关联复杂挖掘困难等等。
利用知识图谱可以有效厘清数据字段与通信网络指标之间的各种关系,并在关系建立的基础上进一步深度挖掘,比如量化关系的关联程度、表征数据字段与指标的特征属性等等。
在一些实现方式中,知识图谱可以通过实体之间的相关性表示,一种表示方式可以为:
(实体1,实体1和实体2之间的相关性系数,实体2)。
例如,实体1是小区吞吐量,实体2是最强波束的L1-RSRP,相关性系数为0.5。
在另一些实现方式中,知识图谱可以通过主体属性对象(Subject,Predicate,Object,SPO)三元组表示,例如,(小区id,赋形码本,码本指示)的SPO三元组。
二、知识向量库
知识向量库是用来存储、检索、分析向量的数据库。之所以称之为数据库,是因为它有下面几个特征:
a)提供标准的访问接口,降低用户的使用门槛;
b)提供高效的数据组织、检索和分析的能力。一般用户在存储和检索向量的同时,还需要管理结构化的数据,如支持传统数据库对结构化数据的管理能力。
因此,知识向量库可以简单理解为用于存储模型输入特征向量的数据库。
例如,在利用图片搜索图片,或者,利用语音搜索语音的时候,在知识向量库中存储和对比的并不是图片和语音片段,而是通过深度学习等算法提取出来的“特征”,例如256或512个浮点数(float)数组,可以用数学中的向量来表示。
在一些情况中,知识向量库可以是一个模型,输入的是一个图片,文字。例如,在通信领域,模型的输入可以是无线信号的测量结果等,输出是一个用于表征该测量结果的特征向量。
在本地可以采用如下方法使用知识图谱或知识向量库:
1)直接作为模型的输入;
2)用知识图谱对模型输入进行处理后,再把处理后的数据和原始的模型输入一起输入到模型中。
3)用知识向量库对模型输入进行处理后,再把处理后的数据和原始的模型输入一起输入到模型中,或者直接使用处理后的数据输入到模型中。
三、提示工程
提示工程是自然语言处理(Natural Language Processing,NLP)中的一种技术,可以制作文本片段、提示或模板,用于指导预训练大语言模型产生特定任务或应用程序的高质量输出,广泛用于问答、摘要、翻译、情感分析和文本生成等领域。提示工程可以利用预训练大语言模型的强大能力来实现各种复杂的自然语言处理任务,减少对标记数据和模型微调的依赖性,降低开发成本和时间,以及提高预训练大语言模型的可解释性和可控性,增加用户信任和满意度。常见的提示可以包括如下零样本提示(Zero-Shot Prompting)、少样本提示(Few-Shot Prompting)、思维链提示(Chain of Thought Prompting)等。
在提示工程中,任务的描述会被嵌入到输入中,从而引导生成式人工智能解决方案生成所需输出。
在移动通信系统中,开始有越来越多的用例结合人工智能(Artificial Intelligence,AI)。例如在物理层有基于AI的信道状态信息(Channel State Information,CSI)预测和反馈压缩,基于AI的波束管理,基于AI的定位等。在一些场景中,也考虑引入基于AI的节能,基于AI的负载均衡等。未来在移动通信系统中会出现更多的结合AI的用例。
近年来,大语言模型在聊天、图像生成等领域获得了广泛关注。因此,考虑将大模型应用到通信网络中,作为一种工具来优化通信网络的性能。但是,通信大模型有两个问题。
问题1:大模型的尺寸很大,如何将大模型部署到例如基站或者终端等存储资源有限的设备进行推理,较为典型的方法是通过量化、剪枝等压缩方法对大模型进行处理后,再部署到通信设备上。
问题2:大模型往往是解决通用问题,而通信中有很多不同的问题,例如波束赋形,资源分配,信道预测。如何将大模型匹配到具体的下游问题上,目前在通信系统中还没有成熟的方案。
下面结合附图,通过一些实施例及其应用场景对本申请实施例提供的通信系统中基于人工智能AI模型的推理方法进行详细地说明。
图2示出了根据本申请实施例的通信系统中的模型推理方法的示意性图。如图2所示,该方法200包括:
S201,第一设备接收第二设备发送的第一信息,其中,所述第一设备上存储有知识信息和知识信息的调用接口信息中的至少之一;
S202,所述第一设备向第三设备发送第二信息,所述第三设备上部署第一模型,所述第二信息是根据所述第一信息和知识信息确定的。
在一些实施例中,所述第一模型是压缩后的AI模型,例如,第一模型是对大模型进行压缩后的AI模型,这里的压缩例如可以包括但不限于量化、剪枝(例如,删除模型中的部分层,或者删除某层中的部分节点)等。
需要说明的是,在本申请实施例中,AI模型也可称为AI单元、ML(machine learning)模型、ML单元、AI结构、AI功能、AI特性、机器学习模型、神经网络、神经网络函数、神经网络功能等,或者,AI模型也可以是指能够实现与AI相关的特定的算法、公式、处理流程、能力等的处理单元,或者AI模型可以是针对特定数据集的处理方法、算法、功能、模块或单元,或者AI模型可以是运行在图形处理单元(Graphics Processing Unit,GPU)、神经网络处理单元(Neural Processing Unit,NPU)、张量处理单元(Tensor Processing Unit,TPU)、专门应用集成电路(Application Specific Integrated Circuit,ASIC)等AI/ML相关硬件上的处理方法、算法、功能、模块或单元,本申请对此不做具体限定。
在本申请实施例中,第一设备可以认为是存储知识相关信息的设备,或称知识存储设备,例如第一设备上存储有知识信息、知识信息的调用接口信息等。第二设备可以认为是需求发起设备,如发起推理需求的设备。第三设备可以认为是推理设备,或者说,AI模型部署设备。其中,第二设备和第三设备可以是同一设备,或者,也可以是不同设备。
在一些实施例中,知识信息包括但不限于知识图谱和知识向量库中的至少之一。
在一些实施例中,所述第三设备可以利用第二信息扩充第一模型的输入,辅助第一模型进行精准的推理。
因此,在本申请实施例中,通过向存储有知识存储设备获取第二信息,进一步使用模型进行推理时,可以使用第二信息和第一模型的输入信息,从而能够使得模型能够获得更多的输入信息,提升模型的推理性能。
在一些实施例中,所述第一设备为接入网设备(例如基站)或核心网功能,例如具有数据存储功能的核心网功能,例如数据库、数据功能或网络存储功能(Network Repository Function,NRF)或UDM,或AI模型库,或AI模型管理功能或终端的第三方服务器等,第二设备是终端,第三设备是终端。
在另一些实施例中,所述第一设备为核心网功能,例如具有数据存储功能的核心网功能,例如数据库、数据功能或网络存储功能(Network Repository Function,NRF)或UDM,或AI模型库,或AI模型管理功能或终端的第三方服务器等,第二设备是终端或接入网设备(例如基站),第三设备是接入网设备(例如基站)。
在又一些实施例中,所述第一设备为核心网功能,例如具有数据存储功能的核心网功能,例如数据库、数据功能或网络存储功能(Network Repository Function,NRF)或UDM,或AI模型库,或AI模型管理功能或终端的第三方服务器等,第二设备是终端或接入网设备(例如基站)、第三方服务器、核心网功能(例如AI控制功能、任务控制功能、协同控制功能或其他核心网功能等),第三设备是核心网的推理功能。
在本申请实施例中,该AI控制节点可以用于AI相关功能的控制,任务控制节点可以用于服务相关功能的控制,或者任务相关的控制功能,例如,该任务可以包括但不限于网络新能力涉及到多节点场景下连接、计算、数据和算法资源的协同和调配,用于共同完成某个特定的目标,协同控制节点可以用于多个功能(例如通信功能,数据功能,算力功能,算法功能,模型功能)之间的协同管理。
在本申请实施例中,终端和接入网设备之间的信息交互可以通过以下至少一种信令:层1信令、层2信令、层3信令、数据面信令、AI层信令。终端和核心网设备之间的信息交互可以通过以下至少一种信令中:NAS信令、数据面信令、AI层信令。终端和核心网设备之间可以直接进行信息的交互,或者,也可以通过其他设备进行信息的转发,例如通过通信控制功能(例如接入移动管理功能AMF)、AI控制功能、任务控制功能、协同控制功能、网络开放功能等进行信息的转发。
应理解,接入网设备和核心网设备,以及核心网设备之间可以直接进行信息的交互,或者,也可以通过其他设备进行信息的转发,例如通过通信控制功能(例如接入移动管理功能AMF)、AI控制功能、任务控制功能、协同控制功能、网络开放功能等进行信息的转发。
在一些实施例中,若所述第一设备为接入网设备,所述第二设备为终端,第一信息携带在以下至少一种信令中:层1信令、层2信令、层3信令、数据面信令、AI层信令。示例性地,该层1信令可以包括但不限于PDCCH。该层2信令例如可以包括但不限于下行MAC CE,该层3信令例如可以包括但不限于RRC信令。
在另一些实施例中,若所述第一设备为核心网功能,所述第二设备为终端,第一信息携带在以下至少一种信令中:NAS信令、数据面信令、AI层信令。
在一些实施例中,若所述第一设备为接入网设备,所述第三设备为终端,第二信息携带在以下至少一种信令中:层1信令、层2信令、层3信令、数据面信令、AI层信令。示例性地,该层1信令可以包括但不限于PDCCH。该层2信令例如可以包括但不限于下行MAC CE,该层3信令例如可以包括但不限于RRC信令。
在另一些实施例中,若所述第一设备为核心网功能,所述第三设备为终端,第二信息携带在以下至少一种信令中:NAS信令、数据面信令、AI层信令。
在本申请一些实施例中,所述第一设备也可以将第一信息发送给所述第三设备,用于辅助第三设备进行模型推理。例如第一信息可以携带在第二信息中,如第二信息包括第一信息中的内容。
在一些实施例中,所述第一信息包括以下至少之一:
用于所述第一模型进行推理的辅助信息;
用于所述第一模型进行推理的需求信息;
所述第一模型的输入信息;
知识信息的调用接口信息;
所述第二信息的接收设备指示;
所述第一模型的输入信息对应的样本指示。
因此,需求发起设备通过向知识存储设备指示用于模型推理的相关信息(如第一信息),知识存储设备可以基于该第一信息和知识信息确定第二信息,例如,根据第一信息获取专业领域的知识信息,从而得到第二信息,进一步向推理设备发送第二信息,这样推理设备在使用第二信息辅助进行模型的推理,能够实现专业领域任务的精准推理,降低模型推理中的幻觉问题,提升模型的推理性能。
在一些实施例中,接收设备指示可以是第二设备的接收设备的标识信息,用于第一设备确定将第二信息发送给哪个设备。
在一些实施例中,样本指示可以是第一信息中的第一模型的输入信息对应哪个样本,避免推理设备使用对应不同的样本的第一信息和第二信息进行模型推理,影响推理结果。例如,该样本指示可以是样本标识。
在一些实施例中,所述用于所述第一模型进行推理的辅助信息包括以下至少之一:
AI模型的标识信息、AI模型的功能指示、AI模型的特性指示、目标知识图谱指示、目标知识向量库指示、小区标识、带宽部分(Band Width Part,BWP)标识。
在一些实施例中,所述AI模型的标识信息例如可以是AI模型的模型标识(model ID)可以用于AI结构标识、AI算法标识,或者AI模型关联的特定数据集的标识,或者AI/ML相关的特定场景、环境、信道特征、设备的标识,或者AI/ML相关的功能、特性、能力或模块的标识,本申请对此不做具体限定。
在一些实施例中,AI模型的特性指示可以显式或隐式指示AI模型支持的特性,例如支持CSI预测和压缩反馈、波束预测、定位、负载均衡、资源分配等。通过向第一设备指示AI模型的特性,可以辅助第一设备获取该特性相关的知识信息,然后根据该知识信息获取相应的第二信息,用于辅助第三设备进行模型的推理。
在本申请实施例中,可以认为在具体配置下的AI模型的特性是AI模型的功能。
例如,AI模型的特性为波束预测,AI模型的功能为基站配置为32发送波束下的时间域波束预测。
在一些实施例中,AI模型的功能指示可以显式或隐式指示AI模型支持的功能,例如支持特定配置下的CSI预测和压缩反馈、波束预测、定位、负载均衡、资源分配等。通过向第一设备指示AI模型的功能,可以辅助第一设备获取该功能相关的知识信息,然后根据该知识信息获取相应的第二信息,用于辅助第三设备进行模型的推理。
在一些实施例中,所述目标知识图谱指示可以用于指示第二设备期待使用的知识图谱。通过向第一设备指示期望的目标知识图谱,可以辅助第一设备使用该目标知识图谱获取相应的第二信息,用于辅助第三设备进行模型的推理。
在一些实现方式中,目标知识图谱指示可以显式指示目标知识图谱。
示例性地,所述目标知识图谱指示用于以下至少之一:
目标知识图谱中的实体之间的关系;
目标知识图谱中的实体的属性信息;
目标知识图谱的数据结构;
目标知识图谱中的实体之间的相关性;
目标知识图谱中的语句的主体属性对象(Subject,Predicate,Object,SPO)三元组;
目标知识图谱文件。
在一些实施例中,目标知识图谱中的实体之间的关系可以采用如下格式表示:
(实体1,关系,实体2)。
可选地,实体之间的关系可以是包括但不限于上下位关系,继承关系等。
在一些实施例中,目标知识图谱中的实体的属性信息可以采用如下格式表示:
(实体,属性名,属性值)。
在一些实施例中,可以通过三元组(triple)来表示节点之间的关系,例如一种关系可以是(节点1,边,节点2)。
在一些实施例中,目标知识图谱中的实体之间的相关性可以采用如下格式表示:
(实体1,实体1和实体2之间的相关性系数,实体2)。
例如,实体1是小区吞吐量,实体2是最强波束的L1-RSRP,相关性系数为0.8。
在一些实施例中,目标知识图谱中的语句的SPO三元组用于描述语句的主语、谓语和宾语。
在另一些实现方式中,目标知识图谱指示可以隐式指示目标知识图谱。
示例性地,所述目标知识图谱指示包括但不限于以下至少之一:
知识图谱标识;
知识图谱文件地址。
在一些实施例中,可以预定义或预配置(例如网络侧设备预先指示)多个知识图谱标识,每个知识图谱标识对应一个知识图谱,例如,第二设备可以在该多个知识图谱标识中选择一个或多个知识图谱标识指示给第一设备,进一步地,该第一设备可以根据指示的知识图谱标识确定对应的知识图谱,进一步基于该知识图谱确定第二信息。
在一些实施例中,可以预定义或预配置(例如网络侧设备预先指示)多个知识图谱文件地址,每个知识图谱文件地址对应一个知识图谱文件,例如,第二设备可以在该多个知识图谱文件地址中选择一个或多个知识图谱文件地址指示给第一设备,进一步地,该第一设备可以根据指示的知识图谱文件地址获取对应的知识图谱文件,进一步基于该知识图谱文件确定第二信息。
在一些实施例中,以AI模型的任务是波束预测为例,知识图谱的使用方式可以为:
根据小区标识和目标任务,利用知识图谱可以检索出小区发送波束配置,例如发送波束的指向或赋形码本。
在一些实现方式中,目标知识向量库指示可以隐式指示第二设备期望使用的目标知识向量库。通过向第一设备指示期望的目标知识向量库,可以辅助第一设备使用该目标知识向量库获取相应的第二信息,用于辅助第三设备进行模型的推理。
示例性地,所述目标知识向量库指示包括但不限于以下至少之一:
知识向量库标识;
知识向量库文件地址。
在一些实施例中,可以预定义或预配置(例如网络侧设备预先指示)多个知识向量库标识,每个知识向量库标识对应一个知识向量库,例如,第二设备可以在该多个知识向量库标识中选择一个或多个知识向量库标识指示给第一设备,进一步地,该第一设备可以根据指示的知识向量库标识确定对应的知识向量库,进一步基于该知识向量库确定第二信息。
在一些实施例中,可以预定义或预配置(例如网络侧设备预先指示)多个知识向量库文件地址,每个知识向量库文件地址对应一个知识向量库文件,例如,第二设备可以在该多个知识向量库文件地址中选择一个或多个知识向量库文件地址指示给第一设备,进一步地,该第一设备可以根据指示的知识向量库文件地址获取对应的知识向量库文件,进一步基于该知识向量库文件确定第二信息。
在一些实施例中,以AI模型的任务是波束预测为例,知识向量库的使用方法可以为:
根据小区标识和目标任务,利用知识向量库获得小区发送波束配置,例如发送波束的指向或赋形码本相关的特征向量。
在一些实施例中,该小区标识可以用于第一设备获取该小区标识对应的小区的知识信息,例如小区配置等,进一步地,可以根据该知识信息确定第二信息。
在一些实施例中,该BWP标识可以用于第一设备获取该BWP标识对应的BWP的知识信息,例如BWP配置等,进一步地,可以根据该知识信息确定第二信息。
在一些实施例中,所述用于所述第一模型进行推理的需求信息包括以下至少之一:
AI模型所需适配的目标任务指示;
AI模型的业务体验质量(Quality of Experience,QoE)要求;
AI模型的业务处理时延;
AI模型的业务处理准确度;
AI模型的业务计算量;
AI模型的数据处理规模。
例如,第一设备可以根据该目标任务指示选择与该目标任务相关的知识信息,然后根据该知识信息确定第二信息。
又例如,第二设备可以根据QoE要求、或处理时延、处理准确度或处理规模等,选择满足QoE要求、或处理时延、处理准确度或处理规模的知识信息,然后根据该知识信息确定第二信息。
在一些实施例中,所述第一模型的输入信息与第一模型所需适配的目标任务有关,以目标任务是波束预测为例,该输入信息可以包括小区标识和小区发送波束的测量质量。
在一些实施例中,所述知识信息的调用接口信息包括但不限于如下至少之一:
调用接口名称、调用接口的输入数据格式、调用接口的输出数据格式、调用接口的输入数据。其中,该输入数据格式、输出数据格式、输入数据与领域、任务、场景等相关。
在一些实施例中,调用接口可以包括但不限于应用程序接口(Application Programming Interface,API)等访问外部设备的接口。
通信系统会不断地迭代更新,通信系统的知识也会更新。在网络部署之后,更新的系统知识,例如来自3GPP等组织的最新知识可以存储在外部的数据库中,此情况下,通过调用API接口的方式,为通信系统提供服务,提升了本申请技术方案的适用性。
在一些实施例中,第二设备和第三设备都为终端,第一信息可以是在特定情况下发送的,例如:
在发生以下事件中的至少之一时,第二设备向第一设备发送所述第一信息:
事件1:接收到第一指示,所述第一指示用于指示终端进行模型的激活或切换;
事件2:所述终端确定进行模型的激活或切换;
事件3:所述终端确定当前使用的模型不满足需求;
事件4:接收到第二指示,所述第二指示用于指示当前终端使用的模型不满足需求。
在一些实施例中,触发第二设备向第一设备发送第一信息的事件可以是预定义的,或者,网络侧设备配置的,当满足触发第一信息的事件时,第二设备可以将该事件的标识信息发送给网络侧设备,例如,携带在第一信息中发送给网络侧设备,从而网络侧设备可以获知第二设备上发生的事件。
在一些实施例中,第一指示可以是网络侧设备发送的,例如,该网络侧设备可以控制AI模型的激活或切换,比如,网络侧设备可以在当前使用的模型不满足需求时,向终端发送第一指示,从而第二设备可以根据第一指示发送第一信息。
在一些实施例中,终端可以在当前使用的模型的性能不满足需求时,自主确定进行模型的激活或切换。
在一些实施例中,第二指示可以是网络侧设备发送的,例如,网络侧设备可以在当前使用的模型不满足需求时,向终端发送第二指示,用于指示终端当前使用的模型不满足需求,进一步地,该终端可以根据该第二指示自主判断是否进行模型的激活或切换。
在一些实施例中,模型不满足需求可以指模型不满足需求信息,例如可以包括但不限于以下至少之一:模型的推理不满足业务体验质量(Quality of Experience,QoE)不满足QoE要求,模型的业务处理时延不满足时延需求,模型的业务处理准确度不满足准确度需求。
在本申请一些实施例中,所述第二信息包括以下至少之一:
隐藏变量,所述隐藏变量是对所述第一模型的输入信息进行映射后的特征向量;
提示信息;
网络侧设备的配置信息;
场景类型信息;
所述第一模型的输入信息对应的样本指示,或者,确定所述第二信息所使用的第一信息中的样本指示。
在一些实施例中,样本指示可以用于指示第二信息是根据哪个样本确定的,该样本指示可以是样本标识。这样,在推理设备获知第一信息和第二信息时,可以通过该样本指示确定第一信息和第二信息对应的样本,进一步在进行模型推理时,可以使用对应相同样本的第一信息和第二信息,避免使用对应不同样本的第一信息和第二信息进行模型推理,影响推理结果。
在一些实施例中,隐藏变量可以指将具有物理含义的输入信息,经过编码器,映射得到的低维特征,或者,原始输入经过去除冗余后的特征向量。
例如,如图3所示,自编码器包括编码器和解码器,编码器将输入的图像映射到低维特征(图3中的隐藏变量),解码器基于隐藏变量去还原输入图像。其中隐藏变量包含了输入图像的特征信息。
在一些实施例中,提示信息可以用于提示或引导第一模型推理出符合预期的输出,提升模型的推理性能。
示例性地,所述提示信息包括以下至少之一:
与所述第一模型适配的目标任务相关的思维链;
与所述目标任务相关的提示文字;
与所述目标任务相关的提示文件;
与所述目标任务相关的提示代码;
与所述目标任务相关的思维链标识,用于标识一个思维链;
与所述目标任务相关的提示文字标识,用于标识一段提示文字;
与所述目标任务相关的提示文件标识,用于标识一个提示文件;
与所述目标任务相关的提示代码标识,用于标识一段提示代码;
与所述目标任务相关的提示文件地址。
在一些实施例中,所述网络侧设备的配置信息包括但不限于以下至少之一:
网络侧设备的发送天线个数;
网络侧设备的发送波束个数;
网络侧设备的天线端口个数;
网络侧设备的发送功率;
网络侧设备的天线增益;
网络侧设备的波束3dB带宽;
网络侧设备的间距;
网络侧设备的频率;
系统带宽;
网络侧设备覆盖下的终端分布特性。
在一些实施例中,所述场景类型信息包括以下至少之一:
视距(line of sight,LOS)、非视距(non line of sight,NLOS)、室外(outdoor)、室内(indoor)、城区微小区(urban microcell,UMi)、城区宏小区(urban maceocell,UMa)、郊区/农村宏小区(rural macrocell,RMa)、郊区/农村微小区(rural microcell,RMi)、高速、低速。
在一些实施例中,第二信息可以是第一设备确定的,例如,第一设备根据第一信息和第一设备上存储的知识信息确定的,或者,根据第一信息和调用的知识信息确定的,例如第一设备可以使用调用接口信息从目标设备获取知识信息,进一步根据该知识信息,或者第一信息和该知识信息确定第二信息。
在另一些实施例中,第二信息也可以是使用调用接口信息所需访问的目标设备确定。可选地,目标设备确定第二信息后,可以直接发送给第三设备,或者,也可以通过其他设备发送给第三设备,例如通过网络开发功能转发,或者,通过第一设备发送给第三设备。
以下,结合具体实施例,对第三设备获取第二信息的具体实现进行说明。
实施例1:第一设备上存储有用于辅助第一模型进行推理的知识信息。
此情况下,在第一设备从第二设备接收第一信息之后,第一设备可以根据第一信息结合该第一设备上存储的知识信息,确定第二信息。进一步地,第一设备可以将第二信息发送给第三设备,从而第三设备可以在使用第一模型进行推理时可以使用第二信息作为辅助信息,能够提升模型的推理性能。
实施例2:第一设备上未存储用于辅助第一模型进行推理的知识信息,并且第一信息包括知识信息的调用接口信息。此情况下,如图4所示,可以通过如下步骤实现:
S211,第二设备向第一设备发送第一信息,第一信息包括知识信息的调用接口信息,该调用接口信息可以是第二设备期望使用的知识信息的调用接口信息。此时,知识信息的调用接口信息包括了调用接口名称和调用接口的输入数据。
S212,第一设备根据该知识信息的调用接口信息,向目标设备发送第一信息,该目标设备是使用调用接口信息所需访问的设备;
S213,目标设备根据第一信息确定第二信息;
S214,目标设备向第一设备发送第二信息;
S215,第一设备向第三设备发送第二信息。
应理解,第一设备和目标设备之间的第一信息和第二信息的交互可以通过其他节点(例如网络开放功能)转发,或者,直接进行信息的交互。
实施例3:第一设备上未存储用于辅助第一模型进行推理的知识信息,并且第一信息不包括知识信息的调用接口信息。
此情况下,作为一个实施例,如图5所示,可以通过如下步骤实现:
S221,第二设备向第一设备发送第一信息,第一信息不包括知识信息的调用接口信息。
S222,第一设备根据第一信息,确定知识信息的调用接口信息。
例如,第一设备可以根据第一信息中的用于第一模型推理的需求信息,在多个调用接口中确定目标调用接口,从而得到该目标调用接口的调用接口名称、调用接口的输入数据格式、调用接口的输出数据格式、调用接口的输入数据等信息。
S223,第一设备向目标设备发送第三信息,第三信息包括所述知识信息的调用接口信息。其中,该目标设备是使用所述调用接口信息所需访问的设备。所述知识信息的调用接口信息包括了调用接口名称和调用接口的输入数据。例如,该调用接口名称根据第一信息中的需求信息或辅助信息或调用接口信息确定,调用接口的输入数据是基于所述第一信息中的第一模型的输入信息确定。
S224,目标设备根据第三信息,确定第二信息;
S225,目标设备向第一设备发送第二信息;
S226,第一设备向第三设备发送第二信息。
如,在第二设备未指示调用接口信息的情况下,第一设备可以自主确定知识信息的调用接口信息,进一步将该调用接口信息,或者调用接口信息和第一信息发送给目标设备,从而目标设备可以基于该调用接口信息获取相应的知识信息,从而得到第二信息。然后目标设备可以将确定的第二信息发送给第一设备,由第一设备将该第二信息发送给第三设备。
应理解,第一设备和目标设备之间的第三信息和第二信息的交互可以通过其他节点(例如网络开放功能)转发,或者,直接进行信息的交互。
此情况下,作为另一个实施例,如图6所示,可以通过如下步骤实现:
S231,第二设备向第一设备发送第一信息,第一信息不包括知识信息的调用接口信息。
S232,第一设备根据第一信息,确定知识信息的调用接口信息。
例如,第一设备可以根据第一信息中的用于第一模型推理的需求信息,在多个调用接口中确定目标调用接口,从而得到该目标调用接口的调用接口名称、调用接口的输入数据格式、调用接口的输出数据格式、调用接口的输入数据等信息。
S233,第一设备向第三设备发送第四信息,第四信息包括所述知识信息的调用接口信息,或者,第四信息包括第一信息和所述知识信息的调用接口信息。其中,所述知识信息的调用接口信息包括目标调用接口的调用接口名称,或者目标调用接口的调用接口名称和调用接口的输入数据格式,或者目标调用接口的调用接口名称和调用接口的输入数据格式和调用接口的输出数据格式。
S234,第三设备根据第四信息向目标设备发送第七信息,其中,第七信息包括所述知识信息的调用接口信息;其中,所述知识信息的调用接口信息包括了调用接口名称和调用接口的输入数据。例如,该调用接口名称包含在第四信息的所述知识信息的调用接口信息中,调用接口的输入数据是基于第四信息中的所述知识信息的调用接口的输入数据格式和第三设备可提供的所述第一模型的输入信息确定。或者,该调用接口名称根据第一信息(包含在第四信息中)中的需求信息或辅助信息或调用接口信息确定,调用接口的输入数据是基于所述第一信息(包含在第四信息中)中的第一模型的输入信息确定。
S235,目标设备根据第七信息确定第二信息,并向第三设备发送第二信息。
如,在第二设备未指示调用接口信息的情况下,第一设备可以自主确定知识信息的调用接口信息,进一步将该调用接口信息,或者该调用接口信息和第一信息发送给第三设备,由第三设备自主向目标设备请求第二信息,然后目标设备可以将确定的第二信息发送给第三设备。
应理解,第三设备和目标设备之间的第七信息和第二信息的交互可以通过其他节点(例如网络开放功能)转发,或者,直接进行信息的交互。
在本申请一些实施例中,所述方法200还包括以下步骤至少之一:
第一设备接收所述目标设备发送的第五信息;
第一设备根据所述第五信息,登记至少一个调用接口信息;
第一设备向所述目标设备发送第六信息。
在一些实施例中,所述第五信息包括但不限于以下至少之一:
调用接口名称,调用接口参数列表,调用接口描述,调用接口的使用样例,调用接口指令集合;其中,调用接口参数列表包括调用接口的输入参数和输出参数,每个参数包括参数名称,参数描述,数据类型,缺省值中的至少之一。调用接口描述用于说明调用接口的功能。
在一些实施例中,所述第六信息包括但不限于以下至少之一:
是否成功登记调用接口,拒绝登记调用接口,已登记的调用接口列表,失败登记的调用接口列表。
进一步地,在后续第一信息不包括调用接口信息时,第一设备可以根据第一信息在登记的调用接口信息中确定目标调用接口信息,使用该目标调用接口信息获取相应的知识信息。
在本申请一些实施例中,若所述第三设备为终端,所述方法200还包括:
所述第三设备向网络侧设备发送第八信息和第一能力信息中的至少之一;
其中,所述第三设备上部署所述第一模型,所述第八信息用于指示所述第三设备基于所述第一模型进行推理所需的辅助信息,所述第一能力信息用于指示所述第三设备基于所述第一模型进行推理的能力信息。
在一些实施例中,该第八信息或第一能力信息可以用于网络侧设备确定向终端指示的第二信息。
在一些实施例中,所述第八信息包括以下至少之一:
第三设备基于所述第一模型进行推理所需的知识信息所在的网元指示;
第三设备基于所述第一模型进行推理所需的知识信息指示;
第三设备基于所述第一模型进行推理所需的知识信息的调用接口指示;
所述第二信息的预处理方式指示;
所述第二信息的格式指示。
可选地,第三设备基于所述第一模型进行推理所需的知识信息所在的网元指示可以是网元标识。例如,若有多个网元用于存储知识信息,第三设备进行模型推理所需的知识存储在第一网元上,则第三设备可以向网络侧设备指示第一网元的网元标识。
在一些实施例中,第三设备基于所述第一模型进行推理所需的知识信息指示可以包括第三设备基于所述第一模型进行推理所需的知识图谱的指示信息和/或知识向量库的指示信息。例如,知识图谱的指示信息是一个或多个知识图谱标识,或者,知识图谱组标识、知识图谱文件地址等,或者显式指示一个或多个知识图谱。例如,知识向量库的指示信息用于指示一个或多个知识向量库标识、或者显式指示一个或多个知识向量库。
示例性地,所述知识图谱的指示信息包括以下至少之一:
知识图谱标识;
知识图谱文件地址。
在一些实施例中,第三设备基于所述第一模型进行推理所需的知识信息的调用接口指示可以显式指示一个或多个调用接口信息,或者,也可以隐式指示一个或多个调用接口标识,该调用接口标识对应一组调用接口信息。
在一些实施例中,所述第二信息的格式指示可以用于指示第二信息的模型输入格式。例如,该模型输入格式可以包括模型输入包括的内容(例如包括提示信息,包括提示信息和调用接口信息,包括提示信息和网络侧设备的配置信息等),以及第二信息的维度,量化精度,大小限制等。
在一些实施例中,所述第二信息的预处理方式指示可以用于指示对第二信息进行预处理的方式,例如对数值做偏置,对数值做缩放,对数值做多项式函数变化,对单个数值做多维映射等。
在一些实施例中,所述第一能力信息包括以下至少之一:
第三设备基于所述第一模型进行推理所支持的知识信息指示;
第三设备基于所述第一模型进行推理所支持的知识信息的调用接口指示;
支持的所述第二信息的预处理方式;
支持的所述第二信息的格式指示。
在一些实施例中,第三设备基于所述第一模型进行推理所支持的知识信息指示可以包括第三设备基于所述第一模型进行推理所支持的知识图谱的指示信息和/或知识向量库的指示信息。例如,知识图谱的指示信息是一个或多个知识图谱标识,或者,知识图谱组标识、知识图谱文件地址等,或者显式指示一个或多个知识图谱。例如,知识向量库的指示信息用于指示一个或多个知识向量库标识、或者显式指示一个或多个知识向量库。
在一些实施例中,第三设备基于所述第一模型进行推理所支持的知识信息的调用接口指示可以显式指示一个或多个调用接口信息,或者,也可以隐式指示一个或多个调用接口标识,该调用接口标识对应一组调用接口信息。
在一些实施例中,所述第三设备支持的第二信息的格式指示可以用于指示第三设备支持的第二信息的模型输入格式,例如该模型输入格式可以包括模式输入包括哪些内容(例如包括提示信息,包括提示信息和调用接口信息,包括提示信息和网络侧设备的配置信息等),以及这些信息的维度,量化精度,大小限制等。
在一些实施例中,所述第三设备支持的所述第二信息的预处理方式可以用于指示第三设备支持的处理方法。这些处理方法是网络侧设备对第二信息进行预处理的方式,例如对数值做偏置,对数值做缩放,对数值做多项式函数变化,对单个数值做多维映射等。
在一些实施例中,第三设备在模型注册(或者说,模型识别(model identification))过程中或能力上报过程中,向所述网络侧设备发送所述第八信息和/或所述第一能力信息。
应理解,第八信息和第一能力信息可以是通过一个信令上报给网络侧设备的,或者,也可以是通过不同的信令上报给网络侧设备的,本申请对此不作限定。
以下,结合图7至图14所示具体实施例,说明本申请提供的通信系统中的模型推理方法。
实施例1:推理设备是终端
如图7所示,第一设备为接入网设备或核心网功能,第二设备和第三设备同一设备,并且为终端,该终端上部署第一模型。该推理方法可以包括如下步骤:
步骤1:终端向接入网设备或核心网设备发送第一信息,用于请求获取第二信息。
步骤2:接入网设备或核心网设备根据第一信息向终端发送第二信息。
从而推理设备可以获知该第二信息,进一步地,推理设备可以使用该第二信息辅助进行模型的推理,能够提升模型的推理性能。
实施例2:推理设备是接入网设备,例如基站。
例如,在图8的示例中,第一设备为核心网功能,第二设备和第三设备为同一设备,并且为接入网设备,该接入网设备上部署第一模型。
如图8所示,该推理方法可以包括如下步骤:
步骤1:接入网设备向核心网设备发送第一信息,用于请求获取第二信息。
步骤2:核心网设备根据第一信息向接入网设备发送第二信息。
从而推理设备可以获知该第二信息,进一步地,推理设备可以使用该第二信息辅助进行模型的推理,能够提升模型的推理性能。
又例如,在图9的示例中,第一设备为核心网功能,第二设备为终端,第三设备为接入网设备,该接入网设备上部署第一模型,如推理设备为接入网设备,例如基站。
如图9所示,该推理方法可以包括如下步骤:
步骤1:终端向核心网功能发送第一信息,用于请求获取第二信息。
可选地,在步骤1中,终端还可以将第一信息发送给接入网设备。
可选地,终端发给接入网设备的第一信息与发送给核心网功能的第一信息可以不同。
例如,终端发送给核心网的第一信息可以包括用于所述第一模型进行推理的辅助信息和所述第一模型的输入信息;而终端发送给接入网设备的第一信息仅包括所述第一模型的输入信息。此外,为了让核心网设备知道将第二信息发给谁,第一信息还可以包括接收设备指示,用于指示第二信息的接收设备,如第三设备。
步骤2:核心网设备根据第一信息向接入网设备发送第二信息。
此时,接入网设备用于第一模型的输入包括两部分,一部分来自终端发送的第一信息,一部分来自核心网发送的第二信息。因此,接入网设备需要将两部分信息进行同步从而生成完整的第一模型的输入信息。从而防止将样本1的第一信息和样本2的第二信息进行合并而生成错误的样本。因此,此时第一信息和第二信息中需要包含样本指示,例如样本标识。
可选地,核心网设备在向接入网设备发送第二信息时,也可以将该第一信息一起发送给接入网设备。从而推理设备可以获知该第一信息和第二信息,进一步地,推理设备可以使用该第一信息和第二信息辅助进行模型的推理,能够提升模型的推理性能。
又例如,在图10的示例中,第一设备为核心网功能,第二设备为终端,第三设备为接入网设备,该接入网设备上部署第一模型,如推理设备为接入网设备,例如基站。
如图10所示,该推理方法可以包括如下步骤:
步骤1:终端向核心网设备发送第一信息,用于请求获取第二信息。
步骤2:核心网设备根据第一信息向终端发送第二信息。
步骤3:终端向接入网设备发送第二信息。
可选地,终端还可以将第一信息一起发送给接入网设备。
如,在该示例中,第二信息不是由知识存储设备直接发送给推理设备的,而是由需求发起设备转发给推理设备的。
从而推理设备可以获知该第一信息和第二信息,进一步地,推理设备可以使用该第一信息和第二信息辅助进行模型的推理,能够提升模型的推理性能。
又例如,在图11的示例中,第一设备为核心网功能,第二设备为终端,第三设备为接入网设备,该接入网设备上部署第一模型,如推理设备为接入网设备,例如基站。
如图11所示,该推理方法可以包括如下步骤:
步骤1:终端向核心网设备发送第一信息,用于请求获取第二信息。
其中,第一信息包括如下至少之一:
接收设备指示,用于指示第二信息的接收设备。
用于所述第一模型进行推理的辅助信息;
用于所述第一模型进行推理的需求信息;
所述第一模型的输入信息。步骤2:核心网设备根据第一信息向接入网设备发送第一信息和第二信息。
从而推理设备可以获知该第一信息和第二信息,进一步地,推理设备可以使用该第一信息和第二信息辅助进行模型的推理,能够提升模型的推理性能。
实施例3:推理设备是核心网功能。
例如,在图12的示例中,第一设备为核心网功能,第二设备为终端或接入网设备或核心网设备,第三设备为核心网功能。
如图12所示,该推理方法可以包括如下步骤:
步骤1:第二设备向第一设备发送第一信息,用于请求获取第二信息。
其中,第一信息包括如下至少之一:
接收设备指示,用于指示第二信息的接收设备;
用于所述第一模型进行推理的辅助信息;
用于所述第一模型进行推理的需求信息;
所述第一模型的输入信息。
可选地,在步骤1中,第二设备还可以将第一信息也发送给第三设备,用于辅助第三设备进行模型推理。此时,第三设备用于第一模型的输入包括两部分,一部分来自第二设备发送的第一信息,一部分来自第一设备发送的第二信息。因此,第三设备需要将两部分信息进行同步从而生成完整的第一模型的输入信息。从而防止将样本1的第一信息和样本2的第二信息进行合并而生成错误的样本。因此,此时第一信息和第二信息中需要包含样本指示,例如样本标识。
步骤2:第一设备根据第一信息向第三设备发送第二信息。
从而推理设备可以获知该第一信息和第二信息,进一步地,推理设备可以使用该第一信息和第二信息辅助进行模型的推理,能够提升模型的推理性能。
又例如,在图13的示例中,第一设备为核心网功能,第二设备为终端或接入网设备或核心网设备,第三设备为核心网功能。
如图13所示,该推理方法可以包括如下步骤:
步骤1:第二设备向第一设备发送第一信息,用于请求获取第二信息。
步骤2:第一设备根据第一信息向第二设备发送第二信息。
步骤3:第二设备向第三设备发送第二信息。可选地,还可以将第一信息也发送给第三设备。
从而推理设备可以获知该第二信息,或者第一信息和第二信息,进一步地,推理设备可以使用该第二信息,或者第一信息和第二信息辅助进行模型的推理,能够提升模型的推理性能。
又例如,在图14的示例中,第一设备为核心网功能,第二设备为终端或接入网设备或核心网设备,第三设备为核心网功能。
如图14所示,该推理方法可以包括如下步骤:
步骤1:第二设备向第一设备发送第一信息,用于请求获取第二信息。
其中,第一信息包括如下至少之一:
接收设备指示,用于指示第二信息的接收设备;
用于所述第一模型进行推理的辅助信息;
用于所述第一模型进行推理的需求信息;
所述第一模型的输入信息。
步骤2:第一设备根据第一信息向第三设备发送第一信息和第二信息。
从而推理设备可以获知该第二信息,或者第一信息和第二信息,进一步地,推理设备可以使用该第二信息,或者第一信息和第二信息辅助进行模型的推理,能够提升模型的推理性能。
应理解,在本申请实施例中,在推理设备获知第一信息和第二信息的情况下,该第一信息和第二信息中包括样本指示,用于避免推理设备使用不同样本的第一信息和第二信息进行推理,影响推理结果。
综上,在本申请实施例中,通过向存储有知识信息的设备获取第二信息,进一步使用模型进行推理时,可以获得更多的信息用作模型的输入,从而能够提升模型的推理性能。
上文结合图2至图14,详细描述了本申请的方法实施例,下文结合图15至图21,详细描述本申请的装置实施例,应理解,装置实施例与方法实施例相互对应,类似的描述可以参照方法实施例。
本申请实施例提供的通信系统中的模型推理方法,执行主体可以为通信装置。本申请实施例中以通信装置执行通信系统中的模型推理方法为例,说明本申请实施例提供的通信装置。
图15示出了根据本申请实施例的通信装置500的示意性框图。如图15所示,该装置500包括:
通信单元510,用于接收第二设备发送的第一信息,其中,所述通信装置500上存储有知识信息和知识信息的调用接口信息中的至少之一;其中,所述知识信息包括以下至少之一:知识图谱、知识向量库;以及
向第三设备发送第二信息,所述第三设备上部署第一模型,所述第二信息是根据所述第一信息和所述通信装置500上存储的或调用的知识信息确定的;
其中,所述第一信息包括以下至少之一:
用于所述第一模型进行推理的辅助信息;
用于所述第一模型进行推理的需求信息;
所述第一模型的输入信息;
知识信息的调用接口信息;
所述第二信息的接收设备指示;
所述第一模型的输入信息对应的样本指示。
在一些实施例中,所述辅助信息包括以下至少之一:
人工智能AI模型的标识信息、AI模型的功能指示、AI模型的特性指示、目标知识图谱指示、小区标识、带宽部分BWP标识。
在一些实施例中,所述需求信息包括以下至少之一:
AI模型所需适配的目标任务指示;
AI模型的业务体验质量QoE要求;
AI模型的业务处理时延;
AI模型的业务处理准确度;
AI模型的业务计算量;
AI模型的数据处理规模。
在一些实施例中,所述第二信息包括以下至少之一:
隐藏变量,所述隐藏变量是对所述第一模型的输入信息进行映射后的特征向量;
提示信息;
网络侧设备的配置信息;
场景类型信息;
所述第一模型的输入信息对应的样本指示。
在一些实施例中,所述提示信息包括以下至少之一:
与所述第一模型适配的目标任务相关的思维链;
与所述目标任务相关的提示文字;
与所述目标任务相关的提示文件;
与所述目标任务相关的提示代码;
与所述目标任务相关的思维链标识;
与所述目标任务相关的提示文字标识;
与所述目标任务相关的提示文件标识;
与所述目标任务相关的提示代码标识;
与所述目标任务相关的提示文件地址。
在一些实施例中,所述网络侧设备的配置信息包括以下至少之一:
网络侧设备的发送天线个数;
网络侧设备的发送波束个数;
网络侧设备的天线端口个数;
网络侧设备的发送功率;
网络侧设备的天线增益;
网络侧设备的波束3dB带宽;
网络侧设备的间距;
网络侧设备的频率;
系统带宽;
网络侧设备覆盖下的终端分布特性。
在一些实施例中,所述场景类型信息包括以下至少之一:
视距LOS、非视距NLOS、室内、室外、城区微小区、城区宏小区、郊区微小区、高速、低速。
在一些实施例中,所述通信装置500还包括:
处理单元,用于根据所述第一信息和所述通信装置500上存储的知识信息确定所述第二信息。
在一些实施例中,所述通信单元510还用于:
从目标设备接收所述第二信息,所述目标设备是使用所述调用接口信息所访问的设备。
在一些实施例中,所述第一信息包括知识信息的调用接口信息,所述通信装置500还包括:处理单元,用于根据所述第一信息和所述通信装置500上存储的知识信息确定所述第二信息。
在一些实施例中,所述通信单元510还用于:根据所述知识信息的调用接口信息,向目标设备发送所述第一信息,其中,所述目标设备是使用所述调用接口信息所访问的设备;以及从所述目标设备接收所述第二信息。
在一些实施例中,所述第一信息不包括知识信息的调用接口信息,所述通信装置500还包括:处理单元,用于根据所述第一信息,确定所述知识信息的调用接口信息;
所述通信装置500还用于:向目标设备发送第三信息,其中,所述目标设备是使用所述调用接口信息所访问的设备,所述第三信息包括所述知识信息的调用接口信息;以及,从所述目标设备接收所述第二信息。
在一些实施例中,所述第一信息不包括知识信息的调用接口信息,所述通信装置500还包括:处理单元,用于根据所述第一信息,确定所述知识信息的调用接口信息;
所述通信单元510还用于:向所述第三设备发送第四信息,所述第四信息包括所述知识信息的调用接口信息,或者,所述第四信息包括所述第一信息和所述知识信息的调用接口信息,所述第四信息用于所述第三设备向目标设备请求获取所述第二信息,所述目标设备是使用所述调用接口信息所访问的设备。
在一些实施例中,所述通信单元510还用于:
从所述目标设备接收第五信息;
所述通信装置500还包括处理单元,用于根据所述第五信息,登记至少一个调用接口信息;
所述通信单元510还用于:向所述目标设备发送第六信息;
其中,所述第五信息包括以下至少之一:调用接口名称,调用接口参数列表,调用接口描述,调用接口的使用样例,调用接口指令集合;
所述第六信息包括以下至少之一:
是否成功登记调用接口,拒绝登记调用接口,已登记的调用接口列表,失败登记的调用接口列表。
在一些实施例中,所述通信单元510还用于:
向所述第三设备发送所述第一信息。
在一些实施例中,所述通信装置500为接入网设备或核心网功能,所述第二设备和所述第三设备为终端;或者
所述通信装置500为核心网功能,所述第二设备为终端或接入网设备,所述第三设备为接入网设备;或者
所述通信装置500为核心网功能,所述第二设备为终端或接入网设备或核心网功能,所述第三设备为核心网功能。
在一些实施例中,所述通信装置500为接入网设备,所述第二设备为终端,所述第一信息携带在以下至少一种信令中:层1信令、层2信令、层3信令、AI层信令、数据面信令;或者
所述通信装置500为核心网功能,所述第三设备为终端,所述第一信息携带在以下至少一种信令中:非接入层NAS信令、AI层信令、数据面信令。
在一些实施例中,所述调用接口信息包括如下至少之一:
调用接口名称、调用接口的输入数据格式、调用接口的输出数据格式、调用接口的输入数据。
可选地,在一些实施例中,上述通信单元可以是通信接口或收发器,或者是通信芯片或者片上系统的输入输出接口。
应理解,根据本申请实施例的通信装置500可对应于本申请方法实施例中的第一设备,并且通信装置500中的各个单元的上述和其它操作和/或功能分别为了实现图2至图14中所示方法实施例中第一设备的相应流程,并达到相同的技术效果,为避免重复,这里不再赘述。
图16示出了根据本申请实施例的通信装置600的示意性框图。如图16所示,该装置600包括:
通信单元610,用于向第一设备发送第一信息,其中,所述第一设备上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,所述知识信息包括以下至少之一:知识图谱、知识向量库;
其中,所述第一信息包括以下至少之一:
用于第一模型进行推理的辅助信息;
用于第一模型进行推理的需求信息;
第一模型的输入信息;
知识信息的调用接口信息;
所述第二信息的接收设备指示;
所述第一模型的输入信息对应的样本指示。
在一些实施例中,所述辅助信息包括以下至少之一:
人工智能AI模型的标识信息、AI模型的功能指示、AI模型的特性指示、目标知识图谱指示、小区标识、带宽部分BWP标识。
在一些实施例中,所述需求信息包括以下至少之一:
AI模型所需适配的目标任务指示;
AI模型的业务体验质量QoE要求;
AI模型的业务处理时延;
AI模型的业务处理准确度;
AI模型的业务计算量;
AI模型的数据处理规模。
在一些实施例中,所述第二信息包括以下至少之一:
隐藏变量,所述隐藏变量是对所述第一模型的输入信息进行映射后的特征向量;
提示信息;
网络侧设备的配置信息;
场景类型信息;
所述第一模型的输入信息对应的样本指示。
在一些实施例中,所述提示信息包括以下至少之一:
与所述第一模型适配的目标任务相关的思维链;
与所述目标任务相关的提示文字;
与所述目标任务相关的提示文件;
与所述目标任务相关的提示代码;
与所述目标任务相关的思维链标识;
与所述目标任务相关的提示文字标识;
与所述目标任务相关的提示文件标识;
与所述目标任务相关的提示代码标识;
与所述目标任务相关的提示文件地址。
在一些实施例中,所述网络侧设备的配置信息包括以下至少之一:
网络侧设备的发送天线个数;
网络侧设备的发送波束个数;
网络侧设备的天线端口个数;
网络侧设备的发送功率;
网络侧设备的天线增益;
网络侧设备的波束3dB带宽;
网络侧设备的间距;
网络侧设备的频率;
系统带宽;
网络侧设备覆盖下的终端分布特性。
在一些实施例中,所述场景类型信息包括以下至少之一:
视距LOS、非视距NLOS、室内、室外、城区微小区、城区宏小区、郊区微小区、高速、低速。
在一些实施例中,所述调用接口信息包括如下至少之一:
调用接口名称、调用接口的输入数据格式、调用接口的输出数据格式、调用接口的输入数据。
在一些实施例中,所述通信装置600为终端,所述通信单元610具体用于:
在发生以下事件中的至少之一时,向所述第一设备发送所述第一信息:
接收到第一指示,所述第一指示用于指示激活或切换至所述第一模型;
所述通信装置600确定激活或切换至所述第一模型;
所述通信装置600确定所述第一模型不满足需求;
接收到第二指示,所述第二指示用于指示所述第一模型不满足需求。
在一些实施例中,所述通信单元610还用于:
向第三设备发送所述第一信息,所述第三设备上部署所述第一模型。
可选地,在一些实施例中,上述通信单元可以是通信接口或收发器,或者是通信芯片或者片上系统的输入输出接口。
应理解,根据本申请实施例的通信装置600可对应于本申请方法实施例中的第二设备,并且通信装置600中的各个单元的上述和其它操作和/或功能分别为了实现图2至图14中所示方法实施例中第二设备的相应流程,并达到相同的技术效果,为避免重复,这里不再赘述。
图17示出了根据本申请实施例的通信装置700的示意性框图。如图17所示,该装置700包括:
通信单元710,用于从第一设备或目标设备获取第二信息,其中,所述通信装置700上部署第一模型,所述第一设备或所述目标设备上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,所述目标设备是使用所述调用接口信息所访问的设备,所述知识信息包括以下至少之一:知识图谱、知识向量库;
其中,所述第二信息根据第一信息和所述第一设备上存储或调用的知识信息确定;
其中,所述第一信息包括以下至少之一:
用于所述第一模型进行推理的辅助信息;
用于所述第一模型进行推理的需求信息;
所述第一模型的输入信息;
知识信息的调用接口信息;
所述第二信息的接收设备指示;
所述第一模型的输入信息对应的样本指示。
在一些实施例中,所述辅助信息包括以下至少之一:
人工智能AI模型的标识信息、AI模型的功能指示、AI模型的特性指示、目标知识图谱指示、小区标识、带宽部分BWP标识。
在一些实施例中,所述需求信息包括以下至少之一:
AI模型所需适配的目标任务指示;
AI模型的业务体验质量QoE要求;
AI模型的业务处理时延;
AI模型的业务处理准确度;
AI模型的业务计算量;
AI模型的数据处理规模。
在一些实施例中,所述第二信息包括以下至少之一:
隐藏变量,所述隐藏变量是对所述第一模型的输入信息进行映射后的特征向量;
提示信息;
网络侧设备的配置信息;
场景类型信息;
所述第一模型的输入信息对应的样本指示。
在一些实施例中,所述提示信息包括以下至少之一:
与所述第一模型适配的目标任务相关的思维链;
与所述目标任务相关的提示文字;
与所述目标任务相关的提示文件;
与所述目标任务相关的提示代码;
与所述目标任务相关的思维链标识;
与所述目标任务相关的提示文字标识;
与所述目标任务相关的提示文件标识;
与所述目标任务相关的提示代码标识;
与所述目标任务相关的提示文件地址。
在一些实施例中,所述网络侧设备的配置信息包括以下至少之一:
网络侧设备的发送天线个数;
网络侧设备的发送波束个数;
网络侧设备的天线端口个数;
网络侧设备的发送功率;
网络侧设备的天线增益;
网络侧设备的波束3dB带宽;
网络侧设备的间距;
网络侧设备的频率;
系统带宽;
网络侧设备覆盖下的终端分布特性。
在一些实施例中,所述场景类型信息包括以下至少之一:
视距LOS、非视距NLOS、室内、室外、城区微小区、城区宏小区、郊区微小区、高速、低速。
在一些实施例中,所述通信单元710还用于:
从所述第一设备接收第四信息,所述第四信息包括所述知识信息的调用接口信息,或者,所述第四信息包括所述第一信息和所述知识信息的调用接口信息;
向所述目标设备发送第七信息,所述第七信息包括所述知识信息的调用接口信息;
从所述目标设备接收所述第二信息。
在一些实施例中,所述通信装置700为终端,所述通信单元710还用于:
向网络侧设备发送第八信息和第一能力信息中的至少之一,其中,所述通信装置700上部署所述第一模型,所述第八信息用于指示所述通信装置700基于所述第一模型进行推理所需的辅助信息,所述第一能力信息用于指示所述通信装置700基于所述第一模型进行推理的能力信息。
在一些实施例中,所述第八信息包括以下至少之一:
通信装置700基于所述第一模型进行推理所需的知识信息所在的网元指示;
通信装置700基于所述第一模型进行推理所需的知识信息指示;
通信装置700基于所述第一模型进行推理所需的知识信息的调用接口指示;
所述第二信息的格式指示。
在一些实施例中,所述第一能力信息包括以下至少之一:
通信装置700基于所述第一模型进行推理所支持的知识信息指示;
通信装置700基于所述第一模型进行推理所支持的知识信息的调用接口指示;
支持的所述第二信息的格式指示。
在一些实施例中,所述通信单元710还用于:
在模型注册过程中或能力上报过程中,向所述网络侧设备发送所述第八信息和所述第一能力信息中的至少之一。
在一些实施例中,所述通信单元710还用于:
从第二设备接收所述第一信息,所述通信装置700上部署所述第一模型。
在一些实施例中,所述通信装置700还包括:
处理单元,用于利用所述第一模型基于所述第二信息和所述第一模型的输入信息进行推理。
在一些实施例中,所述调用接口信息包括如下至少之一:
调用接口名称、调用接口的输入数据格式、调用接口的输出数据格式、调用接口的输入数据。
可选地,在一些实施例中,上述通信单元可以是通信接口或收发器,或者是通信芯片或者片上系统的输入输出接口。
应理解,根据本申请实施例的通信装置700可对应于本申请方法实施例中的第三设备,并且通信装置700中的各个单元的上述和其它操作和/或功能分别为了实现图2至图14中所示方法实施例中第三设备的相应流程,并达到相同的技术效果,为避免重复,这里不再赘述。
在一些实施例中,本申请实施例中的装置500、装置600、装置700可以是电子设备,例如具有操作系统的电子设备,也可以是电子设备中的部件,例如集成电路或芯片。该电子设备可以是终端,也可以为除终端之外的其他设备。示例性的,终端可以包括但不限于上述所列举的终端11的类型,其他设备可以为服务器、网络附属存储器(Network Attached Storage,NAS)等,本申请实施例不作具体限定。
如图18所示,本申请实施例还提供一种通信设备1000,包括处理器1001和存储器1002,存储器1002上存储有可在所述处理器1001上运行的程序或指令,例如,该通信设备1000为第一设备时,该程序或指令被处理器1001执行时实现上述推理方法实施例中由第一设备执行的步骤,且能达到相同的技术效果。例如,该通信设备1000为第二设备时,该程序或指令被处理器1001执行时实现上述推理方法实施例中由第二设备执行的步骤,且能达到相同的技术效果。例如,该通信设备1000为第三设备时,该程序或指令被处理器1001执行时实现上述推理方法实施例中由第三设备执行的步骤,且能达到相同的技术效果。为避免重复,这里不再赘述。
本申请实施例还提供一种终端,包括处理器和通信接口,所述通信接口和所述处理器耦合,所述处理器用于运行程序或指令,实现如图2至图14所示方法实施例中的步骤。该终端实施例与上述终端侧方法实施例对应,上述方法实施例的各个实施过程和实现方式均可适用于该终端实施例中,且能达到相同的技术效果。具体地,图19为实现本申请实施例的一种终端的硬件结构示意图。
该终端1100包括但不限于:射频单元1101、网络模块1102、音频输出单元1103、输入单元1104、传感器1105、显示单元1106、用户输入单元1107、接口单元1108、存储器1109以及处理器1110等中的至少部分部件。
本领域技术人员可以理解,终端1100还可以包括给各个部件供电的电源(比如电池),电源可以通过电源管理系统与处理器1110逻辑相连,从而通过电源管理系统实现管理充电、放电以及功耗管理等功能。图19中示出的终端结构并不构成对终端的限定,终端可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置,在此不再赘述。
应理解的是,本申请实施例中,输入单元1104可以包括图形处理单元(Graphics Processing Unit,GPU)11041和麦克风11042,图形处理器11041对在视频捕获模式或图像捕获模式中由图像捕获装置(如摄像头)获得的静态图片或视频的图像数据进行处理。显示单元1106可包括显示面板11061,可以采用液晶显示器、有机发光二极管等形式来配置显示面板11061。用户输入单元1107包括触控面板11071以及其他输入设备11072中的至少一种。触控面板11071,也称为触摸屏。触控面板11071可包括触摸检测装置和触摸控制器两个部分。其他输入设备11072可以包括但不限于物理键盘、功能键(比如音量控制按键、开关按键等)、轨迹球、鼠标、操作杆,在此不再赘述。
本申请实施例中,射频单元1101接收来自网络侧设备的下行数据后,可以传输给处理器1110进行处理;另外,射频单元1101可以向网络侧设备发送上行数据。通常,射频单元1101包括但不限于天线、放大器、收发信机、耦合器、低噪声放大器、双工器等。
存储器1109可用于存储软件程序或指令以及各种数据。存储器1109可主要包括存储程序或指令的第一存储区和存储数据的第二存储区,其中,第一存储区可存储操作系统、至少一个功能所需的应用程序或指令(比如声音播放功能、图像播放功能等)等。此外,存储器1109可以包括易失性存储器或非易失性存储器。其中,非易失性存储器可以是只读存储器(Read-Only Memory,ROM)、可编程只读存储器(Programmable ROM,PROM)、可擦除可编程只读存储器(Erasable PROM,EPROM)、电可擦除可编程只读存储器(Electrically EPROM,EEPROM)或闪存。易失性存储器可以是随机存取存储器(Random Access Memory,RAM),静态随机存取存储器(Static RAM,SRAM)、动态随机存取存储器(Dynamic RAM,DRAM)、同步动态随机存取存储器(Synchronous DRAM,SDRAM)、双倍数据速率同步动态随机存取存储器(Double Data Rate SDRAM,DDRSDRAM)、增强型同步动态随机存取存储器(Enhanced SDRAM,ESDRAM)、同步连接动态随机存取存储器(Synch link DRAM,SLDRAM)和直接内存总线随机存取存储器(Direct Rambus RAM,DRRAM)。本申请实施例中的存储器1109包括但不限于这些和任意其它适合类型的存储器。
处理器1110可包括一个或多个处理单元;可选的,处理器1110集成应用处理器和调制解调处理器,其中,应用处理器主要处理涉及操作系统、用户界面和应用程序等的操作,调制解调处理器主要处理无线通信信号,如基带处理器。可以理解的是,上述调制解调处理器也可以不集成到处理器1110中。
可以理解,本实施例中提及的各实现方式的实现过程可以参照图2至图14所示方法实施例的相关描述,并达到相同或相应的技术效果,为避免重复,在此不再赘述。
本申请实施例还提供一种网络侧设备,包括处理器和通信接口,所述通信接口和所述处理器耦合,所述处理器用于运行程序或指令,实现如图2至图14所示所示的方法实施例的步骤。该网络侧设备实施例与上述接入网设备侧或核心网功能侧方法实施例对应,上述方法实施例的各个实施过程和实现方式均可适用于该网络侧设备实施例中,且能达到相同的技术效果。
具体地,本申请实施例还提供了一种网络侧设备。如图20所示,该网络侧设备1200包括:天线1201、射频装置1202、基带装置1203、处理器1204和存储器1205。天线1201与射频装置1202连接。在上行方向上,射频装置1202通过天线1201接收信息,将接收的信息发送给基带装置1203进行处理。在下行方向上,基带装置1203对要发送的信息进行处理,并发送给射频装置1202,射频装置1202对收到的信息进行处理后经过天线1201发送出去。
以上实施例中网络侧设备执行的方法可以在基带装置1203中实现,该基带装置1203包括基带处理器。
基带装置1203例如可以包括至少一个基带板,该基带板上设置有多个芯片,如图20所示,其中一个芯片例如为基带处理器,通过总线接口与存储器1205连接,以调用存储器1205中的程序,执行以上方法实施例中所示的网络设备操作。
该网络侧设备还可以包括网络接口1206,该接口例如为通用公共无线接口(Common Public Radio Interface,CPRI)。
具体地,本申请实施例的网络侧设备1200还包括:存储在存储器1205上并可在处理器1204上运行的指令或程序,处理器1204调用存储器1205中的指令或程序执行图15至图17所示各模块执行的方法,并达到相同的技术效果,为避免重复,故不在此赘述。
具体地,本申请实施例还提供了一种网络侧设备。如图21所示,该网络侧设备1300包括:处理器1301、网络接口1302和存储器1303。其中,网络接口1302例如为通用公共无线接口(common public radio interface,CPRI)。
具体地,本申请实施例的网络侧设备1300还包括:存储在存储器1303上并可在处理器1301上运行的指令或程序,处理器1301调用存储器1303中的指令或程序执行图15至图17所示各模块执行的方法,并达到相同的技术效果,为避免重复,故不在此赘述。
本申请实施例还提供一种可读存储介质,所述可读存储介质上存储有程序或指令,该程序或指令被处理器执行时实现上述通信系统中的模型推理方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
其中,所述处理器为上述实施例中所述的通信装置、通信设备、终端或网络侧设备中的处理器。所述可读存储介质,包括计算机可读存储介质,如计算机只读存储器ROM、随机存取存储器RAM、磁碟或者光盘等。在一些示例中,可读存储介质可以是非瞬态的可读存储介质。
本申请实施例另提供了一种芯片,所述芯片包括处理器和通信接口,所述通信接口和所述处理器耦合,所述处理器用于运行程序或指令,实现上述通信系统中的模型推理方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
应理解,本申请实施例提到的芯片还可以称为系统级芯片,系统芯片,芯片系统或片上系统芯片等。
本申请实施例另提供了一种计算机程序/程序产品,所述计算机程序/程序产品被存储在存储介质中,所述计算机程序/程序产品被至少一个处理器执行以实现上述通信系统中的模型推理方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
本申请实施例还提供了一种通信系统,包括:第一设备、第二设备和第三设备,所述第一设备可用于执行如上所述的通信系统中模型推理方法中由第一设备执行的步骤,第二设备可用于执行如上所述的通信系统中的模型推理方法中由第二设备执行的步骤,第三设备可用于执行如上所述的通信系统中的模型推理方法中由第三设备执行的步骤。
需要说明的是,在本文中,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。此外,需要指出的是,本申请实施方式中的方法和装置的范围不限按示出或讨论的顺序来执行功能,还可包括根据所涉及的功能按基本同时的方式或按相反的顺序来执行功能,例如,可以按不同于所描述的次序来执行所描述的方法,并且还可以添加、省去或组合各种步骤。另外,参照某些示例所描述的特征可在其他示例中被组合。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助计算机软件产品加必需的通用硬件平台的方式来实现,当然也可以通过硬件。该计算机软件产品存储在存储介质(如ROM、RAM、磁碟、光盘等)中,包括若干指令,用以使得终端或者网络侧设备执行本申请各个实施例所述的方法。
上面结合附图对本申请的实施例进行了描述,但是本申请并不局限于上述的具体实施方式,上述的具体实施方式仅仅是示意性的,而不是限制性的,本领域的普通技术人员在本申请的启示下,在不脱离本申请宗旨和权利要求所保护的范围情况下,还可做出很多形式的实施方式,这些实施方式均属于本申请的保护之内。
Claims (25)
- 一种通信系统中的模型推理方法,包括:第一设备接收第二设备发送的第一信息,其中,所述第一设备上存储有知识信息和知识信息的调用接口信息中的至少之一;其中,所述知识信息包括以下至少之一:知识图谱、知识向量库;所述第一设备向第三设备发送第二信息,所述第三设备上部署第一模型,所述第二信息是根据所述第一信息和所述第一设备上存储的或调用的知识信息确定的;其中,所述第一信息包括以下至少之一:用于所述第一模型进行推理的辅助信息;用于所述第一模型进行推理的需求信息;所述第一模型的输入信息;知识信息的调用接口信息;所述第二信息的接收设备指示;所述第一模型的输入信息对应的样本指示。
- 根据权利要求1所述的方法,其中,所述辅助信息包括以下至少之一:人工智能AI模型的标识信息、AI模型的功能指示、AI模型的特性指示、目标知识图谱指示、小区标识、带宽部分BWP标识。
- 根据权利要求1或2所述的方法,其中,所述需求信息包括以下至少之一:AI模型所需适配的目标任务指示;AI模型的业务体验质量QoE要求;AI模型的业务处理时延;AI模型的业务处理准确度;AI模型的业务计算量;AI模型的数据处理规模。
- 根据权利要求1-3中任一项所述的方法,其中,所述第二信息包括以下至少之一:隐藏变量,所述隐藏变量是对所述第一模型的输入信息进行映射后的特征向量;提示信息;网络侧设备的配置信息;场景类型信息;所述第一模型的输入信息对应的样本指示;其中,所述提示信息包括以下至少之一:与所述第一模型适配的目标任务相关的思维链;与所述目标任务相关的提示文字;与所述目标任务相关的提示文件;与所述目标任务相关的提示代码;与所述目标任务相关的思维链标识;与所述目标任务相关的提示文字标识;与所述目标任务相关的提示文件标识;与所述目标任务相关的提示代码标识;与所述目标任务相关的提示文件地址;其中,所述网络侧设备的配置信息包括以下至少之一:网络侧设备的发送天线个数;网络侧设备的发送波束个数;网络侧设备的天线端口个数;网络侧设备的发送功率;网络侧设备的天线增益;网络侧设备的波束3dB带宽;网络侧设备的间距;网络侧设备的频率;系统带宽;网络侧设备覆盖下的终端分布特性;其中,所述场景类型信息包括以下至少之一:视距LOS、非视距NLOS、室内、室外、城区微小区、城区宏小区、郊区微小区、高速、低速。
- 根据权利要求1-4中任一项所述的方法,其中,所述方法还包括:所述第一设备根据所述第一信息和所述第一设备上存储的知识信息确定所述第二信息;或者,所述第一设备从目标设备接收所述第二信息,所述目标设备是使用所述调用接口信息所访问的设备。
- 根据权利要求5所述的方法,其中,所述方法还包括:所述第一设备接收所述目标设备发送的第五信息;所述第一设备根据所述第五信息,登记至少一个调用接口信息;所述第一设备向所述目标设备发送第六信息;其中,所述第五信息包括以下至少之一:调用接口名称,调用接口参数列表,调用接口描述,调用接口的使用样例,调用接口指令集合;所述第六信息包括以下至少之一:是否成功登记调用接口,拒绝登记调用接口,已登记的调用接口列表,失败登记的调用接口列表。
- 根据权利要求1-6中任一项所述的方法,其中,所述方法还包括:所述第一设备向所述第三设备发送所述第一信息。
- 根据权利要求1-7中任一项所述的方法,其中,所述调用接口信息包括如下至少之一:调用接口名称、调用接口的输入数据格式、调用接口的输出数据格式、调用接口的输入数据。
- 一种通信系统中的模型推理方法,包括:第二设备向第一设备发送第一信息,其中,所述第一设备上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,所述知识信息包括以下至少之一:知识图谱、知识向量库;其中,所述第一信息包括以下至少之一:用于第一模型进行推理的辅助信息;用于第一模型进行推理的需求信息;第一模型的输入信息;知识信息的调用接口信息;第二信息的接收设备指示;所述第一模型的输入信息对应的样本指示。
- 根据权利要求9所述的方法,其中,所述第二设备为终端,所述第二设备向第一设备发送第一信息,包括:在发生以下事件中的至少之一时,向所述第一设备发送所述第一信息:接收到第一指示,所述第一指示用于指示激活或切换至所述第一模型;所述第二设备确定激活或切换至所述第一模型;所述第二设备确定所述第一模型不满足需求;接收到第二指示,所述第二指示用于指示所述第一模型不满足需求。
- 根据权利要求9或10所述的方法,其中,所述方法还包括:所述第二设备向第三设备发送所述第一信息,所述第三设备上部署所述第一模型。
- 一种通信系统中的模型推理方法,包括:第三设备从第一设备或目标设备获取第二信息,其中,所述第三设备上部署第一模型,所述第一设备或所述目标设备上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,所述目标设备是使用所述调用接口信息所访问的设备,所述知识信息包括以下至少之一:知识图谱、知识向量库;其中,所述第二信息根据第一信息和所述第一设备上存储或调用的知识信息确定;其中,所述第一信息包括以下至少之一:用于所述第一模型进行推理的辅助信息;用于所述第一模型进行推理的需求信息;所述第一模型的输入信息;知识信息的调用接口信息;所述第二信息的接收设备指示;所述第一模型的输入信息对应的样本指示。
- 根据权利要求12所述的方法,其中,所述第三设备从所述目标设备获取第二信息,包括:所述第三设备接收所述第一设备发送的第四信息,所述第四信息包括所述知识信息的调用接口信息,或者,所述第四信息包括所述第一信息和所述知识信息的调用接口信息;所述第三设备向所述目标设备发送第七信息,所述第七信息包括所述知识信息的调用接口信息;所述第三设备接收所述目标设备发送的所述第二信息。
- 根据权利要求12或13所述的方法,其中,所述第三设备为终端,所述方法还包括:所述第三设备向网络侧设备发送第八信息和第一能力信息中的至少之一,其中,所述第三设备上部署所述第一模型,所述第八信息用于指示所述第三设备基于所述第一模型进行推理所需的辅助信息,所述第一能力信息用于指示所述第三设备基于所述第一模型进行推理的能力信息。
- 根据权利要求14所述的方法,其中,所述第八信息包括以下至少之一:第三设备基于所述第一模型进行推理所需的知识信息所在的网元指示;第三设备基于所述第一模型进行推理所需的知识信息指示;第三设备基于所述第一模型进行推理所需的知识信息的调用接口指示;所述第二信息的预处理方式指示;所述第二信息的格式指示。
- 根据权利要求14或15所述的方法,其中,所述第一能力信息包括以下至少之一:第三设备基于所述第一模型进行推理所支持的知识信息指示;第三设备基于所述第一模型进行推理所支持的知识信息的调用接口指示;支持的所述第二信息的预处理方式指示;支持的所述第二信息的格式指示。
- 根据权利要求14-16中任一项所述的方法,其中,所述第三设备向网络侧设备发送第八信息和第一能力信息中的至少之一,包括:所述第三设备在模型注册过程中或能力上报过程中,向所述网络侧设备发送所述第八信息和所述第一能力信息中的至少之一。
- 根据权利要求12-17中任一项所述的方法,其中,所述方法还包括:所述第三设备接收第二设备发送的所述第一信息,所述第三设备上部署所述第一模型。
- 根据权利要求12-18中任一项所述的方法,其中,所述方法还包括:所述第三设备利用所述第一模型基于所述第二信息和所述第一模型的输入信息进行推理。
- 根据权利要求12-19中任一项所述的方法,其中,所述调用接口信息包括如下至少之一:调用接口名称、调用接口的输入数据格式、调用接口的输出数据格式、调用接口的输入数据。
- 一种通信装置,包括:通信单元,用于接收第二设备发送的第一信息,其中,所述通信装置上存储有知识信息和知识信息的调用接口信息中的至少之一;其中,所述知识信息包括以下至少之一:知识图谱、知识向量库;以及向第三设备发送第二信息,所述第三设备上部署第一模型,所述第二信息是根据所述第一信息和所述通信装置上存储的或调用的知识信息确定的;其中,所述第一信息包括以下至少之一:用于所述第一模型进行推理的辅助信息;用于所述第一模型进行推理的需求信息;所述第一模型的输入信息;知识信息的调用接口信息;所述第二信息的接收设备指示;所述第一模型的输入信息对应的样本指示。
- 一种通信装置,包括:通信单元,用于向第一设备发送第一信息,其中,所述第一设备上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,所述知识信息包括以下至少之一:知识图谱、知识向量库;其中,所述第一信息包括以下至少之一:用于第一模型进行推理的辅助信息;用于第一模型进行推理的需求信息;第一模型的输入信息;知识信息的调用接口信息;第二信息的接收设备指示;所述第一模型的输入信息对应的样本指示。
- 一种通信装置,包括:通信单元,用于从第一设备或目标设备获取第二信息,其中,所述通信装置700上部署第一模型,所述第一设备或所述目标设备上存储有知识信息和知识信息的调用接口信息中的至少之一,其中,所述目标设备是使用所述调用接口信息所访问的设备,所述知识信息包括以下至少之一:知识图谱、知识向量库;其中,所述第二信息根据第一信息和所述第一设备上存储或调用的知识信息确定;其中,所述第一信息包括以下至少之一:用于所述第一模型进行推理的辅助信息;用于所述第一模型进行推理的需求信息;所述第一模型的输入信息;知识信息的调用接口信息;所述第二信息的接收设备指示;所述第一模型的输入信息对应的样本指示。
- 一种通信设备,包括:存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序,所述计算机程序被所述处理器执行时实现如权利要求1至8中任一项所述的方法中的步骤,或如权利要求9至11中任一项所述的方法中的步骤,或者,如权利要求12至20中任一项所述的方法中的步骤。
- 一种可读存储介质,所述可读存储介质上存储程序或指令,所述程序或指令被处理器执行时实现如权利要求1至8中任一项所述的方法中的步骤,或如权利要求9至11中任一项所述的方法中的步骤,或者,如权利要求12至20中任一项所述的方法中的步骤。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311551035.X | 2023-11-20 | ||
| CN202311551035.XA CN120020824A (zh) | 2023-11-20 | 2023-11-20 | 通信系统中的模型推理方法、装置、设备以及介质 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2025108271A1 true WO2025108271A1 (zh) | 2025-05-30 |
Family
ID=95704642
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2024/133006 Pending WO2025108271A1 (zh) | 2023-11-20 | 2024-11-19 | 通信系统中的模型推理方法、装置、设备以及介质 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN120020824A (zh) |
| WO (1) | WO2025108271A1 (zh) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4999833A (en) * | 1985-05-06 | 1991-03-12 | Itt Corporation | Network connectivity control by artificial intelligence |
| CN111865446A (zh) * | 2020-07-29 | 2020-10-30 | 中南大学 | 利用网络环境上下文信息实现的智能波束配准方法与装置 |
| CN112287114A (zh) * | 2020-09-28 | 2021-01-29 | 珠海大横琴科技发展有限公司 | 一种知识图谱服务处理方法和装置 |
| CN112805743A (zh) * | 2018-10-16 | 2021-05-14 | 三星电子株式会社 | 用于基于知识图谱来提供内容的系统和方法 |
| CN114205852A (zh) * | 2022-02-17 | 2022-03-18 | 网络通信与安全紫金山实验室 | 无线通信网络知识图谱的智能分析与应用架构及方法 |
| WO2023186099A1 (zh) * | 2022-04-02 | 2023-10-05 | 维沃移动通信有限公司 | 信息反馈方法、装置及设备 |
-
2023
- 2023-11-20 CN CN202311551035.XA patent/CN120020824A/zh active Pending
-
2024
- 2024-11-19 WO PCT/CN2024/133006 patent/WO2025108271A1/zh active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4999833A (en) * | 1985-05-06 | 1991-03-12 | Itt Corporation | Network connectivity control by artificial intelligence |
| CN112805743A (zh) * | 2018-10-16 | 2021-05-14 | 三星电子株式会社 | 用于基于知识图谱来提供内容的系统和方法 |
| CN111865446A (zh) * | 2020-07-29 | 2020-10-30 | 中南大学 | 利用网络环境上下文信息实现的智能波束配准方法与装置 |
| CN112287114A (zh) * | 2020-09-28 | 2021-01-29 | 珠海大横琴科技发展有限公司 | 一种知识图谱服务处理方法和装置 |
| CN114205852A (zh) * | 2022-02-17 | 2022-03-18 | 网络通信与安全紫金山实验室 | 无线通信网络知识图谱的智能分析与应用架构及方法 |
| WO2023186099A1 (zh) * | 2022-04-02 | 2023-10-05 | 维沃移动通信有限公司 | 信息反馈方法、装置及设备 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN120020824A (zh) | 2025-05-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20250055763A1 (en) | Model information obtaining method and apparatus, model information sending method and apparatus, node, and storage medium | |
| CN117440449A (zh) | Ai模型信息传输方法、装置及设备 | |
| WO2023125594A1 (zh) | Ai模型传输方法、装置、设备及存储介质 | |
| WO2024088119A1 (zh) | 数据处理方法、装置、终端及网络侧设备 | |
| WO2024208260A1 (zh) | 基于ai的csi预测的数据采集方法及装置 | |
| CN118158812A (zh) | 信息发送、信息接收方法、装置、终端及网络侧设备 | |
| CN116867020A (zh) | 终端路由选择策略ursp确定方法、终端及网络侧设备 | |
| WO2025108271A1 (zh) | 通信系统中的模型推理方法、装置、设备以及介质 | |
| WO2024078400A1 (zh) | 模型请求方法、装置、通信设备及可读存储介质 | |
| CN116866886A (zh) | 数据获取方法及通信设备 | |
| CN117858072A (zh) | 信息传输方法、装置及设备 | |
| CN116847330A (zh) | 终端不可用时期的协商方法、终端及网络侧设备 | |
| WO2025108272A1 (zh) | 基于ai模型的推理方法、通信装置、通信设备及介质 | |
| CN116418467B (zh) | 探测参考信号的配置方法、装置、终端及网络侧设备 | |
| WO2025209284A1 (zh) | 通信数据处理方法、装置及通信设备 | |
| WO2024078589A1 (zh) | 信息上报方法、装置、通信设备及存储介质 | |
| WO2025026299A1 (zh) | Ai业务建立方法、装置及网络侧设备 | |
| CN119011044A (zh) | 模型处理方法、装置、终端及网络侧设备 | |
| WO2025026315A1 (zh) | 策略协商方法、策略生成方法及设备 | |
| WO2025140070A1 (zh) | 网元查找方法、网元注册方法、装置及通信设备 | |
| CN120128881A (zh) | 信息传输方法、装置及设备 | |
| WO2023213242A1 (zh) | Ue策略信息的配置方法、装置、第一网络功能及终端 | |
| WO2025036206A1 (zh) | 模型的识别方法、装置、终端及网络侧设备 | |
| WO2025026194A1 (zh) | 业务处理方法、装置、通信设备及可读存储介质 | |
| WO2024208308A1 (zh) | 人工智能ai单元的标识方法及通信设备 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 24893428 Country of ref document: EP Kind code of ref document: A1 |