CN116418797A

CN116418797A - AI network information transmission method and device and communication equipment

Info

Publication number: CN116418797A
Application number: CN202111666991.3A
Authority: CN
Inventors: 任千尧; 孙鹏
Original assignee: Vivo Mobile Communication Co Ltd
Current assignee: Vivo Mobile Communication Co Ltd
Priority date: 2021-12-31
Filing date: 2021-12-31
Publication date: 2023-07-11
Also published as: WO2023125869A1

Abstract

The application discloses an AI network information transmission method, an AI network information transmission device and communication equipment, which belong to the technical field of communication, and the AI network information transmission method in the embodiment of the application comprises the following steps: the first end converts the AI network information of the target AI network into an open neural network exchange ONNX file structure; the first end sends the ONNX file structure to a second end.

Description

AI network information transmission method and device and communication equipment

Technical Field

The application belongs to the technical field of communication, and particularly relates to an AI network information transmission method, an AI network information transmission device and communication equipment.

Background

Artificial intelligence (Artificial Intelligence, AI) is a new technical science for researching and developing theories, methods, techniques and application systems for simulating, extending and expanding human intelligence, and is receiving a great deal of attention from people, and the application of AI is becoming more and more widespread. At present, research into the use of AI networks in communication systems, for example, communication data may be transmitted between a network-side device and a terminal through the AI network, has been begun. In the communication system, the neural network frameworks applied by different communication devices are different, and the file structures stored by the AI network information under the different neural network frameworks are different, so that the communication devices may not be able to read the AI network information transmitted by other communication devices different from the neural network frameworks, and the transmission of the AI network information between the communication devices is limited.

Disclosure of Invention

The embodiment of the application provides an AI network information transmission method, an AI network information transmission device and communication equipment, which can solve the problem that AI network information transmission between communication equipment is limited in the related technology.

In a first aspect, an AI network information transmission method is provided, including:

the first end converts the AI network information of the target AI network into an open neural network exchange ONNX file structure;

the first end sends the ONNX file structure to a second end.

In a second aspect, there is provided an AI network information transfer method, including:

and the second end receives an ONNX file structure sent by the first end, wherein the ONNX file structure is obtained by converting the AI network information of the target AI network by the first end.

In a third aspect, there is provided an AI network information transfer apparatus including:

the switching module is used for switching the AI network information of the target AI network into an open neural network exchange ONNX file structure;

and the sending module is used for sending the ONNX file structure to the second end.

In a fourth aspect, there is provided an AI network information transfer apparatus including:

and the receiving module is used for receiving an ONNX file structure sent by the first end, wherein the ONNX file structure is obtained by converting the AI network information of the target AI network by the first end.

In a fifth aspect, there is provided a communication device comprising a processor and a memory storing a program or instructions executable on the processor, which when executed by the processor, implement the steps of the AI-network-information-transmission method of the first aspect, or implement the steps of the AI-network-information-transmission method of the second aspect.

In a sixth aspect, there is provided a readable storage medium storing thereon a program or instructions which, when executed by a processor, implement the steps of the AI-network-information-transmission method as set forth in the first aspect, or implement the steps of the AI-network-information-transmission method as set forth in the second aspect.

In a seventh aspect, there is provided a chip, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being configured to execute a program or instructions to implement the steps of the AI-network information transfer method as described in the first aspect, or to implement the steps of the AI-network information transfer method as described in the second aspect.

In an eighth aspect, there is provided a computer program/program product stored in a storage medium, the computer program/program product being executed by at least one processor to implement the steps of the AI-network-information-transmission method as set forth in the first aspect, or to implement the steps of the AI-network-information-transmission method as set forth in the second aspect.

In the embodiment of the application, the first end converts the AI network information of the target AI network based on the self-neural network frame into an ONNX file structure, and then sends the ONNX file structure to the second end, so that the second end can convert the ONNX file structure into the AI network under the self-neural network frame. Therefore, even two communication devices with different neural network frameworks can transmit the AI network information based on the ONNX file structure, and the transmission of the AI network information between the communication devices is prevented from being blocked.

Drawings

Fig. 1 is a block diagram of a wireless communication system to which embodiments of the present application are applicable;

fig. 2 is a flowchart of an AI network information transmission method provided in an embodiment of the present application;

fig. 3 is a flowchart of another AI-network information transmission method provided by an embodiment of the application;

fig. 4 is a block diagram of an AI network information transmission apparatus provided in an embodiment of the application;

fig. 5 is a block diagram of another AI-network information transfer apparatus provided by an embodiment of the application;

fig. 6 is a block diagram of a communication device according to an embodiment of the present application;

fig. 7 is a block diagram of a terminal according to an embodiment of the present application;

fig. 8 is a block diagram of a network side device according to an embodiment of the present application;

Fig. 9 is a block diagram of another network side device according to an embodiment of the present application.

Detailed Description

Technical solutions in the embodiments of the present application will be clearly described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application are within the scope of the protection of the present application.

The terms first, second and the like in the description and in the claims, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances such that the embodiments of the application are capable of operation in sequences other than those illustrated or otherwise described herein, and that the terms "first" and "second" are generally intended to be used in a generic sense and not to limit the number of objects, for example, the first object may be one or more. Furthermore, in the description and claims, "and/or" means at least one of the connected objects, and the character "/" generally means a relationship in which the associated object is an "or" before and after.

It is noted that the techniques described in embodiments of the present application are not limited to long term evolution (Long Term Evolution, LTE)/LTE evolution (LTE-Advanced, LTE-a) systems, but may also be used in other wireless communication systems, such as code division multiple access (Code Division Multiple Access, CDMA), time division multiple access (Time Division Multiple Access, TDMA), frequency division multiple access (Frequency Division Multiple Access, FDMA), orthogonal frequency division multiple access (Orthogonal Frequency Division Multiple Access, OFDMA), single carrier frequency division multiple access (Single-carrier Frequency Division Multiple Access, SC-FDMA), and other systems. The terms "system" and "network" in the embodiments of the present application are often used interchangeably, and the techniques described may be used for both the above-mentioned systems and radio technologies,but may also be used in other systems and radio technologies. The following description describes a New air interface (NR) system for purposes of example and uses NR terminology in much of the description that follows, but these techniques are also applicable to applications other than NR system applications, such as generation 6 (6) ^th Generation, 6G) communication system.

Fig. 1 shows a block diagram of a wireless communication system to which embodiments of the present application are applicable. The wireless communication system includes a terminal 11 and a network device 12. The terminal 11 may be a mobile phone, a tablet (Tablet Personal Computer), a Laptop (Laptop Computer) or a terminal-side Device called a notebook, a personal digital assistant (Personal Digital Assistant, PDA), a palm top, a netbook, an ultra-mobile personal Computer (ultra-mobile personal Computer, UMPC), a mobile internet appliance (Mobile Internet Device, MID), an augmented reality (augmented reality, AR)/Virtual Reality (VR) Device, a robot, a Wearable Device (weather Device), a vehicle-mounted Device (VUE), a pedestrian terminal (PUE), a smart home (home Device with a wireless communication function, such as a refrigerator, a television, a washing machine, or a furniture), a game machine, a personal Computer (personal Computer, PC), a teller machine, or a self-service machine, and the Wearable Device includes: intelligent wrist-watch, intelligent bracelet, intelligent earphone, intelligent glasses, intelligent ornament (intelligent bracelet, intelligent ring, intelligent necklace, intelligent anklet, intelligent foot chain etc.), intelligent wrist strap, intelligent clothing etc.. Note that, the specific type of the terminal 11 is not limited in the embodiment of the present application. The network-side device 12 may comprise an access network device or core network device, wherein the access network device may also be referred to as a radio access network device, a radio access network (Radio Access Network, RAN), a radio access network function or a radio access network element. The access network device may include a base station, a WLAN access point, a WiFi node, or the like, where the base station may be referred to as a node B, an evolved node B (eNB), an access point, a base transceiver station (Base Transceiver Station, BTS), a radio base station, a radio transceiver, a basic service set (Basic Service Set, BSS), an extended service set (Extended Service Set, ESS), a home node B, a home evolved node B, a transmission receiving point (Transmitting Receiving Point, TRP), or some other suitable terminology in the field, and the base station is not limited to a specific technical vocabulary so long as the same technical effect is achieved, and it should be noted that in the embodiment of the present application, only the base station in the NR system is described by way of example, and the specific type of the base station is not limited. The core network device may include, but is not limited to, at least one of: core network nodes, core network functions, mobility management entities (Mobility Management Entity, MME), access mobility management functions (Access and Mobility Management Function, AMF), session management functions (Session Management Function, SMF), user plane functions (User Plane Function, UPF), policy control functions (Policy Control Function, PCF), policy and charging rules function units (Policy and Charging Rules Function, PCRF), edge application service discovery functions (Edge Application Server Discovery Function, EASDF), unified data management (Unified Data Management, UDM), unified data repository (Unified Data Repository, UDR), home subscriber server (Home Subscriber Server, HSS), centralized network configuration (Centralized network configuration, CNC), network storage functions (Network Repository Function, NRF), network opening functions (Network Exposure Function, NEF), local NEF (or L-NEF), binding support functions (Binding Support Function, BSF), application functions (Application Function, AF), and the like. In the embodiment of the present application, only the core network device in the NR system is described as an example, and the specific type of the core network device is not limited.

For a better understanding, the following explanation is made on related concepts that may be involved in the embodiments of the present application.

Neural network framework

Neural networks have a wide variety of implementation frameworks including TensorFlow, pyTorch, keras, MXNet, caffe2, etc., each with varying emphasis. For example, the deep learning framework with Caffe2 and Keras being high-level can quickly verify the model, and the deep learning framework with TensorFlow and pyrerch being bottom-level can implement modification of the details of the neural network bottom-level. For another example, the emphasis of PyTorch is on supporting a dynamic graph model, the emphasis of TensorFlow is on supporting a variety of hardware, the running speed is fast, and Caffe2 is on a lightweight scale, etc. Each implementation framework uses a method to describe the neural network, and the operations of network construction, training, inference and the like are completed.

In general, the file structures of the network information storage under two different neural network frameworks are different and cannot be directly read, and interaction is required through other standard structures.

Open neural network interactions (Open Neural Network Exchange, ONNX)

ONNX is an AI interaction network, and ONNX itself is only a data structure for describing an AI network, and does not include implementation schemes. The ONNX stores the whole AI network in a computational graph protocol class (GraphProto), including network structure and network parameters, and completes the basic description of the AI network through a node protocol class (NodeProto), a parameter information protocol class (ValueInfoProto) and a tensor protocol class (TensorProto). The node proto is used for describing a network structure, the ONNX describes each operator of the AI network as a node, the input and output names of each node are globally unique, and the network structure of the whole AI network is described through the matching relation of the input and output names. All network parameters are considered as inputs or outputs, also retrieved by name, so ONNX can express the structure of the whole network through a series of nodeboto entities. ValueInfoProto is used to describe all input and output information, including dimensions and element types, indicating the size of the corresponding element, the name of each input and output and the correspondence recorded in the nodoproto. The TensorProto is used for storing the numerical value of a specific network parameter, and obtaining the corresponding parameter according to the name input and output by each node and the storage position.

In addition, ONNX assigns functions of the nodes, such as a convolution layer, a multiplication layer, and the like, to the corresponding nodes through attribute protocol class (AttributeProto), and values of super parameters required by the functions are all stored in TensorProto, and dimensions of the super parameters are all stored in ValueInfoProto.

In some embodiments, the Protocol class (Proto) may also be referred to as a structure, such as a node structure (nodoproto), a parameter information structure (ValueInfoProto), a tensor structure (TensorProto), an attribute structure (AttributeProto), and so on. Proto may be considered a type of data, file, etc. formed based on a particular protocol.

The AI network information transmission method provided by the embodiments of the present application is described in detail below with reference to the accompanying drawings through some embodiments and application scenarios thereof.

Referring to fig. 2, fig. 2 is a flowchart of an AI network information transmission method according to an embodiment of the disclosure, and as shown in fig. 2, the method includes the following steps:

step 201, a first end converts AI network information of a target AI network into an ONNX file structure;

step 202, the first end sends the ONNX file structure to a second end.

In this embodiment, the first end converts AI network information of the target AI network into an ONNX file structure, for example, if the AI network information includes a complete network structure and all network parameters of the target AI network, the first end expresses the network structure and the network parameters into the ONNX file structure based on the structural form of ONNX, for example, describes the network structure through a node protocol class (nodoproto) and a parameter information protocol class (ValueInfoProto), describes the network parameters through a tensor protocol class (tensorpro), and so on.

Further, the first end sends the ONNX file structure to the second end, and the second end converts the ONNX file structure into an AI network under the self neural network framework so as to realize training and application of the second end to the AI network.

In this embodiment of the present application, the first end converts AI network information of a target AI network applicable to a self-neural network framework into an ONNX file structure, and then sends the ONNX file structure to the second end, so that the second end can convert the ONNX file structure into an AI network under the self-neural network framework. Therefore, two communication devices with different neural network frameworks can transmit the AI network information based on the ONNX file structure, and the transmission of the AI network information between the communication devices is prevented from being blocked.

The AI network information includes at least one of a network structure and a network parameter of the target AI network. For example, the AI network information includes a complete network structure of the target AI network and all network parameters, or the AI network information includes a complete network structure of the target AI network, or includes only updated AI network structures in the target AI network, or includes only partial network parameters of the target AI network, or includes only partial values of network parameters of the target AI network, and so on. Therefore, the network structure and the network parameters of the AI network can be sent separately, and all AI networks including the whole network structure and the network parameters are not required to be transmitted together in the communication process, so that the transmission cost in the communication process can be effectively reduced.

In this embodiment of the present application, the ONNX file structure includes a target protocol class, where the target protocol class includes at least one of the following: node protocol class (NodeProto), parameter information protocol class (ValueInfoProto), tensor protocol class (TensorProto), attribute protocol class (AttributeProto).

Wherein, nodeProto is used to describe the network structure; valueInfoProto is used for describing information of all input, output and network parameters, including dimension and element types, and indicates the size of corresponding elements, and the name or index of each input, output and network parameter corresponds to the record in NodeProto; the TensorProto is used for storing the numerical value of a specific network parameter, and obtaining a corresponding parameter by going to a storage position according to the name input and output by each node; attributeProto is used for recording the functions of nodes, such as a convolution layer, a multiplication layer and the like, and endowing corresponding node functions, wherein the values of super parameters required by the functions are all stored in TensorProto, and the dimensions of the super parameters are all stored in ValueInfoProto.

The object protocol class includes at least one of the above-mentioned node protocol class, parameter information protocol class, tensor protocol class and attribute protocol class, and the ONNX file structure also includes at least one of the above-mentioned node protocol class, parameter information protocol class, tensor protocol class and attribute protocol class. For example, if the target protocol class includes a node protocol class, the node protocol class may form a complete onnx.proto file to generate an ONNX file structure; if the target protocol class comprises a node protocol class, a parameter information protocol class and a tensor protocol class, generating an ONNX file structure based on the node protocol class, the parameter information protocol class and the tensor protocol class to form an ONNX. Of course, the protocol class included in the ONNX file structure may be other possible situations, which are not listed here too much.

Optionally, the ONNX file structure includes at least one computational graph protocol class (GraphProto), the computational graph protocol class including the target protocol class. That is, in the case where the target protocol class includes at least one of the above-described node protocol class, parameter information protocol class, tensor protocol class, and attribute protocol class, the content included in the target protocol class may be written in the calculation map protocol class.

For example, the target protocol class includes a node protocol class, a parameter information protocol class and a tensor protocol class, and then the node protocol class, the parameter information protocol class and the tensor protocol class are written into a computation graph protocol class, and an ONNX file structure is generated based on the computation graph protocol class, and the first end sends the ONNX file structure to the second end. Thus, the node protocol class, the parameter information protocol class and the tensor protocol class are combined together for transmission.

Alternatively, the ONNX file structure may include a plurality of image protocol classes, and the target protocol class included in each computation graph protocol class may be different. For example, the ONNX file structure includes two computation graph protocol classes, where one computation graph protocol class includes a node protocol class and a parameter information protocol class, and the other computation graph protocol class includes a tensor protocol class. In addition, each computation graph protocol class can also comprise a plurality of node protocol classes, parameter information protocol classes, tensor protocol classes and attribute protocol classes.

It should be noted that, the ONNX file structure may further include a model protocol class (model proto), where the model protocol class includes a computation graph protocol class, and the computation graph protocol class includes a plurality of node protocol classes, parameter information protocol classes, and tensor protocol classes, and the node protocol class may include a plurality of attribute protocol classes. In this embodiment of the present application, the ONNX file structure may include a plurality of calculation map protocol classes, whether or not there is a model protocol class; that is, the ONNX file structure in the embodiment of the present application may not include a model protocol class, that is, a transmission model protocol class is not required, so that transmission overhead of the communication device can be saved, and application of the ONNX file structure in air interface transmission is more facilitated.

Optionally, the target protocol class may include any one of a node protocol class (nodoproto), a parameter information protocol class (ValueInfoProto), a tensor protocol class (tensorpro to), and an attribute protocol class (attributepro to), where the number of ONNX file structures is at least two, one of the ONNX file structures corresponds to one target protocol class. In this case, the first end sends the ONNX file structure to the second end, including any one of the following:

The first end merges and sends the at least two ONNX file structures to a second end;

and the first end sends the at least two ONNX file structures to the second end respectively.

In this embodiment, each proto may be independently formed into an ONNX file structure, and thus one ONNX file structure corresponds to one type of proto. For example, if the AI network information of the target AI network includes a network structure and a weight parameter, the AI network information generates a node protocol class, a parameter information protocol class, and a tensor protocol class based on the ONNX structure, the node protocol class correspondingly generates an ONNX file structure, the parameter information protocol class correspondingly generates an ONNX file structure, and the tensor protocol class correspondingly generates an ONNX file structure, so that three ONNX file structures are also obtained. Therefore, the ONNX file structure is more flexibly generated, the second end is favorable for distinguishing the ONNX file structure, and the ONNX file structure can be transmitted in different time slots and time-frequency positions, so that time-frequency resources are more effectively utilized, and the scheduling is convenient.

Alternatively, the first end may send the three ONNX file structures to the second end, for example, send the three ONNX file structures sequentially; alternatively, the first end may combine the three ONNX file structures and send the combined three ONNX file structures to the second end at a time. Furthermore, for the received ONNX file structure, the second end can determine that one ONNX file structure corresponds to one proto type, so that the second end can convert the ONNX file structure into AI network information under the self neural network framework more conveniently.

In this embodiment, in a case where the ONNX file structure includes at least one of a node protocol class and a parameter information protocol class, after the first end sends the ONNX file structure to the second end, the method further includes:

and the first end sends the value of the network parameter of the target AI network to the second end.

Specifically, after the first end transmits the ONNX file structure including the node protocol class and the parameter information protocol class to the second end, the first end may further send the value of the network parameter of the target AI network to the second end. It will be appreciated that, since the first end has sent the ONNX file structure including the node protocol class and the parameter information protocol class to the second end, and thus the number and cost of the network parameters by the second end are known, the first end may directly send the specific values of the network parameters in the order of the parameter information protocol class, without converting to the ONNX file structure, and the second end sequentially receives the specific values of each parameter. The node protocol class and the parameter information protocol class may be contained in one ONNX file structure, or may be respectively corresponding to one ONNX file structure.

Optionally, the first end sends the value of the network parameter of the target AI network to the second end, including:

And the first end sends the network parameter values of the target AI network to the second end through a data channel.

In this embodiment of the present application, the ONNX file structure includes a first digital index and a second digital index of integer types, where the first digital index is used to identify the target protocol class, one of the target protocol classes corresponds to one of the first digital indexes, and the second digital index is used to identify at least one of network parameters, inputs, and outputs of the target AI network.

It will be appreciated that the protocol classes included in the ONNX file structure may be identified by a first numerical index. In this embodiment of the present application, the target protocol class includes at least one of a node protocol class, a parameter information protocol class, a tensor protocol class, and an attribute protocol class, and one protocol class may correspond to one first digital index, for example, the ONNX file structure includes a total of 5 protocol classes, that is, the 5 protocol classes may be represented by 5 first digital indexes (for example, 1, 2, 3, 4, and 5).

Alternatively, a protocol class type may correspond to a digital index, for example, a total of 5 protocol classes, and 4 protocol class types, and the 5 protocol classes may be characterized by the 4 first digital indexes.

Alternatively, each protocol class type may correspond to a digital index sequence, for example, a total of 3 node protocol classes, 4 parameter information protocol classes and 1 tensor protocol class are included, the digital indexes 0, 1, 2 represent 3 node protocol classes, 0, 1, 2, 3 represent 4 parameter information protocol classes, 0 represents 1 tensor protocol class, and the sequences of the three protocol classes are distinguished by a specific identifier.

In the embodiment of the application, the content described for the target protocol class can be identified by a second digital index. For example, the node protocol class is used to describe the network structure of the target AI network, the names of the inputs and outputs of each node of the network structure are unique, one node protocol class may be used to describe the inputs and related network parameters of one node in the network structure, and the inputs, outputs and network parameters corresponding to the node protocol class may be identified by the second digital index, for example, the inputs, outputs and network parameters respectively correspond to three different second digital indexes. Likewise, the content of the other protocol class description may be identified by a second digital index.

Thus, for the ONNX file structure, the protocol class included in the ONNX file structure is identified by the first digital index, and the content described by the protocol class is identified by the second digital index. The first digital index and the second digital index are both integer (int) type digital indexes, and compared with the identification through the character string, the identification through the integer type digital indexes can save transmission overhead more effectively.

Optionally, in case the ONNX file structure comprises a node protocol class, the node protocol class comprises the second digital index for characterizing the input and output of the target AI network. It should be noted that other protocol classes, such as parameter information protocol class, tensor protocol class, and attribute protocol class, may also be the content described by the protocol class, such as network parameters, inputs, outputs, etc. of the target AI network, which are identified by the second digital index.

Optionally, in a case where the number of inputs and outputs of the target AI network is plural and continuous, the node protocol class includes a target digital index, and at least one of the number of inputs and the number of outputs of the target AI network, wherein the target digital index is a digital index of the second digital index for identifying a first input and a first output of the plurality of inputs and outputs of the target AI network.

It should be noted that, in the ONNX file structure, the network structure of the target AI network is described by a node protocol class, where the network structure includes a plurality of nodes, one node corresponds to at least one input and at least one input, in a case where the number of nodes included in the network structure is a plurality of and consecutive nodes, the input and output of one node corresponds to one second digital index, and thus should correspond to a plurality of consecutive second digital indexes, in this case, the node protocol class in the ONNX file structure may include only the second digital index (i.e. the target digital index) for identifying the input and output of the first node in the plurality of nodes and the total number of nodes.

For example, the network structure of the target AI network includes 3 nodes, and the inputs and outputs of the same node correspond to one second digital index, and thus to 3 second digital indexes, e.g., 11, 12, 13; in this case, the node protocol class includes the second digital index for identifying the input and output correspondence of the first node and the total number of nodes (i.e., the number of inputs and outputs), and the node protocol class of the ONNX file structure also includes 11 and 3, so that it is not necessary to include all the second digital indexes. Under the condition that the input and output quantity of the target AI network is more, the second digital index for identifying each input and output is not needed, and the node protocol class only needs to comprise the target digital index and the total input and output quantity, so that the content of the ONNX file structure can be saved, and the transmission cost of the ONNX file structure is effectively saved. And the second end can obtain a second digital index corresponding to each input and output based on the target digital index and the total input and output number after receiving the ONNX file structure.

In this embodiment of the present application, one parameter information protocol class is used to describe at least one of an input, an output and a network parameter corresponding to a network node in the target AI network, where the parameter information protocol class includes a first element type and dimension information. Specifically, the parameter information protocol class is used for describing information of all inputs and outputs of the target AI network, and includes dimension information and a first element type, which indicates the size of a corresponding element, and the name of each input and output corresponds to the input and output recorded in the node protocol class.

Optionally, the first element type is characterized by an integer length of the quantization bit number. That is, the first element type is replaced by the number of quantization bits, so that the first element type may be omitted, and a default type, e.g. integer (int) length, is used, i.e. the number of quantization bits is one integer length, and the first element type may be characterized by specifying the integer length.

Optionally, the first element type corresponds to a network node or a network parameter of the target AI network. The integer with the length of 1bit may be offset by a predetermined correspondence, for example, a certain kind of network node or network parameter uses a corresponding integer length, for example, the integer with the length of 3 bits (bits) is used by the weight of the convolution node.

It should be noted that, at present, the element types in the ONNX file structure are defined as follows:

enum DataType{UNDEFINED＝0；FLOAT＝1；UINT8＝2；...INT32＝6；INT64＝7；STRING＝8；BOOL＝9；...COMPLEX128＝15；...}

in this embodiment of the present application, the element types may be modified to be int 1=0, int2=1, int3=2, and int4=3, where int1 represents an index of 1bit, that is, only values of 0 and 1, int2 corresponds to an index of 00 01 10 11 or 0123, and a quantization table corresponding to a specific index is agreed by a protocol according to a function of the AI network, or the base station is configured. Alternatively, the first element type and subsequent second element types may be characterized by such integer lengths.

In this embodiment, one network node of the target AI network corresponds to a parameter information protocol class, and one parameter information protocol class characterizes at least one of input, output and network parameters of the corresponding network node by dimension information, where the dimension information includes a dimension value and a dimension size value, for example, for a 3-dimensional matrix (2×6×14), the dimension value is 3, that is, the dimension is 3, and the dimension size value is 2×6×14=168, and the dimension information includes (3,2,6,14), or (3, (2,6,14)).

Optionally, the dimension value is the number of values included in the corresponding dimension, and the values are normalized values. For example, for a 3-dimensional matrix (2×6×14), the dimension size value is 168, i.e., the dimension includes 168 values, and these values are normalized values.

Optionally, the dimension information in the parameter information protocol class includes a corresponding dimension number value and a dimension size value, the dimension number value in the parameter information protocol class is arranged based on a preset sequence, the dimension size value in the parameter information protocol class is arranged based on the preset sequence, and the preset sequence is a preset arrangement sequence of network parameters of the target AI network.

In this embodiment, one network parameter of the target AI network corresponds to a parameter information protocol class, where the parameter information protocol class includes dimension information, and the dimension information includes dimension value and dimension size value, and the network parameter corresponding to the parameter information protocol class is described through the dimension value and the dimension size value, so that each network parameter corresponds to one dimension value and one dimension size value. In the case that the target AI network includes a plurality of network parameters, the ONNX file structure includes a plurality of corresponding parameter information protocol classes, and each parameter information protocol class includes a dimension value and a dimension size value, the ONNX file structure includes a plurality of dimension values and a plurality of dimension size values accordingly; in this case, the dimension values and dimension values may be arranged according to a preset arrangement sequence of network parameters in the target AI network, for example, all dimension values may be arranged and placed according to a preset arrangement sequence of network parameters, and then all dimension values may be arranged and placed according to a preset arrangement sequence of network parameters. Therefore, the dimension number adopts bits with the same length, the second end knows the bit length corresponding to the dimension size value after analyzing the dimension number value, and further, the identification of the network parameter can be omitted, so that the transmission cost is saved.

Optionally, when the target parameter information protocol class is used for characterizing a target network parameter, the target parameter information protocol class includes a numerical position of a maximum numerical value, the maximum numerical value is a maximum value of the numerical values of the target network parameter, the target tensor protocol class corresponding to the target network parameter includes a quantized value of a ratio of each remaining numerical value in the target network parameter to the maximum numerical value, and the remaining numerical values are other numerical values except the maximum numerical value in a dimension corresponding to the target parameter information protocol class.

The target network parameter is any network parameter of a target AI network, the target parameter information protocol class is a parameter information protocol class used for representing the target network parameter in an ONNX file structure, and the target tensor protocol class is a tensor protocol class corresponding to the target network parameter represented by the target parameter information protocol class. For example, if the target network parameter includes 168 values, the target parameter information protocol class corresponding to the target network parameter includes the numerical position of the largest value of the 168 values, and the remaining 167 values are characterized by the target tensor protocol class corresponding to the target network parameter, where the quantized value of the ratio of the 167 values to the largest value is included in the target tensor protocol class.

In this embodiment of the present application, the parameter information protocol class further includes an indication parameter for indicating a location of a non-zero value in the network parameter. Accordingly, the quantized value of 0 or the value smaller than the quantized minimum threshold value in the tensor protocol class corresponding to the network parameter may not be included.

It should be noted that, the definition of the parameter information protocol class in the present ONNX file structure is:

{optional string name＝1；

optional Typeproto type＝2；

optional string doc_string＝3；}

in this embodiment of the present application, typeproco may be modified into int, which is used to record element types, that is, quantization bits, and may be the same as the previous parameter information protocol class when there is no record; or deleting the type, and saving the typeproco in the calculation map protocol class, wherein the calculation map protocol class indicates that all parameter information protocol classes use the same quantization bit number.

In this embodiment of the present application, the tensor protocol class includes a second element type and a numerical value, or the tensor protocol class may include only a numerical value. The tensor protocol class is used for representing network parameters of the target AI network, and the name, the value and other information of the network parameters are represented through the second element type and the value.

Optionally, the second element type is characterized by an integer length of the quantization bit number, i.e. the second element type only needs to indicate the length.

Optionally, the second element type and the first element type in the parameter information protocol class have a corresponding relationship, for example, the second element type and the first element type are used to describe the same network parameter, and the corresponding relationship may be ordered according to an index order of the network parameters of the target AI network.

Optionally, the arrangement order of the second element types in the tensor protocol class is the same as the arrangement order of the first element types in the parameter information protocol class. It should be noted that the name and/or index of the network parameter may be omitted in the tensor protocol class.

In this embodiment of the present application, the ONNX file structure includes a protocol class list, where one protocol class list corresponds to one protocol class type, and the protocol class type includes the node protocol class, the parameter information protocol class, the tensor protocol class, and the attribute protocol class. That is, the proto may be stored by a category type, and one type of proto corresponds to one protocol class list, for example, the ONNX file structure includes a node protocol class, a parameter information protocol class, and a tensor protocol class, and the ONNX file structure includes three lists, that is, a node protocol list, a parameter information protocol class list, and a tensor protocol class list, respectively. It should be noted that, the protocol class list may include a plurality of corresponding protocol classes, for example, the node protocol class list stores a plurality of node protocol classes, and the plurality of protocol classes are arranged in a certain order. The protocol classes stored in the protocol class list comprise corresponding list indications by which the positions of the protocol classes in the list are indicated.

It should be noted that, for an AI network, the reading order of the nodes of the agreed network may be implemented, for example, from inside to outside or from top to bottom, for example, all the node protocol classes are arranged in order, i.e. the input of each node must be some node or some nodes before itself, so that the list indication of each node is not needed.

Optionally, if the length of each node protocol class is also agreed, or the base station is configured, the corresponding node protocol class can be found through offset, and when the node protocol class is deleted, inserted and added, the corresponding position can be directly found. Likewise, for the parameter information protocol class and the tensor protocol class, the parameter information protocol class and the tensor protocol class may be arranged into corresponding lists according to the order recorded in the node protocol class.

Optionally, the method may further include:

the first end sends indication information to the second end; or,

the first end sends the ONNX file structure to a second end, including:

the first end sends the ONNX file structure carrying the indication information to a second end;

the indication information is used for indicating at least one of a starting position and an ending position of the protocol class list in the ONNX file structure.

That is, the indication information may be sent separately or carried in an ONNX file structure. Alternatively, each protocol class list may be formed separately into an ONNX file structure, and the second end can determine the start position and/or the end position of the protocol class list in the ONNX file structure based on the indication information.

Optionally, the second end may add a portion of the content to the ONNX file structure after receiving the ONNX file structure. For example, the first end is core network equipment, the second end is a base station, the ONNX file structure sent to the base station by the core network equipment only comprises node protocol type information, and the base station can supplement parameter information protocol type information and tensor protocol type information after receiving the ONNX file structure. For another example, the ONNX file structure sent to the base station by the core network device includes all node protocol information, all parameter information protocol information and part of tensor protocol information corresponding to the target AI network, and the base station supplements other tensor protocol information at the relevant position according to the starting position and the ending position of the list corresponding to the tensor protocol.

The protocol class list may be a packet of proto, for example, a packet or a segment according to the function or the requirement of proto.

In this embodiment, the ONNX file structure includes a predefined index, where the predefined index is used to characterize a preset network node. For example, the first end may define a number of network nodes with common functions or historical functions in advance, use the network nodes as preset network nodes, and characterize the preset network nodes through a predefined index, and the second end can obtain information such as functions, input/output dimensions and the like of the corresponding preset network nodes based on the predefined index. Through the setting of the predefined index, all the used network nodes do not need to be characterized by the node protocol class, so that the number of the node protocol classes in the ONNX file structure can be reduced, and the transmission overhead of the ONNX file structure is further saved.

Optionally, the network nodes include discrete fourier transform (Discrete Fourier Transform, DFT) nodes, inverse discrete fourier transform (Inverse Discrete Fourier Transform, IDFT) nodes, filtering nodes, and the like.

In the case that the update occurs in the target AI network, the preset network node may be an un-updated network node in the target AI network. Alternatively, when updating the target AI network, the non-updated network nodes may be regarded as preset network nodes, and these non-updated network nodes may be further characterized by predefined indexes in the ONNX file structure. Alternatively, the network nodes that are not updated may be combined into one node, which is indicated by a predefined index, for example, an index corresponding to the history node.

In this embodiment of the present application, the first end converts AI network information of a target AI network into an ONNX file structure, including:

the first end converts first AI network information of a target AI network into a first ONNX file structure and converts second AI network information of the target AI network into a second ONNX file structure, wherein the target protocol class included in the first ONNX file structure is different from the target protocol class included in the second ONNX file structure;

the first end sends the ONNX file structure to a second end, including:

the first end sends the first ONNX file structure and the second ONNX file structure to a second end.

That is, when generating the ONNX file structures, the first end can generate different ONNX file structures based on the types of the protocol classes, each of the ONNX file structures includes different protocol classes, and then send the ONNX file structures respectively. For example, the first ONNX file structure includes a node protocol class, the second ONNX file structure includes a parameter information protocol class, and the first end sends the two ONNX file structures to the second end respectively.

It should be noted that, based on the number of types of the protocol classes, the first end may generate a plurality of ONNX file structures, for example, if AI network information of the target AI network corresponds to a node protocol class, a parameter information protocol class, and a tensor protocol class, respectively, based on the three different protocol class types, the first end may generate three ONNX file structures, where the three ONNX file structures correspond to the node protocol class, the parameter information protocol class, and the tensor protocol class, respectively, and send the three ONNX file structures, respectively.

Of course, the first end may also include all protocol classes in an ONNX file structure.

Alternatively, there may be multiple cases in which the first ONNX file structure and the second ONNX file structure correspond to a protocol class. For example, the first ONNX file structure includes a node protocol class, and the second ONNX file structure includes a parameter information protocol class; or, the first ONNX file structure includes a node protocol class and a parameter information protocol class, and the second ONNX file structure includes a tensor protocol class; alternatively, the first ONNX file structure includes a node protocol class, and the second ONNX file structure includes a parameter information protocol class and a tensor protocol class.

For better understanding, the technical solutions of the embodiments of the present application are described below through specific embodiments.

The terminal and the network side equipment use a combined AI network to feed back channel state information (Channel State Information, CSI), namely, the terminal converts the channel information into CSI feedback information with a plurality of bits (bits) through the AI network and reports the CSI feedback information to the base station, and the base station receives the bit information fed back by the terminal and recovers the channel information through the AI network at the base station side.

Because the AI networks of the base station and the terminal need to perform joint training, different cell channel conditions may also need new network parameters, when the terminal accesses the network, the base station needs to send the network parameters used by the terminal to the terminal.

The network for CSI feedback can be divided into two parts, a terminal coding part and a base station decoding part, and the terminal only needs to obtain the network structure of the coding part.

The core network device sends the network structure of the coding part to the terminal through Non-access service (NAS) information, and the specific information may be an ONNX file structure including only the nodeboto information, or a file structure including a nodeboto list. And the core network equipment transmits the ONNX file structure to the base station side, and then the base station forwards the ONNX file structure to the terminal.

After receiving the NAS information of the core network equipment, if the NAS information is transparent to the base station, the base station directly forwards the NAS information, then saves the weight parameters into ONNX file structures only including ValueInfoProto and TensorProto, and sends the ONNX file structures to the terminal through radio resource control (Radio Resource Control, RRC) signaling.

If the NAS signaling base station can interpret, the base station can send the ONNX file structure of NodeProto and the ONNX file structure of the NodeProto including ValueInfoProto and TensorProto to the terminal together through RRC signaling, or can combine the ONNX file structure and the ValueInfoProto into one ONNX file structure to be sent to the terminal.

Optionally, the base station may supplement ValueInfoProto and tensorpro information into the ONNX file structure of NAS information, and still forward the ValueInfoProto and tensorpro information to the user according to NAS information.

In addition, the core network device may also send an ONNX file structure including NodeProto, valueInfoProto and tensorproco, where ValueInfoProto and tensorproco may be part or all of them. After receiving the NAS information, the base station sends the ONNX file structures of ValueInfoProto and TensorProto through RRC signaling if the NAS information is transparent and directly forwarded. If the network is non-transparent, the base station can send the ONNX file structure of the base station and the ONNX file structure of the core network equipment to the terminal through RRC, or send the ONNX file structure and the ONNX file structure of the core network equipment to the terminal after combining.

Referring to fig. 3, fig. 3 is a flowchart of another AI network information transmission method provided by an embodiment of the disclosure, and as shown in fig. 3, the method includes the following steps:

step 301, a second end receives an ONNX file structure sent by a first end, where the ONNX file structure is obtained by converting AI network information of a target AI network by the first end.

In this embodiment, the first end converts AI network information of the target AI network into an ONNX file structure, for example, if the AI network information includes a complete network structure and all network parameters of the target AI network, the first end expresses the network structure and the network parameters into the ONNX file structure based on the structural form of ONNX, for example, describes the network structure through a node protocol class (nodoproto) and a parameter information protocol class (ValueInfoProto), describes the network parameters through a tensor protocol class (tensorpro), and so on. The specific implementation manner of the first end to convert the AI network information into the ONNX file structure may refer to the description in the embodiment illustrated in fig. 2, which is not repeated in this embodiment.

In the embodiment of the application, the second end receives the ONNX file structure sent by the first end, and the second end converts the ONNX file structure into the AI network under the self neural network framework so as to realize training and application of the second end to the AI network. Therefore, two communication devices with different neural network frameworks can transmit the AI network information based on the ONNX file structure, and the transmission of the AI network information between the communication devices is prevented from being blocked.

Optionally, the target protocol class includes any one of the node protocol class, parameter information protocol class, tensor protocol class and attribute protocol class, and in the case that the number of the ONNX file structures is at least two, one of the ONNX file structures corresponds to one target protocol class;

the second end receives the ONNX file structure sent by the first end, and the ONNX file structure comprises any one of the following components:

the second end receives at least two ONNX file structures which are sent by the first end in a merging mode;

The second end receives at least two ONNX file structures sent by the first end respectively.

In this embodiment, each proto may be independently formed into an ONNX file structure, and thus one ONNX file structure corresponds to one type of proto. For example, if the AI network information of the target AI network includes a network structure and a weight parameter, the AI network information generates a node protocol class, a parameter information protocol class, and a tensor protocol class based on the ONNX structure, the node protocol class correspondingly generates an ONNX file structure, the parameter information protocol class correspondingly generates an ONNX file structure, and the tensor protocol class correspondingly generates an ONNX file structure, so that three ONNX file structures are also obtained.

Alternatively, the first end may send the three ONNX file structures to the second end, for example, send the three ONNX file structures sequentially; alternatively, the first end may combine the three ONNX file structures and send the combined three ONNX file structures to the second end at a time. Furthermore, the second end can execute corresponding receiving actions, and for the received ONNX file structure, it can be determined that one ONNX file structure corresponds to one proto type, so that the second end can convert the ONNX file structure into AI network information under the self neural network frame more conveniently.

Optionally, in a case that the ONNX file structure includes at least one of a node protocol class and a parameter information protocol class, after the second end receives the ONNX file structure sent by the first end, the method further includes:

and the second end receives the value of the network parameter of the target AI network sent by the first end.

Specifically, after the first end transmits the ONNX file structure including the node protocol class and the parameter information protocol class to the second end, the first end may further send the value of the network parameter of the target AI network to the second end. It will be appreciated that since the first end has sent the ONNX file structure containing the node protocol class and the parameter information protocol class to the second end, the number and overhead of network parameters by the second end are also known. The node protocol class and the parameter information protocol class may be contained in one ONNX file structure, or may be respectively corresponding to one ONNX file structure.

Optionally, the second end receives the value of the network parameter of the target AI network sent by the first end through the data channel.

Optionally, in case the ONNX file structure comprises a node protocol class, the node protocol class comprises the second digital index for characterizing the input and output of the target AI network.

Optionally, in a case where the number of inputs and outputs of the target AI network is plural and continuous, the node protocol class includes a target numerical index, and at least one of the number of inputs and the number of outputs of the target AI network;

wherein the target numerical index is a numerical index in the second numerical index that identifies a first one of a plurality of inputs and outputs of the target AI network.

Optionally, one of the parameter information protocol classes is used for characterizing at least one of an input, an output and a network parameter corresponding to one network node in the target AI network, and the parameter information protocol class includes a first element type and dimension information.

Optionally, the first element type is characterized by an integer length of the quantization bit number.

Optionally, the first element type corresponds to a network node or a network parameter of the target AI network.

Optionally, the dimension value is the number of values included in the corresponding dimension, and the values are normalized values.

Optionally, the parameter information protocol class includes an indication parameter for indicating a position of the non-zero value.

Optionally, the tensor protocol class includes a second element type and a value, or the tensor protocol class includes a value.

Optionally, the second element type is characterized by an integer length of the quantization bit number.

Optionally, the second element type corresponds to the first element type.

Optionally, the arrangement order of the second element types in the tensor protocol class is the same as the arrangement order of the first element types in the parameter information protocol class.

Optionally, the ONNX file structure includes a protocol class list, one of the protocol class lists corresponds to one of protocol class types, and the protocol class types include the node protocol class, the parameter information protocol class, the tensor protocol class, and the attribute protocol class.

Optionally, the method further comprises:

the second end receives the indication information sent by the first end; or,

the second end receives the ONNX file structure sent by the first end, and the ONNX file structure comprises:

the second end receives an ONNX file structure carrying indication information sent by the first end;

In this embodiment, the ONNX file structure includes a predefined index, where the predefined index is used to characterize a preset network node.

Optionally, in the case that the target AI network is updated, the preset network node is an un-updated network node in the target AI network.

It should be noted that, when the AI network information transmission method provided in the embodiment of the present application is applied to the second end, corresponding to the AI network information transmission method provided in the embodiment of fig. 2 and applied to the first end, the specific implementation process of the relevant steps and the relevant concepts related to the ONNX file structure in the embodiment of the present application may refer to the description in the embodiment of the method described in fig. 2, and for avoiding repetition, a detailed description is omitted here.

In the embodiment of the application, the second end receives the ONNX file structure sent by the first end, and then the second end can convert the ONNX file structure into the AI network under the self neural network framework, so that training and application of the second end to the AI network are realized. Therefore, two communication devices with different neural network frameworks can transmit the AI network information based on the ONNX file structure, and the transmission of the AI network information between the communication devices is prevented from being blocked.

According to the AI network information transmission method provided by the embodiment of the application, the execution main body can be an AI network information transmission device. In the embodiment of the present application, an AI network information transmission device executes an AI network information transmission method by taking an AI network information transmission device as an example, and the AI network information transmission device provided in the embodiment of the present application is described.

Referring to fig. 4, fig. 4 is a block diagram of an AI-network information transmission apparatus according to an embodiment of the disclosure, and as shown in fig. 4, an AI-network information transmission apparatus 400 includes:

a conversion module 401, configured to convert AI network information of a target AI network into an open neural network switched ONNX file structure;

a sending module 402, configured to send the ONNX file structure to a second end.

Optionally, the ONNX file structure includes a target protocol class, and the target protocol class includes at least one of the following: a node protocol class;

parameter information protocol class;

tensor protocol class;

attribute protocol class.

Optionally, the ONNX file structure includes at least one computational graph protocol class, and the computational graph protocol class includes the target protocol class.

The sending module 402 is further configured to perform any one of the following:

merging and transmitting at least two ONNX file structures to a second end;

and respectively sending at least two ONNX file structures to a second end.

Optionally, in case the ONNX file structure includes at least one of a node protocol class and a parameter information protocol class, the sending module 402 is further configured to:

and sending the numerical value of the network parameter of the target AI network to a second end.

Optionally, the sending module 402 is further configured to:

and sending the numerical value of the network parameter of the target AI network to a second end through a data channel.

Optionally, the ONNX file structure includes a first digital index of integer type and a second digital index, where the first digital index is used to identify the target protocol class, one of the target protocol classes corresponds to one of the first digital indexes, and the second digital index is used to identify at least one of network parameters, inputs, and outputs of the target AI network.

Optionally, the second element type corresponds to the first element type.

Optionally, the sending module is further configured to:

sending indication information to the second end; or,

sending the ONNX file structure carrying the indication information to a second end;

Optionally, the ONNX file structure comprises a predefined index, the predefined index being used to characterize a preset network node.

Optionally, the conversion module 401 is further configured to:

converting first AI network information of a target AI network into a first ONNX file structure, and converting second AI network information of the target AI network into a second ONNX file structure, wherein the target protocol class included in the first ONNX file structure is different from the target protocol class included in the second ONNX file structure;

The sending module 402 is further configured to:

transmitting the first ONNX file structure and the second ONNX file structure to a second end;

optionally, the first ONNX file structure includes a node protocol class, and the second ONNX file structure includes a parameter information protocol class; or,

the first ONNX file structure comprises a node protocol class and a parameter information protocol class, and the second ONNX file structure comprises a tensor protocol class; or,

the first ONNX file structure comprises a node protocol class, and the second ONNX file structure comprises a parameter information protocol class and a tensor protocol class.

In this embodiment of the present application, the device converts AI network information of a target AI network applicable to a self-neural network framework into an ONNX file structure, and then sends the ONNX file structure to a second end, so that the second end can convert the ONNX file structure into an AI network under the self-neural network framework. Therefore, two communication devices with different neural network frameworks can transmit the AI network information based on the ONNX file structure, and the transmission of the AI network information between the communication devices is prevented from being blocked.

The AI-network information transmission apparatus 400 in this embodiment of the application may be an electronic device, for example, an electronic device with an operating system, or may be a component in an electronic device, for example, an integrated circuit or a chip. The electronic device may be a terminal, or may be other devices than a terminal. By way of example, terminals may include, but are not limited to, the types of terminals 11 listed above, other devices may be servers, network attached storage (Network Attached Storage, NAS), etc., and embodiments of the application are not specifically limited.

The AI network information transmission apparatus 400 provided in this embodiment of the present application can implement each process implemented by the method embodiment described in fig. 2, and achieve the same technical effects, and for avoiding repetition, a detailed description is omitted herein.

Referring to fig. 5, fig. 5 is a block diagram of another AI-network information transmission apparatus provided in an embodiment of the disclosure, and as shown in fig. 5, the AI-network information transmission apparatus 500 includes:

the receiving module 501 is configured to receive an ONNX file structure sent by a first end, where the ONNX file structure is obtained by converting AI network information of a target AI network by the first end.

parameter information protocol class;

tensor protocol class;

attribute protocol class.

The receiving module 501 is further configured to perform any one of the following:

receiving at least two ONNX file structures which are transmitted by a first end in a merging way;

and receiving at least two ONNX file structures respectively sent by a first end.

Optionally, in case the ONNX file structure includes at least one of a node protocol class and a parameter information protocol class, the receiving module 501 is further configured to:

and receiving the value of the network parameter of the target AI network sent by the first end.

Optionally, the receiving module 501 is further configured to:

and receiving the numerical value of the network parameter of the target AI network sent by the first end through a data channel.

Optionally, the second element type corresponds to the first element type.

Optionally, the receiving module 501 is further configured to:

receiving indication information sent by the first end; or,

receiving an ONNX file structure carrying indication information sent by a first end;

In this embodiment of the present application, the device may receive an ONNX file structure sent by the first end, and then the device may convert the ONNX file structure into an AI network under a self neural network framework, so as to implement training and application on the AI network. Therefore, even two communication devices with different neural network frameworks can transmit the AI network information based on the ONNX file structure, and the transmission of the AI network information between the communication devices is prevented from being blocked.

The AI-network information transfer apparatus 500 in this embodiment of the application may be an electronic device, for example, an electronic device with an operating system, or may be a component in an electronic device, for example, an integrated circuit or a chip. The electronic device may be a terminal, or may be other devices than a terminal. By way of example, terminals may include, but are not limited to, the types of terminals 11 listed above, other devices may be servers, network attached storage (Network Attached Storage, NAS), etc., and embodiments of the application are not specifically limited.

The AI network information transmission apparatus 500 provided in this embodiment of the present application can implement each process implemented by the method embodiment illustrated in fig. 3, and achieve the same technical effects, and for avoiding repetition, a detailed description is omitted herein.

Optionally, as shown in fig. 6, the embodiment of the present application further provides a communication device 600, including a processor 601 and a memory 602, where the memory 602 stores a program or an instruction that can be executed on the processor 601, and the program or the instruction implements each step of the embodiment of the AI network information transmission method described in fig. 2 or fig. 3 when executed by the processor 601, and achieves the same technical effects. In order to avoid repetition, a description thereof is omitted.

The embodiment of the application further provides a terminal, and each implementation process and implementation manner of the embodiment of the method of fig. 2 or fig. 3 are applicable to the embodiment of the terminal, and the same technical effects can be achieved. Specifically, fig. 7 is a schematic hardware structure of a terminal for implementing an embodiment of the present application.

The terminal 700 includes, but is not limited to: at least some of the components of the radio frequency unit 701, the network module 702, the audio output unit 703, the input unit 704, the sensor 705, the display unit 706, the user input unit 707, the interface unit 708, the memory 709, and the processor 710.

Those skilled in the art will appreciate that the terminal 700 may further include a power source (e.g., a battery) for powering the various components, and that the power source may be logically coupled to the processor 710 via a power management system so as to perform functions such as managing charging, discharging, and power consumption via the power management system. The terminal structure shown in fig. 7 does not constitute a limitation of the terminal, and the terminal may include more or less components than shown, or may combine certain components, or may be arranged in different components, which will not be described in detail herein.

It should be appreciated that in embodiments of the present application, the input unit 704 may include a graphics processing unit (Graphics Processing Unit, GPU) 7041 and a microphone 7042, with the graphics processor 7041 processing image data of still pictures or video obtained by an image capturing device (e.g., a camera) in a video capturing mode or an image capturing mode. The display unit 706 may include a display panel 7061, and the display panel 7061 may be configured in the form of a liquid crystal display, an organic light emitting diode, or the like. The user input unit 707 includes at least one of a touch panel 7071 and other input devices 7072. The touch panel 7071 is also referred to as a touch screen. The touch panel 7071 may include two parts, a touch detection device and a touch controller. Other input devices 7072 may include, but are not limited to, a physical keyboard, function keys (e.g., volume control keys, switch keys, etc.), a trackball, a mouse, a joystick, and so forth, which are not described in detail herein.

In this embodiment, after receiving downlink data from the network side device, the radio frequency unit 701 may transmit the downlink data to the processor 710 for processing; in addition, the radio frequency unit 701 may send uplink data to the network side device. Typically, the radio unit 701 includes, but is not limited to, an antenna, an amplifier, a transceiver, a coupler, a low noise amplifier, a duplexer, and the like.

The memory 709 may be used to store software programs or instructions and various data. The memory 709 may mainly include a first storage area storing programs or instructions and a second storage area storing data, wherein the first storage area may store an operating system, application programs or instructions (such as a sound playing function, an image playing function, etc.) required for at least one function, and the like. Further, the memory 709 may include volatile memory or nonvolatile memory, or the memory 709 may include both volatile and nonvolatile memory. The nonvolatile Memory may be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically Erasable EPROM (EEPROM), or a flash Memory. The volatile memory may be random access memory (Random Access Memory, RAM), static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (ddr SDRAM), enhanced SDRAM (Enhanced SDRAM), synchronous DRAM (SLDRAM), and Direct RAM (DRRAM). Memory 709 in embodiments of the present application includes, but is not limited to, these and any other suitable types of memory.

Processor 710 may include one or more processing units; optionally, processor 710 integrates an application processor that primarily processes operations involving an operating system, user interface, application programs, and the like, and a modem processor that primarily processes wireless communication signals, such as a baseband processor. It will be appreciated that the modem processor described above may not be integrated into the processor 710.

In one implementation of the embodiments of the present application, the terminal 700 is the first end. The processor 710 is configured to convert AI network information of a target AI network into an open neural network switched ONNX file structure;

and the radio frequency unit 701 is configured to send the ONNX file structure to a second end.

Optionally, the ONNX file structure includes a target protocol class, and the target protocol class includes at least one of the following:

a node protocol class;

parameter information protocol class;

tensor protocol class;

attribute protocol class.

The radio frequency unit 701 is configured to perform any one of the following:

merging and transmitting at least two ONNX file structures to a second end;

and respectively sending at least two ONNX file structures to a second end.

Optionally, in case the ONNX file structure includes at least one of a node protocol class and a parameter information protocol class, the radio frequency unit 701 is configured to:

Optionally, the radio frequency unit 701 is configured to:

Optionally, the second element type corresponds to the first element type.

Optionally, the radio frequency unit 701 is further configured to:

sending indication information to the second end; or,

Optionally, the processor 710 is further configured to:

The radio frequency unit 701 is further configured to:

In another implementation of the embodiment of the present application, the terminal 700 may be used as the second terminal. The radio frequency unit 701 is configured to receive an ONNX file structure sent by a first end, where the ONNX file structure is obtained by converting AI network information of a target AI network by the first end.

a node protocol class;

parameter information protocol class;

tensor protocol class;

attribute protocol class.

The radio frequency unit 701 is configured to perform any one of the following:

Optionally, in the case that the ONNX file structure includes at least one of a node protocol class and a parameter information protocol class, the radio frequency unit 701 is further configured to:

Optionally, the radio frequency unit 701 is further configured to:

receiving indication information sent by the first end; or,

The terminal 700 provided in this embodiment of the present application may be used as the first end or the second end to execute the AI network information transmission method described in fig. 2 or fig. 3, and may achieve the same technical effects, which are not described herein again.

The embodiment of the application further provides a network side device, and each implementation process and implementation manner of the embodiments of the methods described in fig. 2 and fig. 3 are applicable to the embodiment of the network side device, and the same technical effects can be achieved.

Specifically, the embodiment of the application also provides network side equipment. As shown in fig. 8, the network side device 800 includes: an antenna 81, a radio frequency device 82, a baseband device 83, a processor 84 and a memory 85. The antenna 81 is connected to a radio frequency device 82. In the uplink direction, the radio frequency device 82 receives information via the antenna 81, and transmits the received information to the baseband device 83 for processing. In the downlink direction, the baseband device 83 processes information to be transmitted, and transmits the processed information to the radio frequency device 82, and the radio frequency device 82 processes the received information and transmits the processed information through the antenna 81.

The method performed by the network side device in the above embodiment may be implemented in the baseband apparatus 83, and the baseband apparatus 83 includes a baseband processor.

The baseband device 83 may, for example, include at least one baseband board, where a plurality of chips are disposed, as shown in fig. 8, where one chip, for example, a baseband processor, is connected to the memory 85 through a bus interface, so as to call a program in the memory 85 to perform the network device operation shown in the above method embodiment.

The network-side device may also include a network interface 86, such as a common public radio interface (common public radio interface, CPRI).

Specifically, the network side device 800 of the embodiment of the present invention further includes: instructions or programs stored in the memory 85 and executable on the processor 84, the processor 84 invokes the instructions or programs in the memory 85 to perform the methods performed by the modules shown in fig. 4 or fig. 5, and achieve the same technical effects, and are not repeated here.

Specifically, the embodiment of the application also provides another network side device. As shown in fig. 9, the network side device 900 includes: a processor 901, a network interface 902, and a memory 903. The network interface 902 is, for example, a common public radio interface (common public radio interface, CPRI).

Specifically, the network side device 900 of the embodiment of the present invention further includes: instructions or programs stored in the memory 903 and executable on the processor 901, the processor 901 invokes the instructions or programs in the memory 903 to perform the methods performed by the modules shown in fig. 4 or fig. 5, and achieve the same technical effects, so that repetition is avoided and thus they are not described herein.

The embodiment of the present application further provides a readable storage medium, where a program or an instruction is stored, and when the program or the instruction is executed by a processor, the processes of the embodiment of the method described in fig. 2 or fig. 3 are implemented, and the same technical effects can be achieved, so that repetition is avoided, and no further description is given here.

Wherein the processor is a processor in the terminal described in the above embodiment. The readable storage medium includes computer readable storage medium such as computer readable memory ROM, random access memory RAM, magnetic or optical disk, etc.

The embodiment of the present application further provides a chip, where the chip includes a processor and a communication interface, where the communication interface is coupled to the processor, and the processor is configured to run a program or an instruction, implement each process of the method embodiment described in fig. 2 or fig. 3, and achieve the same technical effect, so that repetition is avoided, and no further description is given here.

It should be understood that the chips referred to in the embodiments of the present application may also be referred to as system-on-chip chips, or the like.

The embodiments of the present application further provide a computer program/program product, which is stored in a storage medium, and executed by at least one processor to implement the respective processes of the embodiments of the methods described in fig. 2 or fig. 3, and achieve the same technical effects, and are not repeated herein.

The embodiment of the application also provides a communication system, which comprises: a terminal and a network side device, where the terminal may be used to perform the steps of the method described in fig. 2, and the network side device may be used to perform the steps of the method described in fig. 3, or the terminal may be used to perform the steps of the method described in fig. 3, and the network side device may be used to perform the steps of the method described in fig. 2.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element. Furthermore, it should be noted that the scope of the methods and apparatus in the embodiments of the present application is not limited to performing the functions in the order shown or discussed, but may also include performing the functions in a substantially simultaneous manner or in an opposite order depending on the functions involved, e.g., the described methods may be performed in an order different from that described, and various steps may also be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solutions of the present application may be embodied essentially or in a part contributing to the prior art in the form of a computer software product stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk), comprising several instructions for causing a terminal (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method described in the embodiments of the present application.

The embodiments of the present application have been described above with reference to the accompanying drawings, but the present application is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and many forms may be made by those of ordinary skill in the art without departing from the spirit of the present application and the scope of the claims, which are also within the protection of the present application.

Claims

1. An artificial intelligence AI network information transmission method, comprising:

the first end sends the ONNX file structure to a second end.

2. The method of claim 1, wherein the ONNX file structure comprises a target protocol class comprising at least one of:

a node protocol class;

parameter information protocol class;

tensor protocol class;

attribute protocol class.

3. The method of claim 2, wherein the ONNX file structure comprises at least one computational graph protocol class, the computational graph protocol class comprising the target protocol class.

4. The method according to claim 2, wherein the target protocol class comprises any one of the node protocol class, parameter information protocol class, tensor protocol class and attribute protocol class, and wherein in case the number of the ONNX file structures is at least two, one of the ONNX file structures corresponds to one of the target protocol classes;

the first end sends the ONNX file structure to a second end, and the ONNX file structure comprises any one of the following:

the first end merges and sends at least two ONNX file structures to a second end;

and the first end sends at least two ONNX file structures to the second end respectively.

5. The method according to any one of claims 2-4, wherein, in case the ONNX file structure comprises at least one of a node protocol class and a parameter information protocol class, the first end sends the ONNX file structure to a second end, the method further comprises:

6. The method of claim 5, wherein the first end transmitting the value of the network parameter of the target AI network to a second end comprises:

7. The method of any of claims 2-4, wherein the ONNX file structure includes a first digital index of an integer type and a second digital index, the first digital index being used to identify the target protocol class, one of the target protocol classes corresponding to one of the first digital indexes, the second digital index being used to identify at least one of network parameters, inputs, and outputs of the target AI network.

8. The method of claim 7, wherein, in the case where the ONNX file structure includes a node protocol class, the node protocol class includes the second digital index for characterizing inputs and outputs of the target AI network.

9. The method of claim 8, wherein the node protocol class includes a target numerical index and at least one of a number of inputs and outputs of the target AI network if the number of inputs and outputs of the target AI network is multiple and continuous;

10. The method of any of claims 2-4, wherein one of the parameter information protocol classes is used to characterize at least one of an input, an output, and a network parameter corresponding to one of the network nodes in the target AI network, the parameter information protocol class including a first element type and dimension information.

11. The method of claim 10, wherein the first element type is characterized by an integer length of a quantization bit number.

12. The method of claim 10, wherein the first element type corresponds to a network node or network parameter of the target AI network.

13. The method of claim 10, wherein the dimension information in the parameter information protocol class includes a corresponding dimension number value and dimension size value, the dimension number value in the parameter information protocol class is arranged based on a preset order, and the dimension size value in the parameter information protocol class is arranged based on the preset order, the preset order being a preset arrangement order of network parameters of the target AI network.

14. The method of claim 13, wherein the dimension size value is a number of values included in the corresponding dimension, and the number is a normalized number.

15. The method according to claim 10, wherein in the case that a target parameter information protocol class is used to characterize a target network parameter, the target parameter information protocol class includes a numerical position of a maximum numerical value, the maximum numerical value is a maximum value of the numerical values of the target network parameter, a target tensor protocol class corresponding to the target network parameter includes a quantized value of a ratio of each remaining numerical value in the target network parameter to the maximum numerical value, and the remaining numerical values are other numerical values except the maximum numerical value in a dimension corresponding to the target parameter information protocol class.

16. The method of claim 10, wherein the parameter information protocol class includes an indication parameter for indicating a location of a non-zero value.

17. The method of claim 10, wherein the tensor protocol class comprises a second element type and a value, or wherein the tensor protocol class comprises a value.

18. The method of claim 17, wherein the second element type is characterized by an integer length of a quantization bit number.

19. The method of claim 17, wherein the second element type corresponds to the first element type.

20. The method of claim 17, wherein the order of the second element types in the tensor protocol class is the same as the order of the first element types in the parametric information protocol class.

21. The method according to any of claims 2-4, wherein the ONNX file structure comprises a list of protocol classes, one of the list of protocol classes corresponding to one of the protocol class types comprising the node protocol class, parameter information protocol class, tensor protocol class and attribute protocol class.

22. The method of claim 21, wherein the method further comprises:

the first end sends indication information to the second end; or,

the first end sends the ONNX file structure to a second end, including:

23. The method according to any of claims 1-4, wherein the ONNX file structure comprises a predefined index, the predefined index being used to characterize a preset network node.

24. The method of claim 23, wherein the predetermined network node is an un-updated network node in the target AI network in the event of an update in the target AI network.

25. The method of any of claims 2-4, wherein the first end converting AI network information of the target AI network into an ONNX file structure, comprising:

the first end sends the ONNX file structure to a second end, including:

26. The method of claim 25, wherein the first ONNX file structure comprises a node protocol class and the second ONNX file structure comprises a parameter information protocol class; or,

27. An AI network information transmission method, comprising:

28. The method of claim 27, wherein the ONNX file structure comprises a target protocol class comprising at least one of:

a node protocol class;

parameter information protocol class;

tensor protocol class;

attribute protocol class.

29. The method of claim 28, wherein the ONNX file structure comprises at least one computational graph protocol class, the computational graph protocol class comprising the target protocol class.

30. The method of claim 28, wherein the target protocol class includes any one of the node protocol class, parameter information protocol class, tensor protocol class, and attribute protocol class, and wherein in the case that the number of the ONNX file structures is at least two, one of the ONNX file structures corresponds to one of the target protocol classes;

31. The method according to any one of claims 28-30, wherein, in case the ONNX file structure comprises at least one of a node protocol class and a parameter information protocol class, the second end receives the ONNX file structure sent by the first end, the method further comprises:

32. The method according to any of claims 28-30, wherein the ONNX file structure comprises a list of protocol classes, one of the list of protocol classes corresponding to one of the protocol class types comprising the node protocol class, parameter information protocol class, tensor protocol class and attribute protocol class.

33. The method of claim 32, wherein the method further comprises:

the second end receives the indication information sent by the first end; or,

34. An AI network information transmission apparatus, characterized by comprising:

35. An AI network information transmission apparatus, characterized by comprising:

36. A communication device comprising a processor and a memory storing a program or instructions executable on the processor, which when executed by the processor, implement the steps of the AI network information transfer method of any of claims 1-26, or implement the steps of the AI network information transfer method of any of claims 27-33.

37. A readable storage medium having stored thereon a program or instructions which, when executed by a processor, implements the steps of the AI network information transmission method of any of claims 1-26, or the steps of the AI network information transmission method of any of claims 27-33.