Method and device for selecting large language model by terminal in wireless cellular network

Abstract

The present application discloses a method for selecting a large language model by a terminal in a wireless cellular network, the method comprising: for a data processing requirement in a terminal application, the terminal periodically sends test data to multiple network nodes that deploy large language models according to network configuration information, and selects one or more network nodes to simultaneously apply the large language model to process the actual data processing requirements of the terminal according to the accuracy and feedback delay of the inference results returned by the network nodes. If the terminal selects multiple network nodes to simultaneously apply the large language model and return results, the terminal selects the result returned first.

Description

Technical Field

[0001] The present invention relates to data processing technology, and more particularly to a method for a terminal to select a large language model in a wireless cellular system. Background Art

[0002] With the advancement of chip computing power and artificial intelligence algorithms, mobile phones and other terminal devices also use large language models (LLM) for text, voice, image processing, wireless signal processing, etc. The large language model here generally refers to a deep learning model trained with a large amount of text data, which can generate natural language text or understand the meaning of language text. The large language model can also be referred to as a large model, which generally refers to an artificial intelligence model with more parameters. The large language model based on mobile phones can have a parameter set of hundreds of millions or even billions, which can handle some simple artificial intelligence applications. However, for some more complex tasks, such as image processing or audio and video processing, due to the limitations of the model, mobile phone hardware and other conditions, it is necessary to apply the large language model on the network side (for example, inference and training for video optimization). There may be multiple nodes on the network side that can provide large language models.

[0003] At present, the research on introducing large language model technology into terminal data processing has just begun, the signaling mechanism has not been determined, and there are many optimization problems that need to be solved. Summary of the invention

[0004] In view of this, the present invention provides the following technical solutions:

[0005] A method for selecting a large language model by a terminal in a wireless cellular network, applied to data processing of the terminal, is characterized by comprising:

[0006] In response to the data processing needs in a terminal application, the terminal periodically sends test data to multiple network nodes that deploy large language models according to network configuration information, and selects one or more network nodes to simultaneously apply the large language model to process the actual data processing needs of the terminal based on the accuracy and feedback delay of the inference results returned by the network nodes. If the terminal selects multiple network nodes to simultaneously apply the large language model and return results, the terminal selects the result that is returned first.

[0007] The method includes data processing requirements in the terminal application, including at least: text summarization, language translation, audio optimization and video optimization.

[0008] The method further includes network configuration information, which at least includes: the network node configures the terminal through wireless resource messages, non-access layer messages or upper layer application data, the period for the terminal to send test data to the network side, and the number and address information of the network nodes where the large language model is deployed.

[0009] The method further includes, based on the accuracy of the inference results and the feedback delay returned by the network nodes, selecting one or more network nodes to simultaneously apply the large language model to process the application requirements of the terminal, at least including: calculating the accuracy of the inference results and the feedback delay according to different weights to obtain an evaluation score for the application of the large language model of the network node, and selecting the network node with an evaluation score greater than a specific threshold value as a node for applying the large language model.

[0010] The method further includes applying a large language model, including at least: training, learning and reasoning of a large language model for text, audio or video processing.

[0011] A processing device for selecting a large language model by a terminal in a wireless cellular network, applied to data processing of the terminal, characterized in that it includes: for a data processing requirement in a terminal application, the terminal periodically sends test data to multiple network nodes that deploy large language models according to network configuration information, and selects one or more network nodes to simultaneously apply the large language model to process the actual data processing requirements of the terminal according to the accuracy and feedback delay of the inference results returned by the network nodes. If the terminal selects multiple network nodes to simultaneously apply the large language model and return results, the terminal selects the result returned first.

[0012] It can be seen from the above technical solution that, compared with the prior art, the embodiment of the present invention discloses a method for a terminal to select a large language model in a wireless cellular network, which is applied to data processing of the terminal. The method includes: for a data processing requirement in a terminal application, the terminal periodically sends test data to multiple network nodes that deploy large language models according to network configuration information, and selects one or more network nodes to simultaneously apply the large language model to process the actual data processing requirements of the terminal according to the accuracy and feedback delay of the inference results returned by the network nodes. If the terminal selects multiple network nodes to simultaneously apply the large language model and return the results, the terminal selects the result returned first. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0014] Figure 1 A flow chart of a method for selecting a large language model by a terminal in a wireless cellular network disclosed in an embodiment of the present invention;

[0015] Figure 2 A schematic diagram of a method for selecting a large language model by a terminal in a wireless cellular network disclosed in an embodiment of the present invention;

[0016] Figure 3 A schematic diagram of another method for selecting a large language model by a terminal in a wireless cellular network disclosed in an embodiment of the present invention;

[0017] Figure 4 A functional schematic diagram of selecting a large language model in a wireless terminal disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Figure 1 This is a flow chart of a method for selecting a large language model by a terminal in a wireless cellular network disclosed in an embodiment of the present invention. Figure 1 As shown, the flow chart includes the following steps:

[0020] Step 101: The network sends a message containing large language model configuration information to the terminal. The message can be a wireless resource message, a non-access layer message, or upper layer application data (for example, a message sent by an APP or an operating system). The message contains the test data sending period (for example, 1 minute or 1 hour), the number of network nodes where the large language model is deployed (for example, 3), and address information (IP addresses of nodes 1 to 3);

[0021] First, the terminal needs to determine which network nodes can perform large language model applications (e.g., inference for video optimization) based on network configuration information, and then send test data to the configured network nodes (step 102);

[0022] After the network node performs application reasoning on the test data, it returns the application reasoning results for the test data to the terminal.

[0023] Step 103: The terminal selects one or more network nodes to apply the large language model simultaneously based on the accuracy and feedback delay of the application reasoning results returned by the network nodes to process the actual needs of the terminal. The accuracy of the application reasoning results is evaluated based on the correlation between the application reasoning results and the standard answers configured in advance, and then the evaluation score is calculated. The following is an example formula,

[0024] M1 = 50% × Q1 + 50% × P1

[0025] M1: Evaluation score, the lowest score is 0 and the highest score is 100

[0026] Q1: The accuracy of the inference results of the large language model application, with the highest score being 100 (completely consistent with the standard answer) and the lowest score being 0 (completely inconsistent with the standard answer);

[0027] P1: response delay caused by large language model application inference, with the highest score being 100 (response delay < 3 seconds) and the lowest score being 0 (response delay greater than 30 seconds);

[0028] Set a threshold T1. If M1>T1, the terminal selects the network node to apply the large language model. If the scores of multiple network nodes are all greater than T1, multiple network nodes are selected to apply the large language model simultaneously.

[0029] After selecting the network node to which the large language model is applied, the terminal simultaneously sends the actual data that the user needs to process to the selected one or more network nodes for application of the large language model (step 104). For the sent actual data, if multiple network nodes return the results of their application reasoning, the terminal selects the most selected result as the final application reasoning result according to the time sequence in which the network nodes return the application reasoning results (step 105).

[0030] Figure 2 A schematic diagram of a method for selecting a large language model by a terminal in a wireless cellular network disclosed in an embodiment of the present invention, comprising a terminal, a network node 1 and a network node 2. The embodiment comprises the following steps:

[0031] Step 1: The network side (network node 1) sends a message containing large language model configuration information to the terminal through a control message or an application message;

[0032] Step 2: The terminal prepares test data and starts selecting network nodes that meet the application conditions of the large language model. In this embodiment, it is assumed that two nodes of the large language model are configured on the network side (network node 1 and network node 2);

[0033] Step 3: The terminal periodically sends test data to network node 1 and network node 2 according to the network configuration;

[0034] Step 4: Network node 1 and network node 2 perform application reasoning on the test data based on the large language model;

[0035] Step 5: Network node 1 and network node 2 return the application inference result of the test data to the terminal;

[0036] Step 6: The terminal calculates the score based on the received application inference result and feedback delay, compares it with the threshold, and determines that both network node 1 and network node 2 meet the conditions for applying the large language model;

[0037] Step 7: The terminal sends the user data that actually needs to be applied to reasoning to network node 1 and network node 2, and simultaneously applies the large language model (e.g., reasoning);

[0038] Step 8: Network node 1 and network node 2 simultaneously perform application reasoning on user data based on the large language model;

[0039] Step 9: For the same user data, both network node 1 and network node 2 return the application reasoning result to the terminal;

[0040] Step 10: The terminal selects the application reasoning result returned first as the application result of the large language model according to the time sequence of the application reasoning results fed back by the network nodes. For example, if network node 1 returns the application reasoning result first, the terminal adopts the reasoning result of network node 1. Otherwise, the terminal adopts the reasoning result of network node 2.

[0041] The terminal repeats steps 3-6 and dynamically selects network nodes to perform inference on user data. For example, when the load of network node 2 is heavy, resulting in a large feedback delay, and the terminal evaluates that the score of its large language model application is less than or equal to the threshold, only network node 1 is selected for the large language model application (e.g. Figure 3 as shown).

[0042] The above embodiments can play a good optimization role for various subsequent applications of the terminal (such as text, image, audio and video, and wireless signal processing), thereby improving user experience.

[0043] For the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the order of the actions described, because according to the present invention, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0044] The above embodiments disclosed in the present invention describe the method in detail. The method of the present invention can be implemented by various devices. Therefore, the present invention also discloses a device, and a specific embodiment is given below for detailed description.

[0045] Figure 4 The present invention is a functional diagram of a terminal selecting a large language model in a wireless cellular network disclosed in an embodiment of the present invention, and is a large language model selection optimization unit disclosed in an embodiment of the present invention. The data receiving module 401 is used to receive various data, such as raw data, network signaling, and internal instructions of the terminal that need to be processed by the large language model; the data processing module 402 is mainly used for processing user data and executing the strategy of selecting a large language model; the module 403 is a data sending module, which includes sending internal instructions to the terminal, interactive signaling with the network layer, raw data, etc.; the signaling processing module 404 is a network signaling parsing and encapsulation module.

[0046] The method and device for selecting a large language model by a terminal in a wireless cellular network described in this embodiment include: for a data processing requirement in a terminal application, the terminal periodically sends test data to multiple network nodes that deploy large language models according to network configuration information, and selects one or more network nodes to simultaneously apply the large language model to process the actual data processing requirements of the terminal according to the accuracy and feedback delay of the inference results returned by the network nodes. If the terminal selects multiple network nodes to simultaneously apply the large language model and return results, the terminal selects the result that is returned first.

[0047] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0048] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0049] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0050] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for selecting a large language model by a terminal in a wireless cellular network, applied to data processing of the terminal, characterized in that: include: In response to a data processing requirement in a terminal application, the terminal periodically sends test data to multiple network nodes that deploy a large language model according to network configuration information, and selects one or more network nodes to simultaneously apply the large language model to process the actual data processing requirements of the terminal according to the accuracy and feedback delay of the inference results returned by the network nodes. If the terminal selects multiple network nodes to simultaneously apply the large language model and return results, the terminal selects the result that is returned first; wherein, The selecting one or more network nodes to simultaneously apply the large language model to process the actual data processing requirements of the terminal according to the accuracy and feedback delay of the inference results returned by the network nodes at least includes: The accuracy and feedback delay of the inference results are calculated according to different weights to obtain the evaluation score of the large language model application of the network node. The network nodes with evaluation scores greater than a specific threshold value are selected as nodes that apply the large language model to process the actual needs of the terminal.

2. The method for selecting a large language model by a terminal in a wireless cellular network according to claim 1, characterized in that: The data processing requirements in the terminal application at least include: Text summarization, language translation, audio optimization, and video optimization.

3. The method for selecting a large language model by a terminal in a wireless cellular network according to claim 1, characterized in that: The network configuration information at least includes: The network node configures the period for the terminal to send test data to the network side and the number and address information of the network nodes deploying the large language model to the terminal through wireless resource messages, non-access layer messages or upper layer application data.

4. The method for selecting a large language model by a terminal in a wireless cellular network according to claim 1, characterized in that: The application of the large language model at least includes: Training, learning, and inference of large language models for text, audio, or video processing.

5. A processing device for selecting a large language model by a terminal in a wireless cellular network, applied to data processing of the terminal, characterized in that: The processing device includes at least one data processor, and the data processor is used to perform the following functions: In response to a data processing requirement in a terminal application, the terminal periodically sends test data to multiple network nodes that deploy a large language model according to network configuration information, and selects one or more network nodes to simultaneously apply the large language model to process the actual data processing requirements of the terminal according to the accuracy and feedback delay of the inference results returned by the network nodes. If the terminal selects multiple network nodes to simultaneously apply the large language model and return results, the terminal selects the result that is returned first; wherein, The selecting one or more network nodes to simultaneously apply the large language model to process the actual data processing requirements of the terminal according to the accuracy and feedback delay of the inference results returned by the network nodes at least includes: The accuracy and feedback delay of the inference results are calculated according to different weights to obtain the evaluation score of the large language model application of the network node. The network nodes with evaluation scores greater than a specific threshold value are selected as nodes that apply the large language model to process the actual needs of the terminal.

Images