CN-122023102-A - Heterogeneous computing force identification method and device apparatus, storage medium, and product

CN122023102ACN 122023102 ACN122023102 ACN 122023102ACN-122023102-A

Abstract

The application discloses a heterogeneous computing power identification method, a heterogeneous computing power identification device, heterogeneous computing power identification equipment, a heterogeneous computing power identification storage medium and a heterogeneous computing power identification product. The method comprises the steps of obtaining a calculation force component template, a first component detail template and calculation force description information to be identified, identifying the calculation force description information by utilizing the calculation force component template to obtain a component list, wherein the component list consists of first components in the calculation force description information, identifying the calculation force description information by utilizing the first component detail template and the component list to obtain hardware index data corresponding to the first components, and converting the component list and the hardware index data into a first multidimensional matrix by utilizing the first component detail template and the calculation force component template, wherein the first multidimensional matrix is structured data. The embodiment can improve the calculation efficiency in calculation force matching.

Inventors

WANG JIAN
LIN JINGYU
YANG JIANQING
WANG XU
WANG PANPAN

Assignees

中移（杭州）信息技术有限公司
中国移动通信集团有限公司

Dates

Publication Date: 20260512
Application Date: 20241112

Claims (11)

1. A heterogeneous computing force identification method, the method comprising: acquiring a computing force component template, a first component detail template and computing force description information to be identified, wherein the computing force description information comprises computing force resources of computing force equipment described by unstructured data, the computing force component template is used for describing components included in the computing force equipment, the first component detail template is used for describing performance indexes of the components, and the number of the components corresponds to the number of the first component detail templates one by one; identifying the computational power description information by using the computational power component template to obtain a component list, wherein the component list consists of first components in the computational power description information; identifying the calculation description information by utilizing the first component detail template and the component list to obtain hardware index data corresponding to the first component; And converting the component list and the hardware index data into a first multidimensional matrix by using the first component detail template and the power calculation component template, wherein the first multidimensional matrix is structured data.
2. The heterogeneous computing power identification method of claim 1, wherein the identifying the computing power description information by using the computing power component template to obtain a component list comprises: assembling the computing force component template and the computing force description information into a first function call request; A preset large language model is called by using the first function call request, and a first component in the computing power description information is determined through the large language model; and combining the first components into the component list.
3. The heterogeneous computational power identification method of claim 2, wherein the first function call request includes a first hint field and a first function call parameter, the first function call parameter being used to instruct the large language model to extract information corresponding to the first function call parameter in the first hint field; the assembling the computing force component template and the computing force description information into a first function call request comprises: And assigning the calculation force description information to the first prompt field, and assigning the calculation force component template to the first function call parameter to obtain the first function call request.
4. The heterogeneous computing force identification method of claim 1, wherein the component list comprises a plurality of first components, the first component detail templates comprising a plurality of first component detail templates that are in one-to-one correspondence with the plurality of first components; the identifying the computing power description information by using the first component detail template and the component list to obtain hardware index data corresponding to the first component comprises the following steps: selecting a second component from the plurality of first components; Selecting a second component detail template corresponding to the second component from the plurality of first component detail templates; combining the second component detail template and the computing power description information into a second function call parameter; And calling a preset large language model by using the second function call parameter, generating the hardware index data corresponding to the second component through the large language model, and re-executing the steps, namely selecting the second component from the plurality of first components until a plurality of hardware index data corresponding to the plurality of first components one by one is obtained.
5. The heterogeneous computing power identification method of claim 4, wherein the second function call parameter comprises a second hint field and a second function call parameter; the composing the second component detail template and the computing power description information into a second function call parameter comprises: and assigning the calculation force description information to the second prompt field, and assigning the second component detail template to the second function call parameter to obtain the second function call parameter.
6. The heterogeneous computing power identification method of claim 1, wherein the converting the component list and the hardware metric data into a first multidimensional matrix using the first component detail template, the computing power component template, comprises: And converting the component list and the hardware index data into a first multidimensional matrix by using the hardware sequence numbers and the index sequence numbers in the first component detail template and the power calculation component template, wherein the hardware sequence numbers are the sequence numbers of the first component, and the index needs to be the sequence numbers of the hardware index data.
7. The heterogeneous computing power identification method of claim 1, wherein after the converting the component list and the hardware metric data into a first multidimensional matrix using the first component detail template, the computing power component template, the method further comprises: Converting the computational power requirements into a second multidimensional matrix; Traversing a plurality of first multidimensional matrixes according to the second multidimensional matrixes to obtain a plurality of third multidimensional matrixes corresponding to the second multidimensional matrixes, wherein each first element in the third multidimensional matrixes is larger than or equal to a second element, and the second element is an element corresponding to the position of the first element in the second multidimensional matrixes; Respectively calculating a plurality of similarities between the second multidimensional matrix and the third multidimensional matrices; and determining a target third multidimensional matrix corresponding to the target similarity with the maximum similarity from the plurality of similarities as a target matrix, wherein the target matrix is a calculation force resource meeting the calculation force requirement.
8. A heterogeneous computing force identification device, the device comprising: The system comprises an acquisition module, a first component detail template and a first component detail template, wherein the acquisition module is used for acquiring a computing force component template, a first component detail template and computing force description information to be identified, the computing force description information comprises computing force resources of computing force equipment described by unstructured data, the computing force component template is used for describing components included in the computing force equipment, the first component detail template is used for describing performance indexes of the components, and the number of the components corresponds to the number of the first component detail templates one by one; The first recognition module is used for recognizing the calculation force description information by utilizing the calculation force component template to obtain a component list, wherein the component list consists of first components in the calculation force description information; The second identification module is used for identifying the calculation force description information by utilizing the first component detail template and the component list to obtain hardware index data corresponding to the first component; The first conversion module is used for converting the component list and the hardware index data into a first multidimensional matrix by utilizing the first component detail template and the power calculation component template, and the first multidimensional matrix is structured data.
9. A terminal device comprising a processor and a memory storing computer program instructions; the heterogeneous computational force identification method of any of claims 1-7 when executed by the processor.
10. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon computer program instructions, which when executed by a processor, implement the heterogeneous computational force identification method of any of claims 1-7.
11. A computer program product, characterized in that the computer product comprises a computer program which, when executed by a processor, implements the heterogeneous computational force identification method of any of claims 1-7.

Description

Heterogeneous computing force identification method and device apparatus, storage medium, and product Technical Field The application belongs to the technical field of cloud computing, and particularly relates to a heterogeneous computing power identification method, a heterogeneous computing power identification device, heterogeneous computing power identification equipment, a heterogeneous computing power storage medium and a heterogeneous computing power product. Background Heterogeneous computing power identification refers to a mechanism for identifying and describing computing resources having different architectures, different hardware components, and different computing capabilities. In a heterogeneous computing environment, computing resources may include a Central Processing Unit (CPU), a Graphics Processor (GPU), a programmable gate array (FPGA), a dedicated accelerator (e.g., an AI accelerator), and so forth. The purpose of heterogeneous computational power identification is to identify and categorize these different types of computing resources for efficient resource management and optimization in programming and task allocation. Heterogeneous computing power identification typically uses specific symbols, codes, or naming conventions to represent different computing resource types and capabilities. The existing calculation force identification method belongs to a descriptive and non-numeric identification method, and generally realizes calculation force identification and modeling by splitting calculation force equipment into sub-objects of different levels and nesting. The descriptive modeling can only be realized by means of semantic matching, intention recognition and the like when calculating the force matching or the force measurement, and cannot be realized by simple and efficient numerical calculation, so that the calculation efficiency is lower when the force calculation is performed. Disclosure of Invention The embodiment of the application provides a heterogeneous computing power identification method, a heterogeneous computing power identification device, heterogeneous computing power identification equipment, a heterogeneous computing power identification storage medium and a heterogeneous computing power identification product, and aims to solve the problem of low computing efficiency during computing of computing power. In a first aspect, an embodiment of the present application provides a heterogeneous computing power identification method, where the method includes: acquiring a computing force component template, a first component detail template and computing force description information to be identified, wherein the computing force description information comprises computing force resources of computing force equipment described by unstructured data, the computing force component template is used for describing components included in the computing force equipment, the first component detail template is used for describing performance indexes of the components, and the number of the components corresponds to the number of the first component detail templates one by one; identifying the computational power description information by using the computational power component template to obtain a component list, wherein the component list consists of first components in the computational power description information; identifying the calculation description information by utilizing the first component detail template and the component list to obtain hardware index data corresponding to the first component; And converting the component list and the hardware index data into a first multidimensional matrix by using the first component detail template and the power calculation component template, wherein the first multidimensional matrix is structured data. In a second aspect, an embodiment of the present application provides a heterogeneous computing force identification device, the device including: The system comprises an acquisition module, a first component detail template and a first component detail template, wherein the acquisition module is used for acquiring a computing force component template, a first component detail template and computing force description information to be identified, the computing force description information comprises computing force resources of computing force equipment described by unstructured data, the computing force component template is used for describing components included in the computing force equipment, the first component detail template is used for describing performance indexes of the components, and the number of the components corresponds to the number of the first component detail templates one by one; The first recognition module is used for recognizing the calculation force description information by utilizing the calculation force component template to obtain a component list, wherein the component list consists of first compon