System and method for deploying small model to examine supplier qualification performance through sideband box

Abstract

The invention discloses a method for examining supplier qualification performance through sideband box deployment of a small model, which comprises the following steps: S1, small model deployment: starting from deep analysis of a business demand end of the small model, selecting different basic models for targeted training according to diversified characteristics of supplier qualification performance files, such as a table type and a text type; s2, index extraction and identification: a system in a sideband box has strong file compatibility, and S3, evaluation assistance: index data of identification of the qualification performance attachment by the model and structured data submitted by a supplier are subjected to deep intelligent comparison. A small model is deployed by applying an edge box, so that model capability application in a closed network environment is realized, review time is greatly shortened, an AI model is utilized to perform index extraction and identification, mistakes and omissions of manual review are reduced, review accuracy is improved, key data and information in a review process are recorded, review work traceability is realized, and review efficiency is improved. And the review transparency is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of soft communication technology, and in particular to a system and method for reviewing supplier qualifications and performance using a small model of a sideband box deployment. Background Art

[0002] As the bidding and procurement industry becomes increasingly standardized and transparent, bidding agencies have an increasingly urgent need to verify supplier qualifications and performance. Traditional manual verification methods are not only inefficient, but also prone to errors, making it difficult to cope with large-scale, high-frequency verification needs. At the same time, the diversity and complexity of supporting materials (such as PDFs, pictures, scans, etc.) increase the difficulty of verification, and an efficient and accurate solution is urgently needed.

[0003] Applying models to identify and extract indicators from supplier supporting materials can greatly improve the efficiency of expert review. However, due to the closed network environment, the review site is often unable to directly use large models on the cloud or remote servers for data processing and analysis, thus limiting the effective application of model capabilities at the review site. To solve the above technical problems, this solution proposes a system and method for reviewing supplier qualifications and performance using a small model deployed in a sideband box. Summary of the invention

[0004] The purpose of the present invention is to solve at least one of the technical problems existing in the prior art, and to provide a system and method for reviewing supplier qualifications and performance by deploying a small model with a sideband box. By applying the sideband box to deploy the small model, the model capability application in a closed network environment is realized, which greatly shortens the review time. The AI ​​model is used to extract and identify indicators, which reduces the errors and omissions in manual review, improves the accuracy of the review, records the key data and information in the review process, realizes the traceability of the review work, and enhances the transparency of the review.

[0005] The present invention also provides a method for reviewing supplier qualifications and performance using the above-mentioned sideband box deployment small model, comprising the following steps:

[0006] S1. Small model deployment. Starting from an in-depth analysis of the business demand side of the small model, different basic models are selected for targeted training based on the diverse characteristics of the supplier's qualification and performance documents, such as table type and text type. The strictly trained and adapted small model is stably deployed in the sideband box.

[0007] S2. Index extraction and identification. The system in the sideband box has strong file compatibility and can smoothly read supplier qualification and performance files in various formats. It can make full use of the small model in the sideband box to perform in-depth index extraction and identification on the read files.

[0008] S3. Review assistance: The model conducts a deep and intelligent comparison of the indicator data identified in the qualification and performance attachments with the structured data submitted by the supplier. The recognition results are fully automatically verified based on the preset comprehensive and detailed review standards and rules. During the entire review process, the system records key data and information.

[0009] According to a method provided by the present invention for reviewing supplier qualifications and performance by deploying a small model in a sideband box, the small model deployment selects a model that is good at processing complex table structures for financial statement files, and selects a conventional OCR recognition model for files such as business licenses, thereby ensuring that the small model can accurately extract key indicators when processing various supplier qualifications and performance files, thereby improving recognition accuracy.

[0010] According to the method for reviewing supplier qualifications and performance by deploying a small model of a sideband box provided by the present invention, the index extraction and identification includes but is not limited to PDF, Word, Excel common formats, and various image format files. Ensure that all file types that may be submitted by suppliers are fully covered, providing a solid data foundation for subsequent accurate identification.

[0011] According to the method provided by the present invention for reviewing the performance of suppliers by deploying a small model in a sideband box, the small model for index extraction and identification integrates multiple cutting-edge algorithms such as advanced text detection, text recognition, semantic entity recognition and relationship extraction. It can comprehensively and accurately extract key information and its internal relationships in the target text in a single processing process, significantly improve the efficiency and accuracy of information extraction, accurately identify core indicators such as basic enterprise information, product quality, on-time delivery rate, service level, etc., and provide comprehensive and accurate decision-making basis for review experts.

[0012] According to the method for reviewing supplier qualifications and performance by deploying a small model of a sideband box provided by the present invention, after the extraction and identification of the indicators, the identification indicator results after complex processing are displayed on the interface in an intuitive and clear manner. It is convenient for review experts to quickly view and understand, whether presenting detailed data in the form of a list or displaying data relationships in the form of images, so that experts can quickly grasp the key points of supplier qualifications and performance at a glance, thereby greatly improving the review efficiency.

[0013] According to the method for reviewing supplier qualifications and performance by deploying a small model of a sideband box provided by the present invention, the data comparison in the review assistance is intelligently matched according to the data type and business logic, such as the unified processing of the date format and the conversion and comparison of numerical units. The review assistance covers multiple aspects of data integrity, logical rationality, and compliance. The accuracy of the comparison results is ensured, and they are presented to the review experts in an intuitive and easy-to-understand form, clearly informing them whether they match, so that the experts can quickly judge the authenticity and compliance of the supplier qualifications.

[0014] According to the present invention, a system for reviewing the qualification and performance of suppliers by deploying a small model in a sideband box includes: hardware components and software components. The hardware components include an AGX ORI N module integrated inside the sideband box, and a pre-installed l Ubuntu 20.04 operating system. The software components have a built-in lightweight operating system, which is optimized and designed to realize the rapid deployment and operation of AI models, and are equipped with a user-friendly operation interface. It has an AI processing capability of up to 200TOPS floating-point operations, is made of ultra-strong and lightweight aluminum alloy materials, and ensures excellent heat dissipation performance through conduction passive heat dissipation. It has a lightweight size and a novel appearance. It has a rich I0 interface type, such as USB, CAN, RS232, GPI0, and synchronization signals. It has a built-in 4G communication module and WI FI module, and supports synchronization signal input and output. A solid mounting bracket is reserved. It has significant characteristics such as strong computing power, high reliability, high integration, and low power consumption. It can fully support offline operation of small AI models. Even in poor network environments or without network connections, it can still run stably to ensure that the review work is not affected by network fluctuations.

[0015] Compared with the prior art, the system and method of the present invention, which uses a sideband box to deploy a small model to review the qualifications and performance of suppliers, realizes the application of model capabilities in a closed network environment by applying an edge box to deploy a small model, greatly shortens the review time, and uses AI models to extract and identify indicators, thereby reducing errors and omissions in manual review, improving the accuracy of the review, recording key data and information in the review process, realizing the traceability of the review work, and enhancing the transparency of the review. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention is further described below in conjunction with the accompanying drawings and embodiments;

[0017] Figure 1 A flow chart of a method for reviewing supplier qualification performance for a sideband box deployment small model of the present invention. DETAILED DESCRIPTION

[0018] This section will describe in detail the specific embodiments of the present invention. The preferred embodiments of the present invention are shown in the accompanying drawings. The purpose of the accompanying drawings is to supplement the description of the text part of the specification with graphics, so that people can intuitively and vividly understand each technical feature and the overall technical solution of the present invention, but it cannot be understood as a limitation on the scope of protection of the present invention.

[0019] Reference Figure 1In an embodiment of the present invention, a method for reviewing supplier qualifications and performance by deploying a small model in a sideband box includes the following steps: S1, small model deployment, starting from the in-depth analysis of the business demand side of the small model, according to the diversified characteristics of the supplier qualification and performance documents, such as table type and text type, different basic models are selected for targeted training, and the strictly trained and adapted small models are stably deployed in the sideband box. The small model deployment selects a model that is good at processing complex table structures for financial statement documents, and selects a conventional OCR recognition model for documents such as business licenses, so as to ensure that the small model can accurately extract key indicators when processing various supplier qualification and performance documents.

[0020] S2. Index extraction and recognition. The system in the sideband box has strong file compatibility and can smoothly read supplier qualification performance files in various formats. It makes full use of the small model in the sideband box to perform in-depth index extraction and recognition on the read files. Index extraction and recognition covers but is not limited to common formats such as PDF, Word, and Excel, as well as various image format files. The small model in index extraction and recognition integrates advanced text detection, text recognition, semantic entity recognition, and relationship extraction algorithms.

[0021] S3, review assistance, deeply and intelligently compares the indicator data identified by the model in the qualification performance attachment with the structured data submitted by the supplier, and fully automatically verifies the identification results according to the preset comprehensive and detailed review standards and rules. During the entire review process, the system records key data and information. The data comparison in the review assistance is intelligently matched according to the data type and business logic, such as the unified processing of date formats and the conversion and comparison of numerical units. The review assistance covers multiple aspects of data integrity, logical rationality, and compliance verification.

[0022] In a system in which a sideband box is deployed with a small model to review supplier qualifications and performance according to an embodiment of the present invention, the sideband box is configured as follows:

[0023] The AI ​​performance in the hardware configuration: It has a powerful computing power of 275TOPS, ensuring rapid response when processing complex supplier qualification performance documents.

[0024] Processor: 2048-core NVID IA Ampere arch itecture GPU with 56TensorCores 1.3GHz is used to provide efficient computing core support for model running.

[0025] GPU: 2048-core NVID IA Ampere architecture GPU equipped with 64 TensorCores, further improving graphics processing and data computing speed.

[0026] Memory: Equipped with 64GB 256-bit LPDDR5 204.8GB / s memory to ensure fast reading, writing and storage of data during system operation.

[0027] Storage: It has 2T large-capacity storage, which is used to safely store the operating system, various model files, and large amounts of data generated during the review process, ensuring that the data is not lost and can be called up at any time.

[0028] Network interface: Although a Wi-Fi module is optional, in order to ensure the stability of the review process, it mainly relies on wired connection to avoid affecting data transmission and model operation due to unstable wireless network signals.

[0029] Input and output interface: Equipped with USB 3.0, HDMI and other interfaces, convenient for connecting keyboard, mouse, monitor and file input devices such as USB flash drive, mobile hard disk, etc., to facilitate the operation of review experts and file import and export.

[0030] Operating system in software configuration: Ubuntu20.04 operating system is used to provide a stable and secure operating environment for the entire system. It also has good compatibility and scalability, which facilitates subsequent software updates and function upgrades.

[0031] Model operating environment: The model operating environment is built based on Python 3.10.9, making full use of Python's rich libraries and tools to ensure that the small model can run stably and efficiently and realize the operation of various complex algorithms.

[0032] User interaction interface: Users interact with the system through a desktop browser, and the system interacts with the AI ​​model service through the HTTP interface for data and command transmission. The interface design is concise and easy to operate, which conforms to the operating habits of the review experts and can guide the experts to successfully complete various review operations.

[0033] The embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in the technical field without departing from the purpose of the present invention.

Claims

1. A method for reviewing supplier qualifications and performance using a small model of sideband box deployment, characterized in that: The following steps are involved: S1. Small model deployment. Starting from an in-depth analysis of the business demand side of the small model, different basic models are selected for targeted training based on the diverse characteristics of the supplier's qualification and performance documents, such as table type and text type. The strictly trained and adapted small model is stably deployed in the sideband box. S2. Index extraction and identification. The system in the sideband box has strong file compatibility and can smoothly read supplier qualification and performance files in various formats. It can make full use of the small model in the sideband box to perform in-depth index extraction and identification on the read files. S3. Review assistance: The model conducts a deep and intelligent comparison of the indicator data identified in the qualification and performance attachments with the structured data submitted by the supplier. The recognition results are fully automatically verified based on the preset comprehensive and detailed review standards and rules. During the entire review process, the system records key data and information.

2. According to claim 1, a method for reviewing supplier qualifications and performance using a small sideband box deployment model, characterized in that: The small model deployment selects a model that is good at processing complex table structures for financial statement files, and selects a conventional OCR recognition model for files such as business licenses, thereby ensuring that the small model can accurately extract key indicators when processing various supplier qualification and performance files.

3. According to the method of claim 1, wherein the method comprises: The index extraction and identification include but are not limited to common formats such as PDF, Word, Excel, and various image format files.

4. According to claim 1, a method for reviewing supplier qualifications and performance using a small sideband box deployment model, characterized in that: The indicator extraction and recognition small and medium models integrate multiple cutting-edge algorithms such as advanced text detection, text recognition, semantic entity recognition and relationship extraction.

5. According to claim 1, a method for reviewing supplier qualifications and performance using a small sideband box deployment model, characterized in that: The indicator extraction and recognition small and medium models integrate multiple cutting-edge algorithms such as advanced text detection, text recognition, semantic entity recognition and relationship extraction.

6. According to claim 1, a method for reviewing supplier qualifications and performance using a small sideband box deployment model, characterized in that: The data comparison in the review assistance is intelligently matched according to the data type and business logic, such as the unified processing of date formats and the conversion and comparison of numerical units. The review assistance covers multiple aspects of verification, including data integrity, logical rationality, and compliance.

7. A system for reviewing supplier qualifications and performance by deploying a small model of a sideband box, which adopts a method for reviewing supplier qualifications and performance by deploying a small model of a sideband box as claimed in any one of claims 1 to 6, comprising: The hardware components and software components are characterized in that the hardware components include an AGX ORIN module integrated inside the sideband box and a pre-installed lbuntu 20.04 operating system. The software components have a built-in lightweight operating system that has been optimized to achieve rapid deployment and operation of AI models and are equipped with a user-friendly operating interface.

Images