Method and device for generating token by computing power acceleration of intelligent computing cloud platform

Abstract

This invention provides a method and apparatus for generating tokens using computing power acceleration on an intelligent computing cloud platform. It relates to the fields of intelligent computing centers, smart computing centers, computing infrastructure, and smart computing cloud technologies. The method for generating tokens using computing power acceleration on an intelligent computing cloud platform includes: Step S1, establishing a full-text index of a target data set; Step S2, responding to a received data query request, determining whether to enable document index acceleration; Step S3, if document index acceleration is enabled, extracting scalar filtering conditions and vector filtering conditions; Step S4, performing a full-text search based on the scalar filtering conditions to obtain a first data set; Step S5, performing a vector search on the vector index of the first data set based on the vector filtering conditions to obtain a matching second data set, and returning it to the user. This invention can improve the speed of vector database processing with scalar constraints on intelligent computing cloud platforms, reduce resource consumption, and increase token generation speed.

Description

Technical Field

[0001] This invention relates to the fields of intelligent computing centers, smart computing centers, computing infrastructure, and smart cloud technologies, specifically to a method and apparatus for generating tokens through computing power acceleration on an intelligent computing cloud platform. Background Technology

[0002] With the rapid development of artificial intelligence technology, "intelligent computing centers" and "smart computing centers" have emerged.

[0003] An "intelligent computing center" refers to a facility that provides the necessary computing power, data, and algorithms for artificial intelligence applications (such as the development, training, and token generation of deep learning models) by utilizing large-scale heterogeneous computing resources, including general-purpose and intelligent computing power. Intelligent computing centers encompass facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enablement.

[0004] "Intelligent computing center" includes, but is not limited to, "intelligent computing center".

[0005] "Intelligent computing center" or artificial intelligence computing center is a type of computing infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications, based on artificial intelligence theory and adopting artificial intelligence computing architecture.

[0006] "Computing power" is the core of "intelligent computing center" and "smart computing center". It is the ability of computer equipment or computing / data center to process parameters. It is the ability of computer hardware and software to work together to execute a certain computing requirement. It is the computing power to achieve the target result output by processing parameter data. It is a new type of productivity that integrates parameter computing power, network carrying capacity and data storage capacity. It mainly provides services to society through computing power infrastructure.

[0007] With the deep integration of artificial intelligence and big data technologies, vector databases, as the core infrastructure supporting high-dimensional semantic retrieval, are increasingly widely used in intelligent computing platforms. Especially in scientific computing fields such as weather forecasting, remote sensing image analysis, environmental monitoring, and urban sensing, for example, satellite cloud image embedding vectors, numerical weather prediction model output, and spatiotemporal sequence encoding, are often stored in vector form and achieve efficient semantic matching through near-nearest neighbor retrieval. Currently, mainstream vector database systems exhibit significant performance bottlenecks when handling such mixed queries: when users enter multiple scalar constraints in their query requests, the system's recall response time increases dramatically, with some queries taking tens of seconds or even minutes. More seriously, this process occurs before the first token is generated, directly lengthening the first token delay—the time from when a user initiates a request to receiving the first valid response. This not only drastically degrades the user experience but also renders the system "almost unusable" in high-concurrency production environments, severely restricting the real-time and large-scale deployment of AI-driven scientific computing services.

[0008] It is evident that since the emergence of intelligent computing centers, vector databases have faced the problem of slow pre-filtering speed and high resource consumption when processing mixed queries with scalar constraints, resulting in very slow token generation speed. This issue has always been a difficult problem to be solved in this field. Summary of the Invention

[0009] This invention provides a method and apparatus for accelerating token generation through computing power in an intelligent computing cloud platform, in order to solve the problem in the prior art where the vector database in the intelligent computing cloud platform has slow pre-filtering speed and high resource consumption when processing mixed queries with scalar constraints, resulting in slow token generation speed.

[0010] To solve the above problems, the present invention is implemented as follows: In a first aspect, the present invention provides a method for generating tokens through computing power acceleration on an intelligent computing cloud platform, comprising: Step S1: Divide the acquired target data set into multiple data segments, generate vector parameters and scalar parameters based on the data segments, update the vector parameters and scalar parameters to the vector database of the intelligent computing cloud platform, establish a vector index based on the vector parameters, and establish a full-text index of the data segments based on the scalar parameters. Step S2: In response to receiving a data query request from a user, determine whether to enable the document indexing acceleration function; Step S3: If it is determined that the document indexing acceleration function is enabled, then extract scalar filtering conditions and vector filtering conditions from the data query request; Step S4: Based on the scalar filtering conditions, perform full-text search in the full-text index of the vector database to obtain the first matching data set, wherein the vector database stores multiple vector data, and each vector data includes vector parameters and scalar parameters; Step S5: Based on the vector filtering conditions, perform vector retrieval on the vector index corresponding to the first data set to obtain a matching second data set, and return the second data set to the user.

[0011] In one embodiment, step S4 includes: Step S4.1: Receive the full-text query request sent by the client; Step S4.1: In the vector database, using the pre-built full-text inverted index and semantic vector index, keyword matching and semantic similarity calculation are performed simultaneously on the full-text index; Step S4.2: Generate a complete first data set based on the matching results, and return the first data set in a single response.

[0012] In one embodiment, step S1 includes: Step S1.1: Determine the scalar feature data and vector feature data of the data segment; Step S1.2: Generate the scalar parameters based on the scalar feature data, and generate the vector parameters based on the vector feature data.

[0013] In one embodiment, step S2 includes: Step S2.1: Obtain the computing power resource value currently occupied by the document indexing acceleration function currently enabled on the intelligent computing cloud platform; Step S2.2: In response to the computing power resource value being greater than the resource threshold, the document indexing acceleration function is disabled.

[0014] In one embodiment, prior to step S2.2, the step of determining the resource threshold includes: Step 2.3: Calculate the resource threshold based on the current GPU cluster utilization and / or remaining video memory and / or network bandwidth load of the intelligent computing cloud platform.

[0015] Secondly, the present invention also provides an apparatus for generating tokens through computing power acceleration on an intelligent computing cloud platform, comprising: A module is established to divide the acquired target data set into multiple data segments, generate vector parameters and scalar parameters based on the data segments, update the vector parameters and scalar parameters to the vector database of the intelligent computing cloud platform, establish a vector index based on the vector parameters, and establish a full-text index of the data segments based on the scalar parameters. The receiving module is used to respond to data query requests sent by users and determine whether to enable the document indexing acceleration function; The determination module is used to extract scalar filtering conditions and vector filtering conditions from the data query request if it is determined that the document indexing acceleration function is enabled. The first retrieval module is used to perform a full-text search in the full-text index of the vector database based on the scalar filtering conditions to obtain a matching first data set, wherein the vector database stores multiple vector data, and each vector data includes vector parameters and scalar parameters; The second retrieval module is used to perform vector retrieval on the vector index corresponding to the first data set based on the vector filtering conditions, obtain a matching second data set, and return the second data set to the user.

[0016] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps in the method for generating tokens by accelerating computing power on an intelligent computing cloud platform as described in the first aspect above.

[0017] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the method for generating tokens by computing power acceleration of the intelligent computing cloud platform described in the first aspect above.

[0018] Fifthly, the present invention also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps in the method for generating tokens by computing power acceleration of the intelligent computing cloud platform as described in the first aspect above.

[0019] In this invention, step S1 involves dividing the acquired target data set into multiple data segments, generating vector parameters and scalar parameters based on the data segments, updating the vector parameters and scalar parameters in the vector database of the intelligent computing cloud platform, establishing a vector index based on the vector parameters, and establishing a full-text index for the data segments based on the scalar parameters; step S2 involves determining whether to enable the document index acceleration function in response to receiving a data query request from a user; step S3 involves extracting scalar filtering conditions and vector filtering conditions from the data query request if the document index acceleration function is enabled; step S4 involves performing a full-text search in the full-text index of the vector database based on the scalar filtering conditions to obtain a matching first data set, wherein the vector database stores multiple vector data, each vector data including vector parameters and scalar parameters; and step S5 involves performing a vector search on the vector index corresponding to the first data set based on the vector filtering conditions to obtain a matching second data set, and returning the second data set to the user. In this way, through dynamic decision-making, conditional separation, phased utilization of indexes, and full-text retrieval, the computing power of the intelligent computing cloud platform is focused on high-value candidate sets, avoiding the consumption of floating-point computing resources on invalid data. While ensuring query accuracy, it achieves a double leap in computing power utilization and service efficiency, overcoming the performance bottlenecks caused by redundant data transportation and inefficient computing paths in existing technologies, and achieving the goal of accelerating token generation. Attached Figure Description

[0020] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the present invention will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a method for generating tokens using computing power acceleration provided by an intelligent computing cloud platform, as provided by the present invention. Figure 2 This is a flowchart of another method for generating tokens using computing power acceleration provided by the present invention; Figure 3 This is a flowchart of another method for generating tokens using computing power acceleration provided by the present invention; Figure 4 This is a structural diagram of a device for generating tokens using computing power acceleration provided by the present invention; Figure 5 This is a structural diagram of an electronic device provided by the present invention. Detailed Implementation

[0022] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] The “computing power” mentioned in this invention refers to: the ability of computer equipment or computing / data center to process information; the ability of computer hardware and software to work together to perform a certain computing requirement; the computing power to achieve the target result output by processing information data; and a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity, mainly providing services to society through computing power infrastructure.

[0024] The "computational power" (CP) described in this invention refers to the ability of a data center server to process data and output results. It is a comprehensive indicator of a data center's computing power, encompassing general computing power, supercomputing power, and intelligent computing power. The commonly used unit of measurement is floating-point operations per second (FLOPS, 1 EFLOPS = 10^18 FLOPS), with higher values ​​indicating stronger overall computing power. It is estimated that 1 EFLOPS is approximately the computing power output of 5 Tianhe-2A supercomputers, 500,000 mainstream server CPUs, or 2 million mainstream laptops. The calculation formula is: CP = CP 通用 +CP 智能 +CP 超级 .

[0025] The "Network Power" (NP) mentioned in this invention refers to the performance of data transmission capability of computing facilities, which includes comprehensive capabilities such as network architecture, network bandwidth, transmission latency, intelligent management and scheduling, and involves network transmission within and between data centers. It is a comprehensive indicator for measuring network transmission scheduling capability.

[0026] The "Storage Power" (SP) described in this invention refers to the comprehensive capabilities of a data center in four aspects: data storage capacity, performance, security and reliability, and green and low-carbon operation. It is a comprehensive indicator for measuring the data storage capacity of a data center, including external storage devices such as storage arrays and internal storage devices within servers. The commonly used unit of measurement for storage capacity is exabytes (EB, 1EB = 2^60 bytes), while the commonly used unit of measurement for performance is the number of read / write operations per second (IOPS / TB). Disaster recovery ratio is an important indicator of security and reliability.

[0027] The "computing infrastructure" mentioned in this invention refers to a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity, enabling centralized computing, storage, transmission, and application of information.

[0028] The "new information infrastructure" mentioned in this invention refers to network infrastructure such as 5G networks, fiber optic broadband networks, backbone networks, international communication networks, and satellite internet; computing infrastructure such as data centers, general computing centers, intelligent computing centers, and supercomputing centers; and new technology facilities such as artificial intelligence, blockchain, and quantum computing.

[0029] The “computing power” mentioned in this invention includes: general computing power, intelligent computing power, and supercomputing power.

[0030] The "general computing power" mentioned in this invention refers to the computing power provided by servers based on CPU (Central Processing Unit) chips, which is used to support basic general computing such as cloud computing and edge computing.

[0031] The "intelligent computing power" mentioned in this invention refers to: a computing platform deployed on a large scale based on dedicated chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), and ASIC (Application Specific Integrated Circuit) for various artificial intelligence innovative applications, such as natural language processing and machine vision.

[0032] The “supercomputing power” mentioned in this invention refers to the computing power provided by high-performance computing clusters such as supercomputers. It utilizes the centralized computing resources of multiple computer systems working in parallel and uses a dedicated operating system to handle extremely complex or data-intensive problems. It is mainly used for computing in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0033] The "intelligent computing center" described in this invention refers to a facility that utilizes large-scale heterogeneous computing resources, including general-purpose computing power (CPU) and intelligent computing power (GPU, FPGA, ASIC, etc.), to primarily provide the necessary computing power, data, and algorithms for artificial intelligence applications (such as the development, training, and token generation of artificial intelligence deep learning models). The intelligent computing center encompasses facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enablement.

[0034] The "intelligent computing cloud platform" mentioned in this invention, abbreviated as "intelligent computing cloud", refers to a cloud computing platform that integrates hardware and software resources based on an intelligent computing center.

[0035] The "intelligent computing center" mentioned in this invention includes, but is not limited to, "smart computing center".

[0036] The "intelligent computing center" mentioned in this invention, also known as an artificial intelligence computing center, is a type of computing infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications, based on artificial intelligence theory and adopting an artificial intelligence computing architecture.

[0037] The "computing center" mentioned in this invention refers to a facility that is mainly composed of infrastructure such as wind, thermal, hydro, and electricity, and IT hardware and software equipment, and has computing power, carrying capacity, and storage capacity, including general data centers, intelligent computing centers, supercomputing centers, etc.

[0038] The "supercomputing center" mentioned in this invention refers to a supercomputing data center, which is a data center based on supercomputers or large-scale computing clusters. It can provide large-scale computing, storage and network services and is widely used in aerospace, defense, oil exploration, climate modeling and genome sequencing and other application scenarios.

[0039] The “computing resources” mentioned in this invention refer to the technologies and facilities required for the development of the digital society that have the ability to compute, transmit, store and apply information, including but not limited to computing resources such as CPUs and GPUs, network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, and supporting and guaranteeing resources such as wind, fire, water and electricity.

[0040] The "model" mentioned in this invention includes, but is not limited to, "large language model" and "multimodal large model".

[0041] The "large language model" mentioned in this invention refers to a large-scale language model (LLM), which is a language model with a large number of parameters. It is designed to understand and generate human language. It is trained with a large amount of text data and can perform a wide range of tasks, including text summarization, translation, and sentiment analysis.

[0042] The “Multimodal Large Models” mentioned in this invention refer to models that combine multimodal information such as text, images, videos, and audio for training, including but not limited to multimodal large language models.

[0043] The “Token” mentioned in this invention refers to the basic unit of measurement when a large language model or multimodal large model processes text, code, image descriptions, document content, or other serialized inputs and outputs.

[0044] It is important to emphasize that different models use different tokenizers. The same text may correspond to different numbers of tokens in different models.

[0045] It is important to emphasize that vector filtering is a core capability supporting AI applications (such as intelligent retrieval, recommendation systems, weather analysis, and remote sensing image recognition) within intelligent computing cloud platforms. Accelerating document indexing is crucial in this process, specifically in ways that include, but are not limited to, the following: 1. Solve the performance bottleneck of "high-dimensional vector + multi-condition filtering".

[0046] Intelligent computing cloud platforms often need to process billions of vector data, and user queries often contain multiple scalar constraints (such as time, region, category, confidence threshold, etc.).

[0047] Without efficient indexes: the system needs to perform a scalar conditional scan on each vector in the entire database, resulting in extremely high latency; Introducing document-level full-text indexes (such as scalar indexes based on inverted or columnar storage): can quickly filter out candidate subsets that meet the scalar conditions (such as reducing from 1 billion to 100,000), and then perform vector ANN retrieval on this small set → query speed is improved by tens to hundreds of times.

[0048] This makes "conditional vector search" go from "unavailable" to "available in real time".

[0049] 2. Reduce computing and storage resource consumption and improve the cost-effectiveness of cloud platforms.

[0050] Vector ANN retrieval (such as HNSW and IVF) is inherently computationally intensive and memory-intensive. Without pre-filtering, a large number of irrelevant vectors are sent to the vector index, wasting GPU / CPU computing power and increasing memory bandwidth pressure. Document indexing acceleration can pre-prune the data stream at the CPU or storage layer, sending only highly relevant candidates to the vector engine.

[0051] Importance: Reduced resource consumption per query, support for higher concurrency on the same hardware, and significantly improved service density and ROI per unit computing power of the cloud platform.

[0052] 3. Support service quality under multi-tenancy and heterogeneous load conditions.

[0053] Intelligent computing cloud platforms typically serve hundreds or thousands of tenants with vastly different query patterns. For example, tenant A might search for "high temperature images of Beijing in the last 7 days" (strong time and region filtering); while tenant B might search for "similar trajectories of all typhoon paths" (weak scalar conditions, strong vector matching).

[0054] The document indexing acceleration scheme can be dynamically enabled / disabled (e.g., based on current load, user priority, and condition complexity), enabling high-value queries to prioritize the indexing acceleration path; at the same time, it automatically degrades when resources are scarce to ensure overall stability.

[0055] It enables refined resource scheduling, which is the foundation for the commercial operation of intelligent computing cloud platforms.

[0056] 4. Meet industry compliance and data governance requirements.

[0057] For example, in sensitive fields such as meteorology, finance, and healthcare, users can only access data within a specific region / time / permission range. Forced filtering must be performed before vector retrieval (rather than post-processing filtering) to prevent information leakage.

[0058] Please see Figure 1 , Figure 1 This is a flowchart of a method for generating tokens using computing power acceleration provided by an intelligent computing cloud platform, as described in this invention. Figure 1 As shown, the method includes: Step S1: Divide the acquired target data set into multiple data segments, generate vector parameters and scalar parameters based on the data segments, update the vector parameters and scalar parameters to the vector database of the intelligent computing cloud platform, and build a vector index based on the vector parameters and a full-text index of the data segments based on the scalar parameters.

[0059] It should be noted that the "target data set" refers to a set of raw business data in the intelligent computing cloud platform that is to be processed, vectorized, and prepared for storage in the vector database. It is the data source for the entire vector retrieval system's construction and query service, and its content and structure directly determine the effectiveness of the subsequently generated vector parameters, scalar parameters, and indexes.

[0060] The target dataset can come from various sources, and no restrictions are imposed here. In one possible implementation, the target dataset can come from multi-source heterogeneous systems, such as: meteorological satellite remote sensing images, numerical weather prediction model outputs, text logs, PDF reports, audio and video metadata, etc.

[0061] In this invention, the target data set can support multiple writing modes such as batch import, streaming access, and real-time writing.

[0062] In this invention, dividing the data into multiple segments supports horizontal scaling, allowing a single dataset to be stored across hundreds of nodes. This also improves parallel processing efficiency, enabling subsequent vectorization and index construction to be performed in a distributed manner.

[0063] Dividing the target dataset into multiple data segments can be done according to a pre-designed sharding strategy, which can be modified according to actual design needs. For example, a sharding strategy might look like this: Segment by time: such as one segment per day / month.

[0064] Spatial segmentation: such as latitude and longitude grid division.

[0065] Tenant-based sharding: Achieves physical isolation and ensures the security of multi-tenant communities.

[0066] In the high-concurrency, large-scale, multi-tenant, and resource-constrained environment of intelligent computing cloud platforms, dividing the target data set into multiple data segments improves the parallel efficiency of index building and querying. At the same time, it can optimize resource scheduling and load balancing, which not only improves system performance but also supports the platform's scalability, stability, and cost-effectiveness.

[0067] In this invention, a pre-trained AI model can be invoked to extract vector and scalar parameters from data fragments. The output is a fixed-dimensional floating-point array (e.g., 512-dimensional), representing the position of the data in the semantic space.

[0068] In this invention, an index can be built independently for each data segment, and the index type can be automatically selected according to the characteristics of the segment.

[0069] It should be noted that the intelligent computing cloud platform supports incremental index merging, avoiding full rebuilding.

[0070] The beneficial technical effects achieved in step S1 are as follows: By dividing the target dataset into multiple data segments and extracting vector parameters for semantic matching and scalar parameters for conditional filtering respectively, and simultaneously constructing vector indexes and full-text indexes, a deep integration of high-dimensional semantic retrieval and structured query capabilities is realized. This design can leverage full-text indexes to quickly respond to multi-dimensional scalar constraints (such as time, region, category, etc.), significantly improving the recall accuracy and execution efficiency of hybrid queries, and effectively alleviating the problem of a surge in first token latency caused by the traditional "recall first, filter later" model.

[0071] Step S2: In response to receiving a data query request from a user, determine whether to enable the document indexing acceleration function.

[0072] In this invention, document indexing acceleration does not bring performance benefits in all query scenarios. Blindly enabling it may lead to wasted system resources, increased latency, or even service degradation. Therefore, upon receiving a data query request from a user, this function is only enabled when the user requests acceleration and the acceleration benefits are significant and the cost is controllable. In other cases, it is proactively avoided or reverted to the basic search path.

[0073] In this invention, under conditions of limited system resources or high concurrency pressure, such as during peak periods of an intelligent computing cloud platform, CPU, memory, or storage I / O may be close to saturation. If an index is enabled for a low-priority query at this time, it may crowd out the resources of high-priority tasks, triggering an avalanche effect.

[0074] In another possible scenario, if the target data set is small (e.g., less than 10,000 records), the time taken by a full table scan is negligible, and it can be considered that the conditions for enabling document indexing acceleration are not met.

[0075] Through the aforementioned refined activation judgment mechanism, this solution avoids the drawbacks of traditional "all on" or "all off" strategies, and achieves maximum resource efficiency, guaranteed service quality, and reliable results.

[0076] Enabling document indexing acceleration becomes a cost-benefit-balancing intelligent scheduling decision. This solution, through multi-dimensional context awareness, ensures that acceleration capabilities are used effectively, truly achieving a smart vector retrieval service that balances high performance and high reliability.

[0077] The beneficial technical effects achieved by step S2 are: dynamically identifying and filtering out low-value or high-cost acceleration requests, reducing unnecessary resource consumption at the source, effectively avoiding overload of the intelligent computing cloud platform caused by the abuse of acceleration functions, ensuring the stable operation of the system under peak pressure, and at the same time, prioritizing the availability of core businesses when resources are limited.

[0078] Step S3: If it is determined that the document indexing acceleration function is enabled, extract scalar filter conditions and vector filter conditions from the data query request.

[0079] It should be noted that a vector is an ordered set of floating-point numbers used to quantitatively characterize the features of an entity from multiple dimensions. Each value corresponds to a feature dimension, which together constitute the mathematical representation of the entity in the feature space. For example, a three-dimensional vector [1.1, 2.3, 4.2] represents the quantified feature values ​​of the entity in three specific dimensions.

[0080] Scalar metadata refers to the structured additional attribute information associated with a vector, typically indexable business fields. For example, if a vector is used to represent a person's semantic or behavioral characteristics, its corresponding scalar may include structured data such as height, weight, age, education level, and city of residence.

[0081] In this invention, a complete data record consists of both a vector part and a scalar part, neither of which can be omitted. The vector carries high-dimensional semantic features, while the scalar provides filterable contextual attributes. The combination of the two supports efficient hybrid queries.

[0082] In intelligent computing cloud platforms, user queries are typically submitted in structured or semi-structured forms (such as JSON, GraphQL, SQL extended syntax, or REST API parameters), which may simultaneously include semantic similarity requirements (vector part) and business constraint rules (scalar part).

[0083] The beneficial technical effect achieved by step S3 is that by extracting scalar and vector filtering conditions from the data query request, a data foundation is provided for subsequent vector filtering.

[0084] Step S4: Based on the scalar filtering conditions, perform full-text search in the full-text index of the vector database to obtain the first matching data set. The vector database stores multiple vector data, and each vector data includes vector parameters and scalar parameters.

[0085] Existing technologies often employ a "full table scan + row-by-row filtering" approach, which involves reading complete records (including vectors and scalars) one by one from vector storage, and then determining whether the scalar conditions are met at the application layer. This process requires multiple random disk reads or object storage GET requests, resulting in high-frequency network transmission of large amounts of irrelevant data (especially the vector portion, which typically accounts for over 90% of the volume), leading to I / O bottlenecks and wasted bandwidth.

[0086] The advantages of this solution are: Full-text indexing stores only lightweight scalar metadata and ID mappings. A single index query can return a complete list of matching IDs (e.g., a bitmap or compressed DocID set), and subsequent queries only load the corresponding vectors as needed. This reduces I / O operations from multiple times to a single operation, decreasing network transmission volume by 10 to 100 times and increasing single-query throughput by tens of times. It significantly compresses the processing chain from request reception to the generation of the first valid response. This mechanism directly optimizes the first token latency, greatly improving its response efficiency and enhancing the smoothness of the interactive experience.

[0087] In this invention, scalar conditions can be used to quickly narrow down the range of candidate data, providing a high-quality, small-scale input set for subsequent vector similarity retrieval.

[0088] Under the distributed architecture of the intelligent computing cloud platform, the "full-text index" can be a highly efficient filtering index built for structured / semi-structured scalar fields. It can perform special optimizations on documents, provide accurate queries, and store the index in a distributed manner according to data shards or tenant isolation units, supporting parallel retrieval.

[0089] The intelligent computing cloud platform can parse the complex scalar conditions submitted by users, decompose the conditions into multiple atomic query conditions through the query engine, and perform set operations such as intersection and union at the index layer.

[0090] For example, if a user queries: "grid data of temperatures in East China that are abnormally higher than 35℃ in July 2024", the full-text index will locate: set A corresponding to the time index for July, set B corresponding to the regional index, and set C corresponding to the temperature index. Finally, the first data set = A∩B∩C.

[0091] In one possible implementation, the full-text index does not store complete data, but rather lightweight metadata identifiers (such as primary key IDs, row keys, or globally unique DocIDs). After retrieval, the system uses this set of IDs to quickly locate the corresponding vector data block (usually stored in columnar storage or a vector-specific engine).

[0092] The beneficial technical effect achieved by step S4 is that by implementing filtering logic in advance and atomic delivery of result sets through full-text indexing, the performance bottleneck caused by redundant data handling and inefficient computing paths in existing technologies is fundamentally avoided.

[0093] Step S5: Based on the vector filtering conditions, perform vector retrieval on the vector index corresponding to the first data set to obtain the matching second data set, and return the second data set to the user.

[0094] After obtaining the first data set, perform vector retrieval on the subset corresponding to the first data set, and match to obtain the second data set.

[0095] The beneficial technical effect achieved by step S5 is as follows: Based on the first data set provided in the previous steps, the second data set is generated by further filtering through vector filtering conditions. This design enables the intelligent computing cloud platform to achieve industrial-grade response speed and resource efficiency while ensuring high recall quality, truly supporting the stringent requirements of key fields such as meteorology, remote sensing, and urban governance for AI retrieval services.

[0096] In this invention, step S1 involves dividing the acquired target data set into multiple data segments, generating vector parameters and scalar parameters based on the data segments, updating the vector parameters and scalar parameters in the vector database of the intelligent computing cloud platform, establishing a vector index based on the vector parameters, and establishing a full-text index for the data segments based on the scalar parameters; step S2 involves responding to a data query request sent by a user by determining whether to enable the document index acceleration function; step S3 involves extracting scalar filtering conditions and vector filtering conditions from the data query request if the document index acceleration function is enabled; step S4 involves performing a full-text search in the full-text index of the vector database based on the scalar filtering conditions to obtain a matching first data set, wherein the vector database stores multiple vector data, each vector data including vector parameters and scalar parameters; step S5 involves performing a vector search on the vector index corresponding to the first data set based on the vector filtering conditions to obtain a matching second data set, and returning the second data set to the user. In this way, through designs such as dynamic decision-making, conditional separation, phased utilization of indexes, and full-text retrieval, the computing power of the intelligent computing cloud platform is focused on high-value candidate sets, avoiding the consumption of floating-point computing resources on invalid data. While ensuring query accuracy, it achieves a double leap in computing power utilization and service efficiency, overcoming the performance bottlenecks caused by redundant data transportation and inefficient computing paths in existing technologies.

[0097] In one embodiment, such as Figure 2 As shown, step S4 includes: Step S4.1: In the vector database, using the pre-built full-text inverted index and semantic vector index, keyword matching and semantic similarity calculation are performed simultaneously on the full-text index.

[0098] In this invention, performing keyword matching and semantic similarity calculation on the full-text index simultaneously is the core operation for achieving hybrid retrieval. The key lies in the deep integration of traditional information retrieval and modern semantic understanding capabilities.

[0099] It should be noted that the intelligent computing cloud platform pre-builds two complementary index structures in the vector database: a full-text inverted index and a semantic vector index (such as HNSW, IVF-PQ, and other approximate nearest neighbor indexes). When a full-text query request is received, the system does not rely solely on literal keyword matching, but triggers two retrieval pathways in parallel: In the full-text inverted index path, the query text is processed by word segmentation, stop word removal, and stemming to quickly recall data fragments containing specified keywords or satisfying Boolean logic conditions, ensuring high-precision literal matching results.

[0100] In the semantic vector indexing pathway, the same query text is converted into a dense vector through an embedding model (such as BERT, Sentence-BERT, or domain fine-tuning encoder), and semantically similar data fragments are retrieved in the vector space, effectively capturing non-explicit matching relationships such as synonyms and contextual relevance.

[0101] In another possible implementation, the intelligent computing cloud platform can also generate a unified relevance ranking by weighted fusion of the two results (such as configurable weights). This joint retrieval mechanism retains the accuracy and interpretability of traditional full-text search while introducing the generalization and robustness of semantic search, making it particularly suitable for complex scenarios where user queries are vague, terminologically inconsistent, or require cross-language / cross-modal understanding.

[0102] It should be noted that since both types of indexes are built on the same data fragment in step S1, they are naturally aligned, avoiding the alignment overhead and consistency risks brought about by the fusion of multi-source heterogeneous data, thereby realizing an integrated retrieval service with low latency, high recall and high accuracy on the intelligent computing cloud platform.

[0103] Step S4.2: Generate a complete first data set based on the matching results, and return the first data set in a single response.

[0104] In this invention, the return method of the first data set does not employ pagination, streaming, or multiple polling by the client. Instead, the server assembles, serializes, and compresses the results all at once, and atomically returns all results to the caller in a single network response. This design brings multiple technical advantages: Reduce end-to-end latency: Avoid the round-trip time overhead caused by multiple rounds of communication, and significantly improve the user experience, especially in high-concurrency or cross-regional scenarios; Ensure result consistency: Generate result sets at the snapshot isolation level to prevent pagination results from jumping or repeating due to data updates; Simplify client-side logic: Upper-layer applications do not need to implement pagination state management, merging logic, or timeout retry mechanisms, improving development efficiency and system reliability; To meet the token generation requirements of large-scale models included in the intelligent computing cloud platform: provide complete contextual input for subsequent search enhancement generation and other scenarios, and avoid the degradation of generation quality due to result truncation.

[0105] In one possible implementation, the intelligent computing cloud platform can also dynamically adjust the maximum number of returned records based on the current resource status (such as memory space and response size limits), but always ensure that "either all records are returned or an explicit rejection is made", thus eliminating the ambiguous state of partial success and partial loss.

[0106] In this invention, step S1 includes: Step S1.1: Determine the scalar feature data and vector feature data of the data segment.

[0107] In this invention, scalar feature data refers to structured, indexable attribute information with clear business meaning, which is typically used for filtering, grouping, or sorting.

[0108] Scalar feature data can be obtained through document metadata, such as collection time, geographical location (latitude and longitude), data source (e.g., "satellite X"), resolution, document format, etc.

[0109] The intelligent computing cloud platform can declare which fields are used as scalars when data is accessed.

[0110] Automatic type inference: The intelligent computing cloud platform parses data fragments to identify candidate scalar fields.

[0111] The intelligent computing cloud platform also allows users to customize "scalar feature extraction rules" through the UI or API, adapting them to vertical domain terminology.

[0112] Vector feature data refers to unstructured / high-dimensional content that carries semantic or perceptual information and needs to be converted into dense vectors through AI models.

[0113] The intelligent computing cloud platform can determine whether data fragments need to be vectorized based on their document type. It can also trigger vector feature extraction processes based on task configurations (such as "This dataset is used for similar image retrieval").

[0114] The intelligent computing cloud platform supports multi-granularity vectorization, such as whole-image vectors, region block vectors, and keyframe vectors.

[0115] Step S1.2: Generate scalar parameters based on scalar feature data, and generate vector parameters based on vector feature data.

[0116] In this invention, the identified scalar feature data can be cleaned, normalized, and type-fixed to generate scalar parameters. For example, the time field can be unified into ISO 8601 format (2024-07-15T08:30:00Z), and the geographical region can be mapped to the standard administrative division code (such as "Guangdong" → CN-GD).

[0117] The vector parameters can be obtained by calling a feature extraction model to generate tokens from the vector feature data. For example, it can be done as follows: The image can be processed to output a 512-dimensional float32 vector; Text can be processed to output multidimensional sentence vectors; Timing can be processed to output a fixed-dimensional timing vector.

[0118] The beneficial technical effects achieved by step S7.2 are: by generating vector parameters and scalar parameters, not only is the data representation generated, but also, through structured, standardized and automated processing, a data foundation is provided for subsequent efficient hybrid queries.

[0119] In one embodiment, such as Figure 3 As shown, step S2 includes: Step S2.1: Obtain the computing resources currently used by the document indexing acceleration function of the intelligent computing center cloud platform; Step S2.2: In response to the computing power resource value being greater than the resource threshold, the document indexing acceleration function is disabled.

[0120] It should be noted that this mechanism constitutes an adaptive admission control strategy based on global load status. In a cloud environment with limited resources, it prevents system overload caused by "abuse of acceleration functions", thereby ensuring the overall stability of the intelligent computing cloud platform and the service quality of critical businesses.

[0121] In this invention, a user who has enabled document indexing acceleration refers to an active session currently executing a hybrid query that relies on full-text indexing. The count of active accelerated users can be maintained in real time through a distributed coordination service or an in-memory state table (such as Redis).

[0122] The intelligent computing cloud platform can support grouped statistics by tenant, project, API key, or QoS level.

[0123] In one possible implementation, the resource threshold can be a pre-designed fixed value, such as limiting the number of accelerated users to 100. Alternatively, the resource threshold can be a non-fixed value, i.e., a configurable and dynamically adjustable system parameter. For example, the resource threshold can be calculated based on the current GPU cluster utilization and / or remaining video memory and / or network bandwidth load of the intelligent computing cloud platform. This is not a simple "one-size-fits-all" approach, but rather supports differentiated degradation.

[0124] By setting resource thresholds to control the document indexing acceleration function, resource avalanche can be prevented, ensuring system stability. Active rate limiting keeps the acceleration function within the system's capacity.

[0125] Please see Figure 4 , Figure 4 This is a structural diagram of a device for generating tokens using computing power acceleration provided by the present invention, such as... Figure 4 As shown, the device 400 for generating tokens through computing power acceleration in an intelligent computing cloud platform includes: Module 401 is used to divide the acquired target data set into multiple data segments, generate vector parameters and scalar parameters based on the data segments, update the vector parameters and scalar parameters to the vector database of the intelligent computing cloud platform, and build a vector index based on the vector parameters and a full-text index of the data segments based on the scalar parameters.

[0126] The receiving module 402 is used to respond to a data query request sent by a user and determine whether to enable the document indexing acceleration function.

[0127] The determination module 403 is used to extract scalar filter conditions and vector filter conditions from the data query request if it is determined that the document indexing acceleration function is enabled.

[0128] The first retrieval module 404 is used to perform full-text retrieval in the full-text index of the vector database based on scalar filtering conditions to obtain a first set of matching data. The vector database stores multiple vector data, each of which includes vector parameters and scalar parameters.

[0129] The second retrieval module 405 is used to perform vector retrieval on the vector index corresponding to the first data set based on vector filtering conditions, obtain the matching second data set, and return the second data set to the user.

[0130] In one embodiment, the first retrieval module 404 is further configured to: In the vector database, using a pre-built full-text inverted index and semantic vector index, keyword matching and semantic similarity calculation are performed simultaneously on the full-text index; A complete first data set is generated based on the matching results, and the first data set is returned in a single response.

[0131] In one embodiment, the establishment module 401 is further configured to: Determine the scalar and vector features of the data segment; Scalar parameters are generated based on scalar feature data, and vector parameters are generated based on vector feature data.

[0132] In one embodiment, the receiving module 402 includes: The first acquisition unit is used to acquire the computing power resource value of the intelligent computing cloud platform that currently has the document indexing acceleration function enabled; The disabling unit is used to disable the document indexing acceleration function in response to the computing power resource value exceeding a resource threshold. The device for generating tokens via computing power acceleration in the intelligent computing cloud platform provided by this invention can implement the various processes of the embodiments of the above-described method for generating tokens via computing power acceleration in the intelligent computing cloud platform. The technical features are one-to-one and can achieve the same technical effect; therefore, to avoid repetition, they will not be described again here.

[0133] It should be noted that the device for generating tokens through computing power acceleration in the intelligent computing cloud platform of this invention can be a device, or it can be a component, integrated circuit, or chip in an electronic device.

[0134] The present invention also provides an electronic device, see [link to relevant documentation]. Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. The electronic device includes a memory 501, a processor 502, and a program or instructions stored in the memory 501 that run on the processor 502. When the program or instructions are executed by the processor 502, they can achieve the following: Figure 1 The corresponding intelligent computing cloud platform accelerates the generation of tokens through computing power, and achieves the same beneficial effects in any of the steps in the method embodiment, which will not be elaborated here.

[0135] The processor 502 can be a CPU, ASIC, FPGA, or GPU.

[0136] Those skilled in the art will understand that all or part of the steps of the above-described method embodiment for generating tokens through computing power acceleration can be implemented by hardware related to program instructions, and the program can be stored in a readable medium.

[0137] The present invention also provides a readable storage medium on which a computer program is stored, and which, when executed by a processor, can perform the above-described functions. Figure 1 The corresponding intelligent computing cloud platform accelerates token generation through computing power, and any step in the method embodiment can achieve the same technical effect. To avoid repetition, it will not be described again here. The storage medium mentioned is such as read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0138] The present invention also provides a computer program product, including computer instructions that, when executed by a processor, implement the above-described... Figure 1 The corresponding intelligent computing cloud platform accelerates the generation of tokens through computing power, and achieves the same technical effect. To avoid repetition, it will not be described in detail here.

[0139] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or second terminal device, etc.) to execute the methods of the various embodiments of the present invention.

[0140] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.

Claims

1. A method for generating tokens using computing power acceleration on an intelligent computing cloud platform, characterized in that, include: Step S1: Divide the acquired target data set into multiple data segments, generate vector parameters and scalar parameters based on the data segments, update the vector parameters and scalar parameters to the vector database of the intelligent computing cloud platform, establish a vector index based on the vector parameters, and establish a full-text index of the data segments based on the scalar parameters. Step S2: In response to receiving a data query request from a user, determine whether to enable the document indexing acceleration function; Step S3: If it is determined that the document indexing acceleration function is enabled, then extract scalar filtering conditions and vector filtering conditions from the data query request; Step S4: Based on the scalar filtering conditions, perform full-text search in the full-text index of the vector database to obtain a matching first data set, wherein the vector database stores multiple vector data, and each vector data includes vector parameters and scalar parameters; Step S5: Based on the vector filtering conditions, perform vector retrieval on the vector index corresponding to the first data set to obtain a matching second data set, and return the second data set to the user.

2. The method according to claim 1, characterized in that, Step S4 includes: Step S4.1: In the vector database, using the pre-built full-text inverted index and semantic vector index, keyword matching and semantic similarity calculation are performed simultaneously on the full-text index; Step S4.2: Generate a complete first data set based on the matching results, and return the first data set in a single response.

3. The method according to claim 1, characterized in that, Step S1 includes: Step S1.1: Determine the scalar feature data and vector feature data of the data segment; Step S1.2: Generate the scalar parameters based on the scalar feature data, and generate the vector parameters based on the vector feature data.

4. The method according to claim 1, characterized in that, Step S2 includes: Step S2.1: Obtain the computing power resource value currently occupied by the document indexing acceleration function currently enabled on the intelligent computing cloud platform; Step S2.2: In response to the computing power resource value being greater than the resource threshold, the document indexing acceleration function is disabled.

5. The method according to claim 4, characterized in that, Prior to step S2.2, the step of determining the resource threshold includes: Step 2.3: Calculate the resource threshold based on the current GPU cluster utilization and / or remaining video memory and / or network bandwidth load of the intelligent computing cloud platform.

6. A device for generating tokens using computing power acceleration on an intelligent computing cloud platform, characterized in that, include: A module is established to divide the acquired target data set into multiple data segments, generate vector parameters and scalar parameters based on the data segments, update the vector parameters and scalar parameters to the vector database of the intelligent computing cloud platform, establish a vector index based on the vector parameters, and establish a full-text index of the data segments based on the scalar parameters. The receiving module is used to respond to data query requests sent by users and determine whether to enable the document indexing acceleration function; The determination module is used to extract scalar filtering conditions and vector filtering conditions from the data query request if it is determined that the document indexing acceleration function is enabled. The first retrieval module is used to perform a full-text search in the full-text index of the vector database based on the scalar filtering conditions to obtain a matching first data set, wherein the vector database stores multiple vector data, and each vector data includes vector parameters and scalar parameters; The second retrieval module is used to perform vector retrieval on the vector index corresponding to the first data set based on the vector filtering conditions, obtain a matching second data set, and return the second data set to the user.

7. An electronic device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the method for generating tokens by computing power acceleration of the intelligent computing cloud platform as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for generating tokens by computing power acceleration of the intelligent computing cloud platform as described in any one of claims 1 to 5.

9. A computer program product, characterized in that, It includes computer instructions, which, when executed by a processor, implement the steps of the method for generating tokens by computing power acceleration of the intelligent computing cloud platform as described in any one of claims 1 to 5.

Images