A full-stack intelligent deployment method and system based on digital genes and LLMs

Abstract

A full-stack intelligent deployment method based on digital genes and LLM (Limited Management Model) belongs to the field of artificial intelligence and automated operation and maintenance technology. Its key feature is that it achieves deep collaborative optimization of software and hardware by unifying the full-stack description and management from software intent to hardware circuitry through digital genes, breaking through the limitations of traditional hierarchical management. Utilizing the complete state and strategies of the digital gene packaging system, high-performance, high-fidelity system replication can be achieved in heterogeneous environments, greatly improving the efficiency and consistency of large-scale deployment. Through the life system compiler, hardware resources are allocated, creating isolated and secure hardware resources, providing practical testing of software and hardware security collaboration, avoiding potential security risks in virtual environments, and ensuring reliability and security during the deployment phase. Through a closed loop of "monitoring-evaluation-decision-execution," the system can autonomously optimize according to the runtime environment, moving from "automation" to "autonomy," greatly reducing the cost of manual intervention.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and automated operation and maintenance technology, and in particular relates to a full-stack intelligent deployment method and system based on digital genes and LLM. Background Technology

[0002] Large Language Models (LLMs), as a deep learning-based artificial intelligence technique, have demonstrated powerful capabilities in natural language understanding and generation, code generation, and reasoning. In recent years, their applications have expanded from plain text dialogues to the processing of complex system tasks, including code writing, logical planning, and decision support. LLMs can parse high-level user intents and generate detailed step plans required to achieve those intents, providing a new technological path for building highly intelligent computer systems.

[0003] Currently, mainstream computer system management (such as deployment, configuration, monitoring, and optimization) typically employs a layered management architecture, with information silos existing between the software and hardware layers, resulting in a lack of a holistic perspective in optimization decisions. Traditional automated operation and maintenance systems largely rely on preset rules for decision-making, making it difficult to adapt to dynamically changing computing environments. Furthermore, the lack of effective mechanisms to ensure system security poses decision-making risks. Existing AI methods are limited to specific levels, lacking the ability to make global intelligent decisions across both software and hardware. Existing replication technologies (such as container images) cannot effectively encapsulate the optimal system state in heterogeneous environments, making it difficult to guarantee performance consistency.

[0004] Furthermore, while there have been preliminary explorations of the concept of "digital genes" in existing technologies, a systematic, unified, full-stack description method has not been formed, nor has digital genes been effectively combined with intelligent decision-making models. Summary of the Invention

[0005] The present invention aims to solve the above problems and provide a full-stack intelligent deployment method and system based on digital genes and LLM.

[0006] In a first aspect, the present invention provides a full-stack intelligent deployment method based on digital genes and LLM, comprising the following steps:

[0007] Step 1: Use a hierarchical coding mechanism to structure the hardware configuration, software environment, and strategy parameters of the computing system to generate the digital gene code of the entire computing system stack.

[0008] Step 2: Using the economic form of digital gene coding and computing systems as input, drive the software and hardware co-evolution LLM to generate multiple software and hardware deployment strategies;

[0009] Step 3: Use multiple sets of software and hardware deployment strategies as a decision reward model, score each unit strategy in the strategy, and finally merge the strategies with higher scores from multiple sets of strategies into the final strategy.

[0010] Step 4: Input the final policy generated by LLM into the Adaptive Life System Compiler, which outputs software and hardware executable scripts based on the policy content and hardware compilation specifications.

[0011] Step 5: Based on the final strategy generated by the LLM, the life system compiler allocates hardware resources on the hardware side;

[0012] Step 6: On the allocated hardware resources, perform a simulated security sandbox test on the aforementioned executable script to enable direct deployment and computational calls from the software side to the required hardware.

[0013] Step 7: Based on the sandbox test results, if an anomaly occurs during the test, repeat steps 2 and 3 until the verification is successful.

[0014] Step 8: Deploy the verified script to the target system.

[0015] Furthermore, the full-stack intelligent deployment method based on digital genes and LLM described in this invention collects the operating data of the computing system in real time. When an update condition is triggered, the digital genes of the computing system are updated, and steps 2 to 8 are repeated until the system reaches a new stable state. By collecting hardware and software operating data in real time, the system can proactively respond to unforeseen load changes, performance bottlenecks, and security threats, maintaining its optimal state through dynamic evolution, and possessing the resilience and adaptability of a biological system.

[0016] Furthermore, the full-stack intelligent deployment method based on digital genes and LLM described in this invention, after the system stabilizes, stores the latest digital genes and corresponding strategy codes in a digital gene library to support rapid cloning and deployment of subsequent systems. The establishment of the data gene library supports rapid cloning, enabling efficient deployment of software and hardware deployment strategies and reducing reliance on manual intervention.

[0017] Furthermore, the full-stack intelligent deployment method based on digital genes and LLM described in this invention involves encoding and compressing hardware, software, strategies, and metadata content into unique digital gene codes. These unique digital gene codes not only compress data but also ensure the security and consistency of the digital genes.

[0018] The digital gene comprises the following components:

[0019] The hardware gene layer describes the characteristics of hardware resources, including processor architecture, memory configuration, storage type, and network topology; for example, the number of CPU cores, memory capacity, FPGA chip type, and GPU model.

[0020] The software gene layer describes software-level characteristics, including operating system version, middleware configuration, and application deployment strategy; for example, Linux kernel version, Kubernetes configuration, Docker image parameters, etc.

[0021] The strategy gene layer describes the characteristics at the decision-making level, including system optimization strategies, security rules, and performance goals; for example, load balancing strategies, security protection levels, and resource allocation priorities.

[0022] The metadata layer is used to describe auxiliary information, including the version information, evolutionary history, and applicable environment of the digital gene; such as gene version number, evolution timestamp, and list of compatible hardware.

[0023] Furthermore, the full-stack intelligent deployment method based on digital genes and LLM described in this invention sets up multiple system economic forms according to task characteristics, including an economy-first model, a performance-first model, and a comprehensive model. The economic form of the computing system defines the resource utilization of the final computing system, ensuring the economic controllability of the computing system.

[0024] Secondly, the present invention provides a full-stack intelligent deployment system based on digital genes and LLM, including a digital gene encoding module, a deployment strategy generation module, an adaptive life system compilation module, and a deployment module;

[0025] The digital gene encoding module is used to provide a structured description of the hardware configuration, software environment, and strategy parameters of the computing system through a hierarchical encoding mechanism, thereby generating the digital gene encoding of the entire computing system stack.

[0026] The deployment strategy generation module is used to drive the software and hardware co-evolution LLM to generate multiple software and hardware deployment strategies by taking the economic form of digital gene coding and computing system as input; the multiple software and hardware deployment strategies are used as a decision reward model to score each unit strategy in the strategy; and finally, the strategy with the higher score among the multiple strategies is merged into the final strategy.

[0027] The adaptive life system compilation module is used to output software and hardware executable scripts based on the final policy content and hardware compilation specifications generated by LLM; allocate hardware resources according to the final policy generated by LLM; perform simulated security sandbox tests on the executable scripts; and realize the direct deployment and computation call of the required hardware by the software. As a security barrier, the life system compilation module separates the cognitive ability of LLM from the deterministic execution of the underlying system, which not only utilizes the intelligence of AI, but also ensures the reliability and security of the system.

[0028] The deployment module is used to deploy scripts that have passed the sandbox test to the target system; after the system is stable, the latest digital genes and corresponding strategy codes are stored in the digital gene library.

[0029] Furthermore, the deployment strategy generation module of the full-stack intelligent deployment system based on digital genes and LLM described in this invention includes the following components:

[0030] The system status analysis module is used to integrate hardware sensor data (such as CPU utilization, memory usage, FPGA temperature) and software logs (such as container performance, security events) into an input format that LLM can understand.

[0031] The optimization strategy generation module is used to generate cross-level optimization strategies based on metadata tags of digital genes (such as security policies and performance targets) and system status analysis results; for example, adjusting FPGA resource configuration, optimizing container scheduling rules, and updating BIOS parameters.

[0032] Execution plan formulation module: used to translate optimization strategies into executable API calls or configuration file changes and record them as a new version of the digital gene;

[0033] Decision reward model: This model scores and merges multiple policy options generated by LLM to produce a final policy output. It scores these options and merges them to create a more reliable policy solution, avoiding the limitations of a single policy.

[0034] Furthermore, the adaptive life system compilation module of the full-stack intelligent deployment system based on digital genes and LLM described in this invention includes the following components:

[0035] The policy verification module is used to verify the syntactic correctness and static constraint satisfaction of the policy using formal methods (such as the B method and Coq).

[0036] The execution sandbox module is used to create isolated hardware and software execution environments based on trusted execution environments (such as Intel SGX and ARM TrustZone), simulate the effects of policy execution, and prevent malicious decisions from affecting the production system.

[0037] The security metadata tagging module is used to embed security attributes (such as prohibiting modification of UEFI boot order) in digital genes, and the compiler enforces checks on whether the policy violates these metadata constraints;

[0038] The version management module is used to record the version change history of digital genes based on blockchain technology, ensuring that each evolution is rollbackable and the strategy is verifiable.

[0039] Thirdly, the present invention provides a full-stack intelligent deployment device based on digital genes and LLM, comprising a processor and a memory electrically connected to each other; the memory is used to store a computer program; when the processor executes the aforementioned computer program, it can realize the full-stack intelligent deployment method based on digital genes and LLM as described in the first aspect.

[0040] Fourthly, the present invention provides a computer-readable storage medium storing a computer program; when the computer program is executed, it can realize the full-stack intelligent deployment method based on digital genes and LLM as described in the first aspect.

[0041] The full-stack intelligent deployment method and system based on digital genes and LLM described in this invention unifies the full-stack description and management from software intent to hardware circuits through digital genes, achieving deep collaborative optimization of software and hardware and breaking through the limitations of traditional hierarchical management. Utilizing the complete state and strategies of the digital gene packaging system, high-performance, high-fidelity system replication can be achieved in heterogeneous environments, greatly improving the efficiency and consistency of large-scale deployment. Through the life system compiler, hardware resources are allocated, creating isolated and secure hardware resources, providing practical testing of secure software-hardware collaboration, avoiding potential security risks in virtual environments, and ensuring reliability and security during the deployment phase. Through a closed loop of "monitoring-evaluation-decision-execution," the system can autonomously optimize according to the runtime environment, moving from "automation" to "autonomy," greatly reducing the cost of manual intervention. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the full-stack intelligent deployment method based on digital genes and LLM as described in an embodiment of the present invention. Detailed Implementation

[0043] The following detailed description of the full-stack intelligent deployment method and system based on digital genes and LLM according to the present invention is provided with reference to the accompanying drawings and embodiments.

[0044] Example 1

[0045] This embodiment discloses a full-stack intelligent deployment method based on digital genes and LLM, such as... Figure 1 As shown, it includes the following steps:

[0046] Step 1: The hardware configuration, software environment and strategy parameters of the computing system are described in a structured manner through a hierarchical coding mechanism to generate the digital gene code of the entire computing system stack.

[0047] In this embodiment, specifically after hardware assembly, the system automatically collects hardware parameters such as CPU model, GPU specifications, memory capacity, and storage type, while simultaneously recording software information such as operating system version, driver version, and middleware configuration. A JSON Schema structured storage method is used, dividing the system configuration into several dimensions: hardware layer, software layer, policy layer, and metadata layer. Each dimension contains necessary attributes and constraints. The hardware layer encoding includes device model, specifications, and performance indicators; the software layer includes system version, security patches, and system configuration; the policy layer encoding includes resource allocation policies, performance optimization policies, and security policies; and the metadata layer contains metadata related to the digital gene.

[0048] The final generated digital gene file includes the following components:

[0049] The hardware gene layer describes the characteristics of hardware resources, including processor architecture, memory configuration, storage type, and network topology; the software gene layer describes the characteristics of the software layer, including operating system version, middleware configuration, and application deployment strategy; the strategy gene layer describes the characteristics of the decision-making layer, including system optimization strategy, security rules, and performance goals; and the metadata layer describes auxiliary information, including version information, evolution history, and applicable environment of the digital gene.

[0050] The aforementioned hardware, software, strategies, and metadata are compressed into a unique digital gene encoding storage using the AES-256 encryption encoding format.

[0051] Step 2: Using the economic form of digital gene coding and computing systems as input, drive the software and hardware co-evolution LLM to generate multiple software and hardware deployment strategies.

[0052] In this embodiment, specifically, after receiving the economic form (economic priority model, performance priority model, comprehensive mode, etc.) of the digital gene and computing system as input, LLM first performs semantic parsing of the system configuration to identify the matching degree between the hardware configuration and task requirements. Then, based on a pre-trained domain knowledge base and historical optimization strategies, it evaluates the feasibility, performance improvement potential, and risk coefficient of the solution through multiple rounds of iteration. LLM integrates with external tools through the Model Context Protocol (MCP) to obtain real-time hardware performance data and task requirement analysis, generating a strategy report containing detailed descriptions of optimization strategies, expected performance improvement indicators, and potential risk assessments. The system sets up a strategy generation quality scoring mechanism to ensure the reliability and effectiveness of the output strategy.

[0053] Step 3: Use multiple sets of software and hardware deployment strategies as a decision reward model, score each unit strategy in the strategy, and finally merge the strategies with higher scores from multiple sets of strategies into the final strategy.

[0054] The decision reward model receives multiple policies generated by LLM, uses the reward model to score the solutions for each unit in the scheme, and selects the optimal policy for each unit. The optimal policies of each unit are then merged to form the final policy scheme.

[0055] Step 4: Input the final policy generated by LLM into the Adaptive Life System Compiler, which outputs software and hardware executable scripts based on the policy content and hardware compilation specifications.

[0056] In this embodiment, the life system compiler, after receiving the optimization strategy generated by the LLM, first performs syntax and semantic verification of the strategy and checks its compatibility with the hardware platform. Then, based on the hardware platform's compilation specifications (such as CUDA version, driver compatibility, system APIs, etc.), it converts the strategy into a specific executable script. In this embodiment, the compiler has a built-in hardware feature library, which can automatically adapt to the features of different hardware platforms. For the software configuration part, the compiler uses predefined templates and variable substitution mechanisms to generate configuration files; for the hardware configuration part, the compiler calls the hardware driver API to generate corresponding configuration instructions. The final output script includes detailed execution steps, parameter descriptions, error handling mechanisms, and version identifiers.

[0057] Step 5: Based on the final strategy generated by the LLM, the life system compiler allocates hardware resources on the hardware side;

[0058] Step 6: On the allocated hardware resources, perform a simulated security sandbox test on the aforementioned executable script.

[0059] In this embodiment, the life system compiler first performs static analysis on the script to check for potential security vulnerabilities and non-compliant configurations. Then, a portion of secure hardware resources are isolated on the hardware side, and the script is executed in a sandbox environment to monitor system resource usage, network access behavior, and the status of critical processes. Sandbox testing includes functional verification testing (verifying whether the script can correctly execute its intended functions), performance stress testing (testing system stability under high load and selecting key parameters that match the economic choices of the computing system), and security boundary testing (testing the script's behavior under abnormal input). The test results generate detailed reports, including execution logs, performance metrics, security risk assessments, and optimization suggestions.

[0060] Step 7: Based on the sandbox test results, if an anomaly occurs during the test, repeat steps 2 and 3 until the verification is successful.

[0061] In this embodiment, when an anomaly is detected during sandbox testing, the system automatically feeds back the anomaly information to the LLM, which then regenerates the optimization strategy based on the anomaly information. Simultaneously, the system records the anomaly type, occurrence conditions, and scope of impact for subsequent optimization strategy improvements. The process of repeatedly executing steps 2 and 3 employs incremental optimization; each iteration only makes local adjustments to the detected problem, avoiding full retries. In this embodiment, the system sets a maximum iteration limit (default 5 times) to prevent infinite loops. After each iteration, the system evaluates the degree of improvement in the optimization strategy to ensure that each iteration brings substantial improvement.

[0062] Step 8: Deploy the verified script to the target system.

[0063] The validated script is transmitted to the target system via a secure transport protocol (such as HTTPS or SFTP). In this embodiment, the deployment process employs atomic operations to ensure system availability during deployment. The deployment script includes version information, dependency checks, a rollback mechanism, and execution logs. Before deployment, the system performs an environment check to ensure the target system meets the prerequisites for script execution. After deployment, the system automatically executes validation tests to confirm that the script execution results are consistent with expectations and generates a deployment report.

[0064] In this embodiment of the disclosure, system performance metrics (such as CPU utilization, memory usage, network throughput, disk I / O, etc.) and task execution metrics (such as task completion time, error rate, resource consumption, etc.) are continuously monitored. The system presets update trigger conditions, including performance degradation exceeding a threshold (such as CPU utilization consistently exceeding 85%), error rate increase (such as task failure rate exceeding 5%), and changes in new task requirements.

[0065] When the triggering conditions are met, the system automatically generates a new digital gene and initiates the optimization process. Time series analysis and anomaly detection algorithms (such as LSTM neural networks) are used to ensure accurate judgment of update conditions. The system maintains service continuity during the update process by adopting a gradual update strategy to progressively apply optimization strategies and avoid drastic system fluctuations.

[0066] After the system is running stably, the digital gene repository will automatically encode and store the latest digital gene versions. The storage process includes digital signature verification, version number management (using semantic versioning, such as v1.2.3), and metadata recording (such as generation time, optimization goals, system configuration, performance improvement metrics, etc.). The digital gene repository adopts a distributed storage architecture (such as a distributed database based on the Raft consensus algorithm) to ensure high availability and data consistency. The system provides a RESTful API interface, supporting other systems to quickly retrieve and clone digital genes. The cloning process includes downloading, verifying, and applying digital genes, ensuring that the cloned system is completely consistent with the original system configuration. The system also supports version comparison and difference analysis functions for digital genes, facilitating the traceability and learning of the system optimization process.

[0067] Example 2

[0068] This embodiment discloses a full-stack intelligent deployment system based on digital genes and LLM, including a digital gene encoding module, a deployment strategy generation module, an adaptive life system compilation module, and a deployment module.

[0069] In this embodiment, the digital gene encoding module is used to structurally describe the hardware configuration, software environment, and strategy parameters of the computing system through a hierarchical encoding mechanism, generating a full-stack digital gene encoding for the computing system. The deployment strategy generation module is used to drive the software-hardware co-evolution LLM to generate multiple software-hardware deployment strategies, taking the digital gene encoding and the economic form of the computing system as input. These multiple software-hardware deployment strategies are used as a decision-making reward model to score each unit strategy within the strategy, and finally, the strategy with the highest score among the multiple strategies is merged into the final strategy. The adaptive life system compilation module is used to output executable software-hardware scripts based on the final strategy content generated by the LLM and the hardware compilation specifications. Hardware resources are allocated according to the final strategy generated by the LLM, and the executable scripts undergo simulated security sandbox testing, enabling direct deployment and computational calls from the software to the required hardware. The life system compilation module acts as a security barrier, separating the cognitive capabilities of the LLM from the deterministic execution of the underlying system, utilizing the intelligence of AI while ensuring the reliability and security of the system. The deployment module is used to deploy the scripts that have passed the sandbox test to the target system. After the system stabilizes, the latest digital gene and corresponding strategy encoding are stored in the digital gene library.

[0070] This embodiment also includes a hardware monitoring and analysis module for continuously monitoring system performance metrics (such as CPU utilization, memory usage, network throughput, disk I / O, etc.) and task execution metrics (such as task completion time, error rate, resource consumption, etc.). The system presets update trigger conditions, including performance degradation exceeding a threshold (e.g., CPU utilization consistently above 85%), an increase in the error rate (e.g., task failure rate exceeding 5%), and changes in new task requirements.

[0071] In this embodiment, the deployment strategy generation module includes the following components: a system status analysis module, used to integrate hardware sensor data (such as CPU utilization, memory usage, FPGA temperature) and software logs (such as container performance, security events) into an input format understandable by LLM; an optimization strategy generation module, used to generate cross-level optimization strategies based on the metadata tags of the digital gene (such as security policies, performance targets) and system status analysis results; for example, adjusting FPGA resource configuration, optimizing container scheduling rules, updating BIOS parameters, etc.; an execution plan formulation module, used to convert optimization strategies into executable API calls or configuration file changes and record them as a new version of the digital gene; and a decision reward model, used to score and fuse multiple strategy schemes generated by LLM to generate the final strategy scheme output. The decision reward model scores multiple strategy schemes generated by LLM and fuses them into a more reliable strategy scheme, avoiding the limitations of a single strategy scheme.

[0072] In this embodiment, the adaptive life system compilation module includes: a policy verification module for verifying the syntactic correctness and static constraint satisfaction of the policy using formal methods (such as the B method or Coq); an execution sandbox module for creating an isolated hardware and software execution environment based on a trusted execution environment (such as Intel SGX or ARM TrustZone) to simulate the policy execution effect and prevent malicious decisions from affecting the production system; a security metadata tagging module for embedding security attributes (such as prohibiting modification of the UEFI boot order) in the digital gene, and the compiler forcibly checking whether the policy violates these metadata constraints; and a version management module for recording the version change history of the digital gene based on blockchain technology to ensure that each evolution is rollbackable and the policy is verifiable.

[0073] The specific operation steps of the full-stack intelligent deployment system based on digital genes and LLM described in this embodiment are the same as those of the full-stack intelligent deployment method based on digital genes and LLM described in Embodiment 1 above, and will not be repeated here.

[0074] Example 3

[0075] This embodiment discloses a full-stack intelligent deployment device based on digital genes and LLM, including a processor and a memory electrically connected to each other; the memory is used to store computer programs; when the processor executes the aforementioned computer programs, it can realize the full-stack intelligent deployment method based on digital genes and LLM as described in Embodiment 1. The specific deployment method is the same as in Embodiment 1, and will not be repeated here.

[0076] Example 4

[0077] This embodiment discloses a computer-readable storage medium storing a computer program. When the computer program is executed, it can implement the full-stack intelligent deployment method based on digital genes and LLM as described in Embodiment 1. The specific deployment method is the same as in Embodiment 1 and will not be repeated here.

[0078] The computer described in this application embodiment can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. The computer-readable storage medium can be any usable medium that a computer can read, or a data storage device such as a server or data center that integrates one or more usable media. The usable medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital versatile optical disc (DVD)), or a semiconductor medium (e.g., solid-state drive (SSD)). The software formed by the computer's stored code can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other storage media that are mature in the art.

[0079] In the various embodiments of this application, the functional modules can be integrated into one processing unit or module, or each module can exist physically separately, or two or more modules can be integrated into one unit or module. In the above embodiments, they can be implemented entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated.

[0080] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A full-stack intelligent deployment method based on digital genes and LLM, characterized in that, Includes the following steps: Step 1: Use a hierarchical coding mechanism to structure the hardware configuration, software environment, and strategy parameters of the computing system to generate the digital gene code of the entire computing system stack. Step 2: Using the economic form of digital gene coding and computing systems as input, drive the software and hardware co-evolution LLM to generate multiple software and hardware deployment strategies; Step 3: Use multiple sets of software and hardware deployment strategies as a decision reward model, score each unit strategy in the strategy, and finally merge the strategies in the multiple strategies that meet the preset conditions into the final strategy. Step 4: Input the final policy generated by LLM into the Adaptive Life System Compiler, which outputs software and hardware executable scripts based on the policy content and hardware compilation specifications. Step 5: Based on the final strategy generated by the LLM, the life system compiler allocates hardware resources on the hardware side; Step 6: On the allocated hardware resources, perform a simulated security sandbox test on the aforementioned executable script; Step 7: Based on the sandbox test results, if an anomaly occurs during the test, repeat steps 2 and 3 until the verification is successful. Step 8: Deploy the verified script to the target system.

2. The full-stack intelligent deployment method based on digital genes and LLM according to claim 1, characterized in that: The system collects operational data in real time. When an update condition is triggered, the system updates its digital genes and repeats steps 2 to 8 until the system reaches a new stable state.

3. The full-stack intelligent deployment method based on digital genes and LLM according to claim 2, characterized in that: Once the system is stable, the latest digital genes and corresponding strategy codes will be stored in the digital gene bank.

4. The full-stack intelligent deployment method based on digital genes and LLM according to any one of claims 1-3, characterized in that: The digital gene encoding is the process of encoding and compressing hardware, software, strategies, and metadata content into a unique digital gene code. The digital gene comprises the following components: The hardware gene layer describes the characteristics of hardware resources, including processor architecture, memory configuration, storage type, and network topology. The software gene layer describes software-level characteristics, including operating system version, middleware configuration, and application deployment strategy. The strategy gene layer is used to describe the characteristics at the decision-making level, including system optimization strategies, security rules, and performance goals. The metadata layer is used to describe auxiliary information, including the version information, evolutionary history, and applicable environment of the digital gene.

5. The full-stack intelligent deployment method based on digital genes and LLM according to claim 4, characterized in that: The economic model of the computing system is set up in various forms according to the characteristics of the task, including the economic priority model, the performance priority model, and the comprehensive model.

6. A full-stack intelligent deployment system based on digital genes and LLM, characterized in that: It includes a digital gene encoding module, a deployment strategy generation module, an adaptive life system compilation module, and a deployment module; The digital gene encoding module is used to provide a structured description of the hardware configuration, software environment, and strategy parameters of the computing system through a hierarchical encoding mechanism, thereby generating the digital gene encoding of the entire computing system stack. The deployment strategy generation module is used to drive the software and hardware co-evolution LLM to generate multiple software and hardware deployment strategies by taking the economic form of digital gene coding and computing system as input; the multiple software and hardware deployment strategies are used as a decision reward model to score each unit strategy in the strategy; and finally, the strategies in the multiple strategies whose scores meet the preset conditions are merged into the final strategy. The adaptive life system compilation module is used to output software and hardware executable scripts based on the final policy content and hardware compilation specifications generated by LLM; allocate hardware resources according to the final policy generated by LLM; and perform simulated security sandbox tests on the executable scripts. The deployment module is used to deploy scripts that have passed the sandbox test to the target system; after the system is stable, the latest digital genes and corresponding strategy codes are stored in the digital gene library.

7. The full-stack intelligent deployment system based on digital genes and LLM according to claim 6, characterized in that, The deployment strategy generation module includes the following components: The system status analysis module is used to integrate hardware sensor data and software logs into an input format that LLM can understand; The optimization strategy generation module is used to generate cross-level optimization strategies based on the metadata tags of digital genes and the results of system status analysis. Execution plan formulation module: used to translate optimization strategies into executable API calls or configuration file changes and record them as a new version of the digital gene; Decision reward model: used to score and merge multiple policy options generated by LLM to generate the final policy option output.

8. The full-stack intelligent deployment system based on digital genes and LLM according to claim 6, characterized in that, The adaptive life system compilation module includes the following components: The policy verification module is used to verify the syntactic correctness and static constraint satisfaction of policies using formal methods. The execution sandbox module is used to create isolated software and hardware execution environments based on the trusted execution environment to simulate the effects of policy execution. The security metadata tagging module is used to embed security attributes in digital genes, and the compiler enforces checks to see if the policy violates these metadata constraints. The version management module is used to record the version change history of digital genes based on blockchain technology.

9. A full-stack intelligent deployment device based on digital gene and LLM, comprising a processor and a memory electrically connected to each other; the memory is used to store computer programs; characterized in that: When the processor executes the aforementioned computer program, it can implement the full-stack intelligent deployment method based on digital genes and LLM as described in any one of claims 1-5.

10. A computer-readable storage medium storing a computer program thereon; characterized in that: When the computer program is executed, it can implement the full-stack intelligent deployment method based on digital genes and LLM as described in any one of claims 1-5.

Images