Artificial intelligence chip fault repair method, device, equipment and medium

Abstract

This invention provides a method, apparatus, device, and medium for repairing AI chip faults, relating to the field of computer technology. The method includes: acquiring a log file; dividing the log into multiple log blocks according to a preset segmentation threshold; performing parallel analysis on each log block using a preset lightweight language model to obtain multiple block summaries; merging all block summaries using a preset high-capacity language model to obtain a diagnostic report; analyzing the diagnostic report using a preset large language model to obtain suggested actions; retrieving logs using a preset rule matching model to determine if restart keywords exist; determining the target action based on the suggested actions and restart keyword detection results; repairing the AI ​​chip fault based on the target action; and terminating the repair process and outputting the repair result when the repair process meets a preset termination condition. This invention can solve the problems of low diagnostic accuracy and easy getting stuck in abnormal loops in existing operation and maintenance systems when repairing AI chip faults.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, device, and medium for repairing AI chip faults. Background Technology

[0002] In real-world production environments, AI server clusters are prone to initialization failures, inner-loop topology anomalies, and hardware detection anomalies. Traditional manual maintenance and repair methods suffer from low efficiency, high labor costs, and the inability to be on duty at night. They are particularly inadequate for meeting the automated diagnosis and repair needs of initialization failures and chip interconnect communication link loop failures in large-scale model computing clusters composed of AI computing power chips such as TPU / NPU (e.g., DLC high-computing-power chips).

[0003] Existing operation and maintenance systems based on large language models (such as the ReAct framework) also have many shortcomings: Existing operation and maintenance systems typically infer fault causes directly from system logs, which is highly prone to discrepancies between the inferred output and the actual logs. For example, the model may detect a fault, but the actual log data may not support this conclusion. In actual testing, models without structured validation have low initial diagnostic accuracy (only around 60% in some scenarios), leading to incorrect subsequent repair strategies.

[0004] Furthermore, if the automated repair system uses "system restart" as a universal repair method, when the root cause of the fault is a problem that cannot be resolved by restarting, such as hardware damage or abnormal network inner loop topology, the system will fall into a vicious cycle of repeated restarts. In a batch of faulty equipment repair tests, some devices failed to repair due to frequent and ineffective restarts, wasting a lot of time and power resources.

[0005] In addition, some faults, such as inner-loop topology anomalies and physical hardware damage, are beyond the scope of automated repair capabilities and must be handled manually. However, the existing system lacks a clear fault handling boundary determination mechanism, causing the system to waste too much time on invalid attempts or prematurely abandon faults that could be automatically repaired. Summary of the Invention

[0006] This invention provides an AI chip fault repair method, apparatus, device, and medium to solve the problems of low diagnostic accuracy and easy getting stuck in abnormal loops in existing operation and maintenance systems when repairing AI chip faults.

[0007] In a first aspect, embodiments of the present invention provide an AI chip fault repair method, including: Obtain the log file and divide the log into multiple log blocks according to the preset block threshold; The log blocks are analyzed in parallel using a pre-set lightweight language model to obtain multiple block summaries. All block summaries are then merged using a pre-set high-capacity language model to obtain a diagnostic report. The diagnostic report is analyzed using a pre-defined large language model to obtain suggested actions; Logs are retrieved using a pre-defined rule-based matching model to determine if any restart keywords exist. Based on the suggested actions and the results of restarting the keyword detection, determine the target action; The AI ​​chip is repaired based on the target operation, and the repair process is terminated when the preset termination condition is met, and the repair result is output.

[0008] In one possible implementation, the step of analyzing the diagnostic report using a preset large language model to obtain suggested actions includes: If the analysis results show that the current situation is any of the following: the system is executing for the first time, the system is not initialized, or the hardware detection failed and the system restarted and recovered, it is recommended to perform an initialization operation. If the analysis results show that there are errors but a restart is not required, it is recommended to perform a repair operation; If the analysis results indicate that a restart is required, it is recommended to perform a restart.

[0009] In one possible implementation, determining the target operation based on the suggested operation and the restart keyword detection results includes: If the keyword "restart" is found in the log, then the target operation is determined to be a restart operation. If the restart keyword is not found in the log, then the target operation is determined to be the suggested operation.

[0010] In one possible implementation, terminating the repair process when the repair process meets a preset termination condition includes: terminating the repair process when a repair success keyword is detected in the log, or when a preset maximum number of repairs is reached.

[0011] In one possible implementation, if a restart operation is recommended, then: Generate manual confirmation information for manual confirmation; If manual confirmation is required to restart, then the restart operation will be performed. If it is manually confirmed that a restart is not required, then perform the repair operation.

[0012] In one possible implementation, the method further includes: Before performing the restart operation, store the current repair data to form a historical evidence chain; After performing the restart operation, based on the historical evidence chain, it is determined whether the number of times the same repair strategy was executed without success has reached a preset threshold. If the preset threshold is reached, a prompt will appear indicating that manual intervention is required to repair the process.

[0013] In one possible implementation, the method further includes: After performing a reboot, read the underlying hardware state; The underlying hardware state is compared with the expected state anchor point to verify the consistency of the hardware physical characteristics. If the hardware physical characteristics are inconsistent, the system initialization is re-executed.

[0014] Secondly, embodiments of the present invention provide an AI chip fault repair device, comprising: The analysis module is used to obtain log files, divide the logs into multiple log blocks according to a preset block threshold, perform parallel analysis on each log block using a preset lightweight language model to obtain multiple block summaries, and merge all block summaries using a preset high-capacity language model to obtain a diagnostic report. The large language diagnostic module analyzes the diagnostic report using a preset large language model to obtain suggested actions; The rule matching module is used to retrieve logs using a preset rule matching model to determine whether restart keywords exist; The repair module is used to determine the target operation based on the suggested operation and the restart keyword detection results; to repair the AI ​​chip based on the target operation; and to terminate the repair process and output the repair result when the repair process meets the preset termination conditions.

[0015] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.

[0016] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect or any possible implementation thereof.

[0017] Fifthly, embodiments of the present invention provide a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect or any possible implementation thereof.

[0018] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows: In this embodiment of the invention, logs are divided into multiple log blocks, and a lightweight language model is used to perform parallel analysis on each log block to obtain multiple block summaries. A high-capacity language model is then used to merge all block summaries to obtain a diagnostic report. This addresses the issue of intermediate loss of information when language models process long texts (i.e., a significant decrease in the model's attention to and retrieval accuracy for information in the middle of the text), ensuring that the information within each block is within the effective range of the model's attention. A pre-set large language model analyzes the diagnostic report to obtain suggested actions. A pre-set rule-matching model retrieves logs to determine if restart keywords exist. Based on the suggested actions and the restart keyword detection results, the target action is determined to repair the AI ​​chip, ensuring that the system will always execute a restart operation when required. This retains the flexibility of the large language model in handling complex scenarios while avoiding the risk of misjudgment by the large language model through rule constraints. Finally, the repair process is monitored in real time, and terminated when a pre-set termination condition is met, suppressing ineffective repair strategy trajectories and preventing the system from falling into an infinite loop. This invention solves the problems of low diagnostic accuracy and easy entry into abnormal loops in existing operation and maintenance systems when repairing AI chip faults. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the implementation of the AI ​​chip fault repair method provided in this embodiment of the invention. Figure 2 This is a flowchart of the repair tool selection process provided in the embodiments of the present invention; Figure 3 This is a schematic diagram of the structure of the AI ​​chip fault repair device provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0021] The AI ​​chip fault repair method of this invention is applicable to intelligent operation and maintenance systems of large-scale AI computing power clusters, especially for scenarios such as initialization faults, inner loop topology anomalies, hardware detection failures, and chip interconnect communication link loop faults of AI chips such as TPU / NPU (e.g., DLC high-computing-power chips). The invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] See Figure 1 The document illustrates a flowchart of the AI ​​chip fault repair method provided in an embodiment of the present invention, which is described in detail below: Step S101: Obtain log files (system logs, hardware detection logs, tool execution logs, etc.) and divide the logs into multiple log blocks according to the preset block threshold.

[0023] Language models suffer from the "lost in the middle" phenomenon when processing long texts, meaning that the model's attention to information in the middle of the text and its retrieval accuracy decrease significantly. This embodiment systematically avoids this problem by dividing large log files into blocks for parallel analysis and generating structured evidence reports for subsequent retrieval and decision-making. The length of each block is controlled within the effective range of the language model's attention.

[0024] Step S102: Perform parallel analysis on each log block using a preset lightweight language model to obtain multiple block summaries. Then, merge all block summaries using a preset high-capacity language model to obtain a diagnostic report.

[0025] To address the issues of low efficiency and the potential for key information to be buried in large log file analysis, this embodiment proposes a structured evidence chain construction method based on parallel sharding parsing. Unlike simple text summarization, this embodiment utilizes a parallel computing architecture to process massive logs in blocks, with the core focus on extracting fault fingerprints and restart signals that drive downstream decisions. Specifically: during the sharding parsing stage, a high-sensitivity scan is performed specifically on restart-related semantic patterns, extracting them as the highest-priority control signals; and combined features including error codes, hardware status register values, and key error phrases are extracted to form a unique fault fingerprint for subsequent deduplication and knowledge base matching. The generated diagnostic report is not merely a text description but structured data containing evidence fields. This field serves as a hash key linking the current fault with historical operations, ensuring that every repair decision is traceable, achieving a dimensional transformation from unstructured logs to structured decision credentials.

[0026] For example, the block digest is shown in Table 1: Table 1 Block Summary Data Structure Table

[0027] For example, the diagnostic report is shown in Table 2: Table 2. Diagnostic Report Data Structure Table

[0028] Step S103: Analyze the diagnostic report using a pre-set large language model to obtain suggested actions.

[0029] This embodiment implements differentiated control based on risk level, intelligently selecting the most suitable repair tool through a large language model. For example, such as... Figure 2 As shown: If the analysis results show that the current situation is any of the following: the system is executing for the first time, the system is not initialized, or the hardware detection failed and the system restarted and recovered, it is recommended to perform an initialization operation. If the analysis results show that there are errors but a restart is not required, it is recommended to perform a repair operation; If the analysis results indicate that a restart is required, it is recommended to perform a restart.

[0030] Step S104: Search the logs using a preset rule matching model to determine if there are any restart keywords.

[0031] In this embodiment, a rule-matching model is used to perform millisecond-level scanning of the original log file to search for the existence of preset restart keywords. Restart keywords include "restart required," "please restart the system," "after restarting the system," and "please restart." If any restart keyword is detected, it is marked as a restart signal; otherwise, it is marked as no restart signal.

[0032] Step S105: Based on the suggested operation and the keyword detection results, determine the target operation.

[0033] To address the issue of significant differences in risk levels among different repair operations and the potential for system damage from indiscriminate execution, this embodiment proposes a hierarchical hybrid strategy based on dual-channel collaboration. This strategy organically integrates probabilistic large-scale semantic decision-making with deterministic rule constraints, achieving collaborative control of intelligent decision-making and safety fallback through a dual-channel architecture and hierarchical constraint mechanism.

[0034] Channel A (Generative Channel): A probabilistic decision-making channel based on a large language model; Inputs: Log context, historical diagnostic reports, tool description; Capabilities: Proficient in handling unstructured, unseen fault modes and identifying complex combination scenarios; Output: Tool recommendations and confidence scores.

[0035] Channel B (Discriminative Channel): A deterministic constraint channel based on a rule engine; Input: raw log stream, list of predefined keywords (e.g., "restart required"); Capability: Performs millisecond-level scanning of high-risk operations to identify clear security boundary triggering conditions; Output: Boolean type blocking signal.

[0036] In one possible implementation, if the restart keyword is present in the log, the target operation is determined to be a restart operation; if the restart keyword is not present in the log, the target operation is determined to be a suggested operation.

[0037] In this embodiment, the large language model first performs semantic analysis and tool recommendation based on the complete context; the rule matching layer acts as a safety valve to verify the output of the large language model. If a restart keyword is detected, the large language model's decision is forcibly overridden, ensuring that the system will always perform a restart operation when a restart is required. This design retains the flexibility of the large language model in handling complex scenarios while avoiding the risks of potential misjudgments by the large language model through rule constraints.

[0038] Specifically: While channel A performs semantic analysis, channel B simultaneously scans the log content. If channel B detects a restart keyword, it immediately acquires a decision mutex lock, blocks the output of channel A, and forces the decision result to overwrite the restart tool; if channel A suggests a restart but channel B does not detect a clear signal, the system will downgrade to execute the repair tool or request manual confirmation to prevent unnecessary restarts.

[0039] Step S106: Perform fault repair on the AI ​​chip based on the target operation, and terminate the repair process when the repair process meets the preset termination conditions, and output the repair result.

[0040] In this embodiment, termination conditions for the repair process are set. For example, the repair process is terminated when a keyword indicating successful repair is detected in the log, or when the preset maximum number of repairs is reached, to prevent the system from getting stuck in an ineffective repair loop.

[0041] In this embodiment of the invention, logs are divided into multiple log blocks, and a lightweight language model is used to perform parallel analysis on each log block to obtain multiple block summaries. A high-capacity language model is then used to merge all block summaries to obtain a diagnostic report. This addresses the issue of intermediate loss of information when language models process long texts (i.e., a significant decrease in the model's attention to and retrieval accuracy for information in the middle of the text), ensuring that the information within each block is within the effective range of the model's attention. A pre-set large language model analyzes the diagnostic report to obtain suggested actions. A pre-set rule-matching model retrieves logs to determine if restart keywords exist. Based on the suggested actions and the restart keyword detection results, the target action is determined to repair the AI ​​chip, ensuring that the system will always execute a restart operation when required. This retains the flexibility of the large language model in handling complex scenarios while avoiding the risk of misjudgment by the large language model through rule constraints. Finally, the repair process is monitored in real time, and terminated when a pre-set termination condition is met, suppressing ineffective repair strategy trajectories and preventing the system from falling into an infinite loop. This invention solves the problems of low diagnostic accuracy and easy entry into abnormal loops in existing operation and maintenance systems when repairing AI chip faults.

[0042] In some embodiments, to address the problem that automated repair systems may get stuck in invalid loops (especially infinite loops of reboot-failure-reboot) and waste resources, this embodiment proposes an invalid repair strategy trajectory suppression mechanism based on hardware feature consistency verification. The core innovation of this mechanism is that the system can transcend physical reboot cycles and determine the validity of repair operations by comparing the consistency between key hardware features and log semantics, thereby making suppression decisions.

[0043] In one possible implementation, the strategy includes: after performing a reboot, reading the underlying hardware state; comparing the underlying hardware state with the expected state anchor to verify the consistency of the hardware physical characteristics; if the hardware physical characteristics are inconsistent, then re-executing system initialization.

[0044] Hardware key characteristic consistency verification is a crucial verification step after system reboot. Unlike simple device presence checks, the system delves into the underlying hardware state, including PCIe link width, link speed, BAR space mapping status, and device register values. The system compares these physical characteristics with persistent expected state anchors to verify whether the hardware has truly recovered to a repairable physical state. If physical characteristics are inconsistent (e.g., PCIe speed reduction or BAR space mapping failure), it is determined that the reboot has not resolved the physical layer problem, triggering suppression or reinitialization.

[0045] In one possible implementation, the strategy includes: before performing the restart operation, storing the current repair data to form a historical evidence chain; after performing the restart operation, determining, based on the historical evidence chain, whether the number of times the same repair strategy has been executed without success has reached a preset threshold; if the preset threshold is reached, prompting that manual intervention is required in the repair process.

[0046] Before performing a restart, the system serializes and stores the current repair data (including executed tools, diagnostic reports, and iteration counts) to a non-volatile medium. After restarting, the system automatically loads the historical evidence chain and restores the logical judgment context. Based on the historical evidence chain, the system can identify patterns where the problem persists after a restart. If it detects that the same repair strategy has been executed for N consecutive iterations without success, suppression is triggered and manual intervention is prompted. That is, this embodiment supports multiple suppression triggers: maximum iteration count limit, successful keyword detection, duplicate strategy detection, and hardware feature verification failure.

[0047] In one specific embodiment, taking the initialization failure of the AI ​​server DLC chip as an example, the repair process of this method is demonstrated: (1) Initialization operation failed on first execution. The system initiated an automatic repair process, detected no historical sessions, and determined it to be the first execution. Perform a hardware check and return the results; The large language model suggests performing an initialization operation, but the rule matching model did not detect a restart signal, so the target operation is the initialization operation. The initialization tool was invoked, but during execution, the log showed "Hardware detection failed. Please try restarting the machine and then executing this script again." Initialization failed. Parallel log block parsing generates diagnostic reports, which are marked as "restart required".

[0048] (2) If a restart signal is detected, a restart operation is forcibly executed. The rule matching model detected the restart keyword in the log and forced the target operation to be a restart operation; Persistently store the historical evidence chain, record hardware state anchor points, and set the session state to wait for restart; Use the restart tool to perform a system restart.

[0049] (3) Restart and restore, re-execute the initialization operation and succeed. After the system restarts, the repair process is initiated, and the system is determined to be undergoing a restart recovery. Perform hardware feature consistency verification, read the underlying hardware state and match it with the anchor point, and the verification passes; Perform a hardware check and return the results; The large language model suggests performing an initialization operation without a restart signal; the target operation is the initialization operation. The initialization tool is invoked. If the keyword indicating successful repair is detected and the termination condition is met, the repair process is terminated immediately, and the repair is marked as successful.

[0050] This embodiment has the following technical advantages: (1) Structured log analysis: Large logs are divided into blocks for parallel analysis through parallel log sharding parsing mode, generating searchable structured diagnostic reports, supporting knowledge base retrieval and accurate fault location; (2) Intelligent tool routing: Based on the big oracle model, the tool selection combines hierarchical strategies and priority rules to ensure that the restart operation is only executed when it is clearly needed, thus avoiding blind restarts; (3) Recovery across restart states: Supports automatic recovery of execution context after system restart, ensuring continuity of the repair process; (4) Invalid repair strategy trajectory suppression mechanism: Combine successful keyword detection, maximum iteration limit and invalid recovery suppression to terminate the repair in time or trigger manual intervention.

[0051] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0052] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0053] Figure 3 A schematic diagram of the AI ​​chip fault repair device provided in an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below: like Figure 3 As shown, the AI ​​chip fault repair device 3 includes: Analysis module 31 is used to obtain log files, divide the logs into multiple log blocks according to a preset block threshold, perform parallel analysis on each log block using a preset lightweight language model to obtain multiple block summaries, and merge all block summaries using a preset high-capacity language model to obtain a diagnostic report. The large language diagnostic module 32 analyzes the diagnostic report through a preset large language model and obtains suggested actions; The rule matching module 33 is used to retrieve logs using a preset rule matching model to determine whether a restart keyword exists. Repair module 34 is used to determine the target operation based on the suggested operation and the restart keyword detection results; perform fault repair on the AI ​​chip based on the target operation; and terminate the repair process and output the repair results when the repair process meets the preset termination conditions.

[0054] In one possible implementation, the large language diagnostic module 32 is used for: If the analysis results show that the current situation is any of the following: the system is executing for the first time, the system is not initialized, or the hardware detection failed and the system restarted and recovered, it is recommended to perform an initialization operation. If the analysis results show that there are errors but a restart is not required, it is recommended to perform a repair operation; If the analysis results indicate that a restart is required, it is recommended to perform a restart.

[0055] In one possible implementation, the repair module 34 is used for: If the keyword "restart" is found in the log, then the target operation is determined to be a restart operation. If the restart keyword is not found in the log, the target operation is determined to be the recommended operation.

[0056] In one possible implementation, the repair module 34 is used to terminate the repair process when a repair success keyword is detected in the log, or when the preset maximum number of repairs is reached.

[0057] In one possible implementation, if a restart operation is recommended, the repair module 34 is used to: Generate manual confirmation information for manual confirmation; If manual confirmation is required to restart, then the restart operation will be performed. If it is manually confirmed that a restart is not required, then perform the repair operation.

[0058] In one possible implementation, the repair module 34 is used for: Before performing the restart operation, store the current repair data to form a historical evidence chain; After performing the restart operation, based on the historical evidence chain, determine whether the number of times the same repair strategy was executed without success has reached a preset threshold. If the preset threshold is reached, a prompt will appear indicating that manual intervention is required to repair the process.

[0059] In one possible implementation, the repair module 34 is used for: After performing a reboot, read the underlying hardware state; The underlying hardware state is compared with the expected state anchor point to verify the consistency of the hardware physical characteristics. If the hardware physical characteristics are inconsistent, the system initialization is re-executed.

[0060] In this embodiment of the invention, logs are divided into multiple log blocks, and a lightweight language model is used to perform parallel analysis on each log block to obtain multiple block summaries. A high-capacity language model is then used to merge all block summaries to obtain a diagnostic report. This addresses the issue of intermediate loss of information when language models process long texts (i.e., a significant decrease in the model's attention to and retrieval accuracy for information in the middle of the text), ensuring that the information within each block is within the effective range of the model's attention. A pre-set large language model analyzes the diagnostic report to obtain suggested actions. A pre-set rule-matching model retrieves logs to determine if restart keywords exist. Based on the suggested actions and the restart keyword detection results, the target action is determined to repair the AI ​​chip, ensuring that the system will always execute a restart operation when required. This retains the flexibility of the large language model in handling complex scenarios while avoiding the risk of misjudgment by the large language model through rule constraints. Finally, the repair process is monitored in real time, and terminated when a pre-set termination condition is met, suppressing ineffective repair strategy trajectories and preventing the system from falling into an infinite loop. This invention solves the problems of low diagnostic accuracy and easy entry into abnormal loops in existing operation and maintenance systems when repairing AI chip faults.

[0061] Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 4 As shown, the electronic device 4 in this embodiment includes a processor 40 and a memory 41. The memory 41 stores a computer program 42. When the processor 40 executes the computer program 42, it implements the steps in the various method embodiments described above. Alternatively, when the processor 40 executes the computer program 42, it implements the functions of each module in the various device embodiments described above.

[0062] For example, computer program 42 may be divided into one or more modules / units, which are stored in memory 41 and executed by processor 40 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 42 in electronic device 4.

[0063] Electronic device 4 may include, but is not limited to, processor 40 and memory 41. Those skilled in the art will understand that... Figure 4 This is merely an example of electronic device 4 and does not constitute a limitation on electronic device 4. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 4 may also include input / output devices, network access devices, buses, etc.

[0064] The processor 40 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0065] The memory 41 can be an internal storage unit of the electronic device 4, such as a hard disk or RAM. The memory 41 can also be an external storage device of the electronic device 4, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory 41 can include both internal and external storage units of the electronic device 4. The memory 41 is used to store the computer program 42 and other programs and data required by the electronic device 4. The memory 41 can also be used to temporarily store data that has been output or will be output.

[0066] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0067] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0068] This invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0069] Computer programs include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0070] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0071] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for repairing AI chip faults, characterized in that, include: Obtain the log file and divide the log into multiple log blocks according to the preset block threshold; The log blocks are analyzed in parallel using a pre-set lightweight language model to obtain multiple block summaries. All block summaries are then merged using a pre-set high-capacity language model to obtain a diagnostic report. The diagnostic report is analyzed using a pre-defined large language model to obtain suggested actions; Logs are retrieved using a pre-defined rule-based matching model to determine if any restart keywords exist. Based on the suggested actions and the results of restarting the keyword detection, determine the target action; The AI ​​chip is repaired based on the target operation, and the repair process is terminated when the preset termination condition is met, and the repair result is output.

2. The AI ​​chip fault repair method according to claim 1, characterized in that, The step of analyzing the diagnostic report using a pre-set large language model to obtain suggested actions includes: If the analysis results show that the current situation is any of the following: the system is executing for the first time, the system is not initialized, or the hardware detection failed and the system restarted and recovered, it is recommended to perform an initialization operation. If the analysis results show that there are errors but a restart is not required, it is recommended to perform a repair operation; If the analysis results indicate that a restart is required, it is recommended to perform a restart.

3. The AI ​​chip fault repair method according to claim 2, characterized in that, The step of determining the target operation based on the suggested operation and the restart keyword detection results includes: If the keyword "restart" is found in the log, then the target operation is determined to be a restart operation. If the restart keyword is not found in the log, then the target operation is determined to be the suggested operation.

4. The AI ​​chip fault repair method according to claim 1, characterized in that, The step of terminating the repair process when it meets the preset termination conditions includes: terminating the repair process when a successful repair keyword is detected in the log, or when the preset maximum number of repairs is reached.

5. The AI ​​chip fault repair method according to claim 2, characterized in that, If a restart is recommended, then: Generate manual confirmation information for manual confirmation; If manual confirmation is required to restart, then the restart operation will be performed. If it is manually confirmed that a restart is not required, then perform the repair operation.

6. The AI ​​chip fault repair method according to any one of claims 1 to 5, characterized in that, The method further includes: Before performing the restart operation, store the current repair data to form a historical evidence chain; After performing the restart operation, based on the historical evidence chain, it is determined whether the number of times the same repair strategy was executed without success has reached a preset threshold. If the preset threshold is reached, a prompt will appear indicating that manual intervention is required to repair the process.

7. The AI ​​chip fault repair method according to any one of claims 1 to 5, characterized in that, The method further includes: After performing a reboot, read the underlying hardware state; The underlying hardware state is compared with the expected state anchor point to verify the consistency of the hardware physical characteristics. If the hardware physical characteristics are inconsistent, the system initialization is re-executed.

8. An AI chip fault repair device, characterized in that, include: The analysis module is used to obtain log files and divide the logs into multiple log blocks according to a preset block threshold. The log blocks are analyzed in parallel using a pre-set lightweight language model to obtain multiple block summaries. All block summaries are then merged using a pre-set high-capacity language model to obtain a diagnostic report. The large language diagnostic module analyzes the diagnostic report using a preset large language model to obtain suggested actions; The rule matching module is used to retrieve logs using a preset rule matching model to determine whether restart keywords exist; The repair module is used to determine the target operation based on the suggested operation and the restart keyword detection results; The AI ​​chip is repaired based on the target operation, and the repair process is terminated when the preset termination condition is met, and the repair result is output.

9. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images