47 results about "Direct execution" patented technology

The invention provides an instant messaging driving type automatic service processing method and system based on a large language model and a model context protocol (MCP), and relates to the technical field of computers, and the method comprises the following steps: a user sends a message in a group of an instant messaging platform and makes a group robot to form a Webhook event; the MCP client exposes an HTTP (Hyper Text Transport Protocol) interface to receive a Webhook event, completes authentication and analyzes a message; the MCP client sends the message to a large language model, ACL permission judgment is carried out, and an MCPServer method and parameters to be called are recognized; if the permission is judged to be'unauthorized ', stopping, and then submitting permission errors to a big language model organization and description and directly executing and returning'insufficient permission'; if the permission is judged to be'authorization ', the next step is carried out; and the MCP client calls the MCP server (through HTTP + SSE or stdio) to obtain a structured result. According to the method, the unstructured chat instruction is stably converted into the standardized MCP request, and coupling of dialect type APIs of all back-end systems is eliminated.

The invention discloses an artificial intelligence large model application method supporting vector graphic input and processing, and relates to the technical field of artificial intelligence, comprising: receiving a natural language instruction and multi-modal data of a user, the multi-modal data comprising vector graphic data selectable to input by the user; analyzing the vector graphic data, extracting a spatial relationship and a semantic attribute, constructing a structured representation of the spatial relationship and the semantic attribute, performing cross-modal semantic alignment and joint coding on the structured representation and a natural language instruction, and inputting into a multi-modal large model to generate a task execution plan comprising a plurality of steps, and scheduling the multi-mode large model and an external vector graphic processing tool to execute to obtain a processing result, and aiming at solving the technical problem that the vector graphic operation is difficult to directly execute by a general large model, the general cognitive ability of the large model and the accurate execution ability of the external professional tool are decoupled and coordinated; and the application potential of the large model in the professional field is effectively expanded.

The invention provides an intelligent caching method and device based on a self-adaptive caching strategy, and the method comprises the steps: collecting an access event flow of each caching key in real time through a sensing probe embedded in a caching client, the event flow comprising Key, operation type, timestamp, request source, hit state and response time; running a streaming processing engine at a client based on an access event stream, and calculating indexes such as instantaneous QPS, continuous miss times and short-term access frequency by adopting a sliding time window; a calculation result is input into a rule engine for online matching, a decision instruction for a single Key is generated, and the decision instruction comprises distributed mutual exclusion lock addition, local TTL dynamic adjustment and read-write mode switching; a decision instruction is directly executed in a client memory through a strategy executor, and nanosecond strategy effectiveness is achieved; and finally, collecting system feedback data after strategy execution, inputting the system feedback data into a machine learning module for online learning, and dynamically optimizing a rule threshold value or recommending a new strategy.

The invention discloses a robot strategy generation method and device, a computer and a storage medium, whether an initial plan generated by a VLM strategy network should be directly executed or not is judged through a confidence trigger, unnecessary reflection is achieved, task execution efficiency is improved, if the confidence is lower than a preset threshold value, an reflection stage is entered, and the task execution efficiency is improved. A plurality of candidate future trajectories guided by a plurality of candidate actions are generated, advantage evaluation is performed on each candidate future trajectory, a more accurate supervision signal is provided, decision robustness is enhanced, an evaluation result is used as feedback of a VLM strategy network, feedback information of all trajectories is aggregated by the VLM strategy network in a decoding stage to correct an initial plan, and the decision robustness is improved. According to the method, the low-quality action planning can be accurately identified and corrected, and the balance of efficiency and performance is realized.

The invention provides a heterogeneous reasoning acceleration method and system for a hybrid expert model, and the method comprises the steps: analyzing the historical activation behavior of each expert based on a representative data set of a multi-task context, constructing an expert portrait representing the relevance between the experts and tasks, and providing priori knowledge for scheduling; a cost prediction model is established for the heterogeneous computing units, and time delays of different execution paths are predicted according to input characteristics (such as data volume) during operation; in the reasoning process, an expert portrait and a cost prediction model are combined, an optimal path is selected from candidate paths (CPU execution, GPU direct execution and weight migration to GPU execution) for an activated expert, and task execution and data migration are controlled; based on real-time expert activation statistics and a system state, an expert portrait and a resident strategy of an expert in a heterogeneous environment are dynamically updated, and a feedback closed loop is formed. According to the method, adaptive optimization of the expert reasoning path is realized, and the problem of performance degradation of a traditional static distribution strategy in a complex input scene is solved.

This invention discloses a method for generating intelligent shell commands in a Linux environment, comprising: acquiring input natural language in a Linux terminal; determining whether a corresponding ready-made command exists; if it exists, directly executing the corresponding ready-made command; if it does not exist, calling a cloud model to perform semantic parsing on the natural language, identifying the user's intent based on the semantic parsing, and performing knowledge retrieval in conjunction with the context; combining the user's intent and the knowledge retrieval results, calling a large model to generate a shell command, and displaying the shell command to the user; after user confirmation, performing sandbox pre-execution; if pre-execution fails, returning the pre-execution result; if pre-execution succeeds, executing the shell command, and returning the execution result to the user after execution. This invention also discloses an intelligent shell command generation system in a Linux environment. This invention uses natural language to operate the terminal, making it convenient to use.

Embodiments of the present application provide a tensor instruction optimization method and device and a storage medium, and relate to the technical field of artificial intelligence chips. The method comprises: sequentially traversing N tensor instructions in a target basic block; if there is a target instruction on which a currently traversed tensor instruction depends, and there are L intermediate instructions between the tensor instruction and the target instruction, obtaining total additional time based on the additional execution time of each of the L intermediate instructions; and if the total additional time is greater than the unit execution time of the currently traversed tensor instruction, it is indicated that the target instruction on which the currently traversed tensor instruction depends has been executed when the pipeline executes the currently traversed tensor instruction, so the dependency descriptor of the currently traversed tensor instruction is set to a first descriptor to represent no dependency relationship. In this way, the pipeline can directly execute the tensor instruction without waiting, thereby reducing the probability of pipeline blocking, reducing the idling of hardware resources in the pipeline, and improving the utilization rate of hardware resources.

This application discloses a method and apparatus for updating a boot program. The method is applied to a storage medium including a first bank and a second bank, wherein the first bank is an active, running bank, and the second bank is an inactive, non-running bank. The method first responds to a boot program update command and queries the burning flag bit of the storage medium. This update command instructs the target boot program to be burned to each bank of the storage medium. Subsequently, when the burning flag bit indicates that the program should be burned to the inactive bank, the target boot program is burned to the second bank. Finally, after the target boot program is successfully burned to the second bank, it is copied to the first bank, thus completing the boot program update. By using the burning flag bit as the execution basis, the target boot program is prioritized for burning to the inactive, non-running second bank. After successful burning, the target boot program is copied to the first bank. The entire process does not require direct burning operations on the active bank, avoiding interference with the currently running boot program during the update process.

Approaches are described for a decentralized computing platform that enables dynamic execution of distributed services through a hybrid on-chain and off-chain architecture. The platform consists of a network of nodes, each running a virtual machine to host and execute service instances. A sequencer component records, sequences, and notarizes function calls to maintain a consistent service-level state across the network. Computation takes place off-chain on nodes that participate dynamically in specific services based on resource availability. A persistent virtual memory structure manages both service-level and instance-level state. The service-level state is synchronized across nodes participating in the same service, while the instance-level state is isolated to individual nodes for independent operations. The memory system is non-volatile, retaining state across function calls and enabling direct execution from memory. A universal procedure call (UPC) framework facilitates decentralized communication through multicast-enabled operations with customizable fault tolerance and supports applications from finance to high-performance computing.

This disclosure provides a data manipulation method, system, storage medium, and computer program product, relating to the field of computer technology. In this disclosure, a corresponding handling method is provided for situations where front-end operation instructions involve multiple syntax types. Therefore, users do not need to use a fixed syntax type during instruction input. When a user wants to perform a first operation, they can choose a first syntax type to write the instruction, which defines instructions that can directly execute the first operation. When the user wants to perform a second operation, if the first syntax type does not define instructions that can directly execute the second operation, the user can choose a second syntax type to write the instruction, which defines instructions that can directly execute the second operation. Thus, for the operation the user wants to perform, the user does not need to break the operation into multiple steps and input multiple instructions, thereby improving the user's operational efficiency.

The application relates to an adaptive wavefront control method and system based on parameter identification and active disturbance rejection. The method comprises a digital architecture adopting PC data processing and FPGA control. The PC data processing module is used to perform complex calculations such as variable forgetting factor recursive least squares, is responsible for dimensionality reduction decoupling and high-precision estimation of disordered data, and utilizes the high-speed parallel processing capability of the FPGA control system to directly execute an active disturbance rejection control algorithm at the bottom layer to offset external interference. The hardware-software collaborative design effectively combines the accuracy of the complex algorithm and the real-time performance of the bottom-layer hardware, thereby significantly enhancing the anti-interference capability and response bandwidth of the system.

This invention relates to the field of information security technology, specifically to a secure computation method for medical privacy data based on CKKS homomorphic encryption. The method involves initializing a polynomial multiplication module and a CKKS module; the user end collects medical data, performs CKKS encoding and encryption, and uploads the ciphertext to the cloud; the server end deploys a disease risk prediction model, performs homomorphic inference analysis based on the ciphertext, and obtains the ciphertext result; the user end decrypts and decodes the ciphertext result, outputs a disease risk score, and obtains health data. This method makes the computation process more compatible with the GPU hardware architecture and scheduling mechanism, directly manipulates registers using assembly language to improve bit reversal efficiency, and designs a resource pool so that threads can place data into the resource pool after processing the current stage. Idle threads can find the data needed for the next stage in the resource pool without waiting and can directly execute the next stage's computation, reducing synchronization time.

The invention provides a risk control rule visualized configuration method and system, and the method comprises the steps: receiving a calculation function selected by a business analyst in a visualized editing interface, and enabling the calculation function to serve as a root node of an n-way tree; the calculation function has a plurality of parameter input bits, and for each parameter input bit, receiving a binding operation on a business data field or a sub-function to form a sub-node; constructing a complete n-way tree structure based on the root node and each child node, wherein the n-way tree structure represents an expression y = f (x1, x2,..., xn); receiving a selection of a risk disposal strategy associated based on the calculation result of the expression; combining the n-way tree structure with the selected strategy to generate a complete risk control rule; and converting the risk control rule into an expression which can be directly executed by the rule engine according to a script language supported by the target rule engine. According to the invention, the problems of high communication cost, long development period, low business participation degree, poor rule confidentiality and the like in a traditional risk control system are solved.

Example processors and methods for executing a looping code segment with zero overhead are provided. An example processor includes a pipelined architecture configured to execute a looping code segment having a first loop instruction at a first loop instruction address, a last loop instruction at a last loop instruction address, and a loop branch instruction. The processor includes a branch target address predictor, having a branch identifier, first instruction operational codes associated with the first loop instruction, a second loop instruction address in sequence with the first loop instruction address, and a branch taken indicator indicating a likelihood that the loop branch instruction directs execution to the first loop instruction. The branch target address predictor enables the pipelined architecture to execute the looping code segment with zero overhead, meaning the last loop instruction and the first loop instruction are executed on consecutive cycles.

This invention provides a script execution method, apparatus, electronic device, and storage medium. The method includes acquiring script execution request information, which includes a script execution request, a target script identifier, and target version information. Based on the target script identifier and target version information, a target script intermediate form is determined from multiple preset script intermediate forms. The preset script intermediate form is obtained by pre-compiling the script code. The target script intermediate form is then executed to obtain an execution result. By pre-compiling the script code, the script code intermediate form is directly executed when responding to the script execution request, thereby avoiding repeated compilation each time the script is executed. This compilation-execution separation architecture can achieve a balance between performance and resources, reducing script processing time and the resource consumption of script execution.

The invention relates to a method for interaction between a vehicle (1) and a user (2), wherein the user (2) carries out an interaction action for interacting with the vehicle (1), and wherein the vehicle (1) detects the interaction action with the aid of detection means (3) and generates an action recommendation (4) for the user (2) on the basis of an analysis of the interaction action and outputs said action recommendation to the user (2) by output means (5) and / or directly implements said action recommendation. The method according to the invention is characterized by the following method steps: A: feeding input data (6) generated by the detection means (3) to a first machine learning model (ML1) in order to generate an interaction text (7); B: processing the interaction text (7) by means of a first large language model (LLM1) in order to generate at least one action recommendation (4); C: processing the action recommendation (4) generated by the first large language model (LLM1) by means of a second large language model (LLM2) in order to generate a probability of success (8) for the respective action recommendation (4); D: comparing the probabilities of success (8) to a defined probability-of-success threshold value; and E1: outputting and / or performing the action recommendation (4) having the highest probability of success (8), if the probability of success (8) is greater than the probability-of-success threshold value; or E2: carrying out steps A to D again with adjustment of a random factor for the first machine learning model (ML1) until step E1 is reached.

The invention discloses a method for solving a machine learning model of mixed integer linear programming for industrial optimization, and belongs to the field of mixed integer linear programming solving of industrial optimization scenes. The method comprises the following steps: step 1, collecting multi-source data and preprocessing to generate a training set; constraint learning model training: dynamically selecting a best matching machine learning model according to industrial scene requirements; step 3, model automatic conversion processing: through automatic identification and analysis of a most matched machine learning model structure, a linearization constraint is obtained based on an SOS1 constraint, and a mixed integer linear programming constraint set, an objective function and a binary auxiliary variable set are output; step 4, optimization solution: carrying out solution through an open source solver to obtain a corresponding optimal decision variable value; and 5, outputting a solving result: analyzing the decision variable value obtained in the step 4 into an industrial optimization scheme which can be directly executed, and outputting the industrial optimization scheme. According to the method, general automatic modeling for mixed integer linear programming can be realized, and the solving efficiency and precision are ensured.

The invention discloses a control method and device, a storage medium and electronic equipment, and relates to the technical field of computer vision, and the method comprises the steps that at least one function area in a first interface is determined, and the function area is an effective control range for triggering function operation; in response to an input operation of the input device, determining at least one target function area based on a preset association table, and executing a function operation corresponding to the target function area; the target function area is one of the at least one function area, and the preset association table is a comparison table of the function area and the input operation. According to the technical scheme, the association table of the input operation of the input device and the function area in the first interface is established, when the input device inputs the operation, the operation corresponding to the target function area can be directly executed, and the problems that an existing physical device is low in virtual interface control speed and accuracy and cannot meet complex operation requirements are solved.

System emulation of a floating-point dot product operation 110 can be performed without directly performing the arithmetic by decomposing 130 the Addend into a constituent sign, an exponent, and a fra

The application relates to a model calling control method and device, equipment and a medium, the method comprising: in response to a streaming inference request sent via a gateway server, creating a calling record for initiating streaming inference calling of a large language model, the calling record comprising an abort flag initialized as not requested to be aborted and an execution flag initialized as not completed execution; starting the streaming inference calling; in response to a control handle reaching event, writing the control handle to the calling record, and detecting whether the abort flag represents having been requested to be aborted, if yes, directly executing the control handle to abort the calling; in response to a user abort instruction relayed via the gateway server, setting the abort flag to represent having been requested to be aborted, and when the execution flag represents not having been completed execution and the control handle has been written, executing the control handle and setting the execution flag to represent having been completed execution. The application realizes low delay, high reliability and precise one-time streaming inference task cancellation.

A method is disclosed for coupling a supporting system (12) to a main system (10) and enabling the supporting system to at least partially control the main system (11), the supporting system (12) including at least one input and output device for interacting with a user and a connection port for connecting the supporting system (12) to the main system (10), the input and output device including a display, the main system (10) being configured to directly execute main software, the method comprising the steps of: connecting the supporting system (12) to the main system (10); generating a user interface (34); displaying the user interface (34) via the supporting system (12); and granting access permissions that enable the supporting system (12) to at least partially control the main software via the displayed user interface (34).

The invention discloses an intelligent terminal interaction method and system based on semantic intention and scene arbitration, and the method comprises the steps: analyzing a user voice input into an initial semantic intention after a system-level voice service receives the user voice input; before the operation corresponding to the initial semantic intention is executed, firstly judging whether a target application is in an active state currently and whether the initial semantic intention can be provided for the target application; if yes, suspending direct execution of the initial semantic intention, and forwarding a data packet carrying the initial semantic intention to the target application to realize first transfer of the control right; and after receiving the initial semantic intention, the target application performs arbitration processing according to the service logic of the target application and the current user interface context. According to the method and the device, the dynamically loaded application can obtain the priority processing right for the system-level voice instruction, and after the instruction is arbitrated and processed according to the logic of the dynamically loaded application, which operation is executed by the system is determined, so that the intellectualization and fluency of user experience are improved.

The invention discloses a dynamic path correction method for an artificial intelligence robot, which comprises the following steps of: 1, generating an initial sequence: receiving initial environment observation O by a correction system, and quickly generating a preliminary complete action sequence A0 = [a10, a20,..., aT0] through an initial predictor; and step 2, iterative optimization circulation: the system does not directly execute the action sequence A0, but sends the action sequence A0 to an iterative optimization module for N rounds of optimization, and N is greater than 1. And step 3, sequence execution: after N rounds of optimization, outputting a final optimization sequence AN, and starting to execute actions in the sequence AN by the robot. According to the method, through a global attention mechanism, an action sequence which is highly consistent and smooth in time can be generated, and motion jitter caused by local decision is fundamentally avoided; the defect that autoregression model sequence generation is irreversible is overcome, and the unexecuted part of the sequence can be re-optimized, so that the unexecuted part can flexibly cope with environmental changes and uncertainty.

The application provides a low-precision approximate tensor operator calculation method based on a block unit, comprising: dividing a tensor operator into a plurality of subtasks according to a block size which can be directly executed by two matrix multiplication and accumulation instructions with different precisions supported by target hardware; mapping each subtask to each block unit in the target hardware, and finding a configuration vector which makes the calculation error of the tensor operator less than or equal to a preset error threshold, wherein an element of the vector indicates that the corresponding block unit uses one of the two instructions with different precision types to perform calculation; and setting the data processing sequence of each block unit for the subtask calculation according to the vector, so as to align the logical positions of the output results of each block unit with different precision types in the same reduction dimension in a register, and to reduce and accumulate the output results in the register under different precisions. The application improves the resource utilization and execution efficiency by blocking and aligning the output results with different precisions.

The application discloses a device diagnosis method and device based on a driving function, and the method comprises the following steps: when a diagnosis service request about a target device sent by a client is detected, at least one driving file matched with the diagnosis service request is loaded from a dynamic link library corresponding to a device server, each driving file comprises at least one driving function; a service corresponding to each driving function is executed to diagnose the target device, and a diagnosis result corresponding to the diagnosis service request is obtained. It can be seen that the driving function used for diagnosing the device can be written in the dynamic link library, the diagnosis service request is received, the driving file in the dynamic link library is called to the device server, and the driving function in the driving file is directly executed in the device server to meet the diagnosis requirement of the device, so that the development difficulty of the diagnosis program is reduced, and the calling efficiency of the diagnosis function is improved.

The invention relates to a method for backing up an operating system under domestic platform firmware, and belongs to the technical field of domestic computers. According to the method, based on a GPT hard disk partition table, a backup partition table, an FAT32 file system, compression and other technologies, the software technical scheme of backing up and restoring a domestic operating system under domestic firmware is achieved. The operation of the backup and restoration tool does not depend on other file libraries, hot keys can be bound as required, the tool can be started as required when the BIOS is started, and an executable. EFI format file can also be loaded to an EFI Shell environment of BIOS firmware by a USB flash disk to be directly executed. By backing up and restoring the operating system in the firmware, the requirement that the operating system can be backed up or restored without logging in the operating system is met.

Processing unit arrangement comprising a data source unit (10), a processing unit (20), and an actuation unit (30). The data source unit (10) is configured to provide the processing unit (20) with direct input data (D). DI ) to provide, wherein the data source unit (10) and the processing unit (20) are directly connected to each other via first data lines (W1). The processing unit (20) includes an arithmetic unit (22) configured to process direct actuation data (D DA ) to determine by performing arithmetic operations on the direct input data (D DI ) performs. The processing unit (20) is configured to process the direct actuation data (D DA) to provide the actuation unit (30), wherein the processing unit (20) and the actuation unit (30) are directly connected to each other via second data lines (W2). The actuation unit (30) is configured to operate an application using the direct actuation data (D DA ) to control.

The present disclosure relates to the technical field of computers. Provided are a data operation method and system, and a storage medium and a computer program product. In the present disclosure, in the case that a front-end operation instruction involves a plurality of types of syntax, a corresponding processing method is provided. Accordingly, a user does not need to use a fixed type of syntax in an instruction input process. As such, when the user wants to perform a first operation, the user can select a first type of syntax to write an instruction, wherein the first type of syntax defines an instruction capable of directly executing the first operation; and when the user wants to further perform a second operation, if the first type of syntax does not define an instruction capable of directly executing the second operation, the user can select a second type of syntax to write an instruction, wherein the second type of syntax defines the instruction capable of directly executing the second operation. In this way, for an operation that the user wants to perform, the user does not need to split the operation into a plurality of steps or input a plurality of instructions. Therefore, the operation efficiency of the user can be improved.

To reuse flushed instructions after an instruction pipeline flush in response to a hazard.SOLUTION: The instruction processing circuit detects a performance-degrading instruction (PDI) that could cause a precise interrupt that would cause a flush of the instruction pipeline at a preliminary execution stage in the instruction pipeline, and in response to the PDI detection, captures the fetched PDI and / or younger fetched instructions that follow the PDI in the pipeline refill circuit. The instruction processing circuit also injects the fetched PDI and / or its younger instructions, previously captured from the pipeline refill circuit, into the instruction pipeline to be processed without being re-fetched if execution after the PDI in the instruction pipeline causes a flush of the instruction pipeline.SELECTED DRAWING: Figure 3

The invention relates to a method for interaction between a vehicle (1) and a user (2), wherein the user (2) performs an interaction action to interact with the vehicle (1), the vehicle (1) detects the interaction action using detection means (3) and, depending on an analysis of the interaction action, generates a recommendation for action (4) for the user (2) and outputs this recommendation to the user (2) via output means (5) and / or implements it directly. The method according to the invention is characterized by the following process steps: A: Feeding the input data generated by the acquisition means (3) (6) to a first machine learning model (ML1) to generate an interaction text (7); B: Processing the interaction text (7) by a first large language model (LLM1) to generate at least one action recommendation (4); C: Processing the action recommendation (4) generated by the first large language model (LLM1) by a second large language model (LLM2) to generate a probability of success (8) for the respective action recommendation (4); D: Comparing the success probabilities (8) with a specified success probability threshold; and E1: Issuance and / or execution of the action recommendation (4) with the highest probability of success (8), if the probability of success (8) is greater than the probability of success threshold; or E2: Repeat steps A to D, adjusting a random factor for the first machine learning model (ML1), until step E1 is reached.